Regional Identity and Spatial Spillovers
Eva van der Wal s2554615
Msc thesis Economics
Supervisor: Prof. dr J.P. Elhorst
June 2018
Keywords: Regional Unemployment –Spatial Spillovers - Regional Identity – Spatial Panel Data model
3
Abstract
5
Introduction
Unemployment policy is one of the priorities of each European national government, and one of the key components of the Europe 2020 Strategy as formulated by the European Union (2017). This strategy sets out a target of 75% of the European population aged between 20 and 64 having paid employment by 2020, which currently is 71.3%.
While national and supranational attention is paid to this topic, unemployment often has regional causes, shown by the large regional unemployment differentials across countries in the European Union. Where some countries stand out for their high (Greece) or low (Germany) regional unemployment rates, other countries’ regional unemployment rates are particularly dispersed (Belgium, Italy, Austria, Hungary). Unemployment differentials between countries are often explained by differences in labour market institutions, for example wage bargaining and tax systems. However, these labour market institutions cannot explain regional unemployment differentials, as they are usually similar across regions within a country. This gives reason to consider unemployment from a regional perspective (Elhorst, 2000).
The regional nature of unemployment emphasizes the importance of recognizing regional characteristics in the design of national unemployment policies. In one region high unemployment may be caused by the presence of a declining industry, where in another region unemployment is related to the demographic structure or the educational attainment of the population (Elhorst, 2000). Both situations require a different policy response.
At the same time, regions are not floating islands, but interdependent parts of a system, because the labour force and firms are mobile between regions. It is important for policy makers to have knowledge about the strength of this interdependence, as it can have large effects on regional unemployment rates and regional development. Take for example the Italian case, where the massive outflow of higher educated workers from the south to the north of Italy causes a serious ‘brain-drain’.
6 provincial unemployment disparities is found by Cracolici et al (2008). Blanchard and Katz (1992) have done this for the US.
These papers focus on one country and therefore cannot compare the estimated spatial dependence with that of other countries. Obstfeld and Peri (1998) compare adjustment to regional unemployment shocks in the European Union with the United States. They find that internal adjustment to a regional unemployment shock is slower in European countries than in the US. Therefore, local unemployment shocks in Europe are more persistent, where in the US the effects will be spread out to other regions. The effect of a regional unemployment shock on the unemployment rate of other regions is called the spatial spillover effect, these are thus higher in the US. Their explanation for this difference is that interregional migration within the United States is much more sensitive to wage and unemployment differentials and thus serves as an effective adjustment mechanism to regional unemployment shocks.
This paper offers a comparative approach for countries within western Europe. Spatial spillover effects between regions will be estimated for seven European countries in a similar way, using spatial econometric techniques. This way, the magnitude of spatial spillover effects of different countries can be observed. The question is which characteristics of countries influence this magnitude. This paper will focus on the influence of regional identity.
In times of European and global political and economic integration, the region is getting more and more important in an emotional sense. Exactly because nations have lost power to this supranational union, and because of the distance between ‘normal citizens’ and this union, people feel threatened and estranged. Therefore the region and regional identities are gaining importance. People find recognition and safety in the region, opposed to the distant nation or continent they live in (Cornips & Stengs, 2010). This also leads to conflicts. There are some very recent examples of situations of large contrast between one or more regions and their nation in western Europe. First of all the Spanish region Catalonia, where the majority of the citizens voted in favour of independency from Spain, which resulted in violent confrontations between the Spanish police and Catalonian citizens. Secondly the Northern Italian regions, who may not strive for independence from their nation, but definitely want more financial autonomy.
7 umbrella organization operating at European level, where regionalist parties that strive for more freedom are members of. Figure 1 shows that in many European countries, such a party is active.
This paper focuses on the influence of regional identity on spatial spillovers in unemployment. In countries where regional identity is strong, it is expected that the labour force is less mobile. People prefer to work in their own region and do not want to move to another region in case of unemployment. Consequently, spatial spillovers between regions in unemployment will be lower. The main hypothesis of this paper can thus be formulated as:
H1: In western European countries where regional identity is stronger, spatial spillover effects are lower
where spatial spillover effects are between NUTS-2 regions and are concerning regional unemployment rates. The remainder of this paper is organized as follows. The next section explains the basic theoretical model of regional spillovers and the role of migration, and which influence regional identity could have on this. The following section describes the data and estimation method, followed by the results. The last sections discuss and conclude.
Literature overview
In economic theory, spatial spillovers between regions are a consequence of the rational actions of agents. Initially, regional unemployment rates are in equilibrium state. Suppose an unemployment shock in region x occurs, caused by for example a large firm going bankrupt. This shock is expected to be an incentive for: 1) workers to move out of the region in search for better opportunities
8 elsewhere; 2) firms to migrate into the region, attracted by the large labour reserves available; and 3) regional wages to fall down in order to equalize labour supply and demand. These three adjustment mechanisms will bring the regional unemployment rate back to its original relative level: the effect of the shock in region x is spread out to other regions where the unemployment rate will increase as well. When the equilibrium is restored, unemployment differentials can only be the result of underlying preferences of workers that influence their choice of residence, for example natural amenities.
Marston (1985) argues that the first mechanism is the most important: labour mobility is the key factor for restoring the unemployment equilibrium. As long as economic and social barriers on labour mobility are low enough, the quantity of migration in response to unemployment shocks is so large that the shock is eliminated within a year. This relationship between unemployment shocks and labour mobility is confirmed by many authors. Molho (1995) finds that an increase in local demand, and thus in employment, in one British local labour market area, will eventually be spread out over the whole country. The local employment shock has an effect in the immediate period on the particular region. During subsequent periods, workers migrate and spillovers increase, so ultimately unemployment levels will be restored to their original values. The spatial interaction effect he finds is lagged in time, which is consistent with this migration behavior. Pissarides and McMaster (1990) confirm this finding, but emphasize that the adjustment is a long-term process that can take at least twenty years. Treyz et al (1993) study the delong-terminants of interregional migration and find that net migration is a response to changes in amenity differentials and relative economic opportunities, mainly employment probability. Also Blanchard and Katz (1992) state that labour mobility is primarily a response to changes in unemployment.
Migration and commuting
9 region 2. The unemployment rate is defined as 𝑈𝐿
𝐿𝐹, where UL equals the level of unemployment and
LF the total labour force per region. It is assumed that the fictive person was unemployed in region
1 because of an unemployment shock, and finds a job in the region 2, thereby replacing another worker.
Table 1 Change in Unemployment rate by migration/commuting from region 1 to region 2
Migration Commuting
Region 1 Region 2 Region 1 Region 2
𝑈𝐿− 1 𝐿𝐹− 1− 𝑈𝐿 𝐿𝐹 < 0 𝑈𝐿+ 1 𝐿𝐹+ 1− 𝑈𝐿 𝐿𝐹 > 0 𝑈𝐿− 1 𝐿𝐹 − 𝑈𝐿 𝐿𝐹 < 0 𝑈𝐿+ 1 𝐿𝐹 − 𝑈𝐿 𝐿𝐹 > 0
After an unemployment shock in region 1, a worker may choose to migrate to region 2. In the first column of table 1 we see that this decreases the unemployment rate in region 1 (because UL<LF by
definition). The second column shows that this in-migrant increases the unemployment rate in region 2. Thus, the unemployment shock in region 1 spreads out to region 2. The same worker can also decide to stay in region 1, but to accept a job in region 2 and travel from home to work every day, a strategy called commuting. In this case, column 3 and 4 show that the unemployment rate in region 1 decreases again, and increases in region 2. Thus, both commuting and migration are expected to create a positive spillover effect between the two regions: the unemployment shock in region 1 leads to an increase in the unemployment rate of region 2. The same mechanisms will also lead to an increase in the unemployment rate in region 3, and so on, until the unemployment shock of region 1 is spread out over the whole country. However, the spillovers caused by commuting are larger than in the case of migration, because 𝑈𝐿−1
𝐿𝐹−1 > 𝑈𝐿−1 𝐿𝐹 and 𝑈𝐿+1 𝐿𝐹+1 < 𝑈𝐿+1 𝐿𝐹 . Another difference
is that the unemployment effect caused by migration is static: the unemployment rate changes at the moment of migration only. The unemployment effect of commuting is more dynamic, during the whole period in which the worker lives in region 1 and works in region 2, the unemployment rate remains lower in region 1 and higher in region 2.
10 place to a limited distance. Following Layos and António (2007), people’s daily routines such as commuting to school and work usually take place within a 50-km radius from home. Although other research predicts that the distances people are willing to commute will increase due to better transportation possibilities and more possibilities to work from home (Schlafer & Victor, 2000; Green, Hogarth, and Schackleton, 1999), it is still expected to be quite exceptional that commuting takes place between two NUTS-2 regions. In most of the countries, these regions represent a fairly larger scale. Only weekly commuting, when a worker turns back home for the weekend and lives close to his/her job during the week, may take place between NUTS-2 regions, but this accounts for no more than a small share of the commuting workers, see again Green et al. (1999). Therefore, in this paper the focus will be on migration as the adjustment mechanism in response to regional unemployment shocks. Commuting is expected to be of greater influence on local and sub-regional level. However, for smaller countries with smaller regions, this argument may not hold. In the Netherlands for example there is intensive commuting between Almere and Amsterdam, cities located close to each other but in different NUTS-2 regions. This must be taken into account when interpreting the results.
Differences in labour mobility
It has been argued that migration is an important factor in the spatial dependence between regions in unemployment. The stronger the labour force responds to unemployment differentials between regions via migration, the higher spatial spillovers between regions will be. When people would behave perfectly rational and move according to unemployment differentials, each unemployment shock will be spread out over all regions, as the theoretical model predicts. Does this hold for all countries? That depends on the mobility of the labour force.
11 the US, interregional migration from the first year after the shock accounts for 52% of the adjustment. In Europe, this proportion is similar only from the third year onwards, changes in the participation rate accounted for a larger proportion in the first years. For example, workers choose to retire at an earlier age or women drop out of the labour force in response to an unemployment shock. De Grauwe and Vanhaverbeke (1993) report results for labour mobility in European countries only. Their results indicate that labour mobility in Europe differs significantly across countries, and that it is higher between regions than between countries. Furthermore, they found interregional labour mobility to be higher in the north of Europe than in the south. However, they did not give an explanation for this finding.
Assuming that spatial dependence between regions depends on the mobility of the labour force, understanding why in one country the population is more mobile than in another might explain differences in spatial spillovers across countries. A more mobile population has a higher willingness to migrate in response to unemployment differentials. In the next section, it will be argued why regional identity may influence this willingness to migrate.
Regional identity
12 distinguish a region from others, for example landscape, local language, local food, or names of places. On the other hand, it can refer to the regional identity of the inhabitants: the extent to which people within a region identify themselves with each other and with the region. This can be labelled as regional consciousness and forms the fourth shape. In the extreme case, each region has a homogenous population with similar characteristics, which are different among regions. Paasi emphasizes that although feelings of unity and solidarity sound positive, regional identity may also be a form of opposition or conflict with other regions or the nation: an ‘identity of resistance.’
In countries where regional identity is strong, we expect larger differences between regions and their inhabitants in terms of nature, culture, politics, and perspective on the rest of country. People will identify themselves mostly with the region where they are born, and feel connected to the people they share this identity with. Regional identity will be expressed strongly, for example by different regional languages, political parties representing regions, and regional traditions. These expressions might be stronger than the expressions of national culture. Because of the value people attach to living in the region where they are born, it is expected that they are less willing to leave this region because of a relatively high unemployment rate. Interregional migration would imply that people leave the region they feel connected to. Furthermore, they move to another region with a strong regional identity as well, which makes integration more difficult. It can even be possible that a new language needs to be learned. The incentive to move to another region (employment) must be traded off against the loss of utility from migrating out of the home region (social familiarity). This loss of utility is bigger in countries where people have a strong regional identity. Therefore, it is expected that interregional migration is less responsive to unemployment differentials in countries with strong regional identity and is thereby less effective as an adjustment mechanism to an unemployment shock. Once the group of people that is willing to migrate has left after an unemployment shock, the remaining unemployment disparity will present the new equilibrium. The unwillingness to migrate thus serves as a distorting factor in order to restore the initial equilibrium (McArthur, Thorsen & Ubøe, 2010).
13 statistically significantly related to the regional unemployment rate. This means that people with greater attachment to the region ‘accept’ a higher unemployment rate and stay, instead of move out of the region.
Another interesting approach to the relation between regional identity and population migratory behaviour is given by Raagmaa (2002). He explains two possible mechanisms in which regional identity itself would influence the incentive to migrate, i.e. economic downturn. Crucial is the type of society in the region. A ‘Gemeinschaft’-type of society is very traditional and closed (Tönnies, 1955). These societies rely on traditional forms of production and are less innovative. As a result, slow stagnation of the economy and of social and cultural life takes place, which leads to an outward migration flow. A second type of regional societies with a strong identity are based on affectual solidarity and shared beliefs, known as ‘Bund’ (Schmalenbach, 1922). This term is more applicable to gatherings of people cooperating and supporting each other in common goals or beliefs. These ‘Bund’-regions are more innovative and entrepreneurial, thereby stimulating economic development, which attracts migrants.
14 Concluding, because of a higher resistance to migrate in response to unemployment shocks, and a higher degree of decentralization, it is expected that in countries with strong regional identity, spatial spillovers between regions in unemployment will be lower.
Data and Method
Regional unemployment
Regions exist at different scales. The European Commission has defined a hierarchical system to divide its territory called the NUTS classification (Nomenclature of Territorial Units for Statistics). Regions are defined at NUTS 2-level in this research, as these are basic regions for the application of regional policies (European Commission, n.d.). NUTS-2 level unemployment data is made available through Eurostat since 1986. For the full period 1986-2016, data of six European countries is complete: Belgium, France, Germany, Italy, the Netherlands, and Spain. For the United Kingdom the data can be made complete by combing Eurostat data and the data used by Overman and Puga (2002) in their study on regional unemployment clusters in Europe. Hence the sample covers 7 European countries, consisting of 154 NUTS-2 regions. Although this number is limited, the sample includes large and small countries, in the north and the south of the continent, and is therefore a good representation of western Europe. Descriptive statistics of the regional unemployment rates per country are given in table 2.
15
Table 2 Descriptive statistics regional unemployment rate 1986-2016
Country Obs Mean Std. Dev Min Max
Belgium 341 8.36 3.82 1.9 19.2 France 648 9.72 2.33 4.73 18.1 Germany 1055 6.9 3.48 2.06 22.4 Italy 602 10.38 5.94 1.98 28.1 Netherlands 372 5.80 2.47 1.2 13.76 Spain 496 16.09 6.69 4.1 36.2 United Kingdom 981 6.91 2.99 1.8 18.6
The distribution of regional unemployment levels across the different countries is given in figure 1, panels A-G. From this figure the following observations can be made. In Belgium (11 regions), regional differences in unemployment exist and have increased relative to the 1990s. However, during the last ten years, the absolute and relative levels tend to be quite stable. In Belgium, Brussels-Capital Region in particular stands out as one region with high unemployment over the whole period. Between this region and the lowest-unemployment region a gap of around 14 percentage points is found. In France (21 regions), an opposite trend is observed. Here regional disparities have decreased relative to the 1990s and the difference between the most extreme regions nowadays is around 7 percentage points.
The graph of Germany (38 regions) looks somewhat puzzling, as in the late 1990s a new group of high-unemployment regions enters the picture. These are the former East-German regions, of which data is available only from 1999 onwards. When entering the dataset, these regions have an unemployment rate far above that of the other regions of Germany, which are all quite close to each other. The Eastern regions however quickly catch up with the others, and only slightly higher unemployment levels remain today. The average regional unemployment rate in Germany is low compared to the other countries.
16 north, is now almost 18 percentage points.
The graph of the Netherlands (12 regions) shows that regional unemployment rates move more-or-less together, with one or two regions always either a bit higher or lower than the others. The difference between those however is only a few percentage points.
In Spain (16 regions) a few regions have a long-lasting higher unemployment rate. Just like in Italy this gap has been decreasing in the beginning of this century, but during last years, when unemployment rates started to increase in the whole country, this difference increased again. In general in Spain regional unemployment rates are high compared to the other countries.
The last panel shows regional unemployment rates in the United Kingdom (36 regions), where regions differed a lot at the beginning of the period, but tend to become more similar over time. With most regions having an unemployment rate around 5 percent, the United Kingdom is one of the countries with the lowest regional unemployment rates.
A Belgium B France
17
E The Netherlands F Spain
Figure 2 Distribution of regional unemployment rates per country, 1986-2016
Regional identity
To evaluate the hypothesis that stronger regional identity will lead to lower interregional migration and therefore lower spatial spillovers, the regional identity variable is introduced. The statistic is retrieved from the European Values Survey (2008). The 2008 edition is the fourth wave of this large population survey that covers a broad range of topics and is administered in all European countries with a population of 100.000 or more. In each country a random sample of the adult population is drawn and approached for face-to-face interviews. Table 3 provides a partial overview of the relative frequency of answers in the sample countries on the following question: ‘to which territorial unit are you belonging the most?’ Respondents could answer this question with
18 ‘the world’, ‘Europe’, ‘own country’, ‘region of country’, or ‘locality or town’. The answer options expected to be relevant with respect to regional identity are presented in table 3. It is assumed that the percentage of respondents answering ‘region’ gives the best indication of the strength of regional identity in the particular country.
Table 3 ‘To which territorial unit are you belonging the most?’: relative frequency of answers
In three countries, the percentage of respondents answering ‘region’ is remarkably low: Netherlands, Italy and France. The lowest value of 8.69 percent is found for the Netherlands. Although in the Netherlands regional cultures exist with its own dialects, they are often seen as part of the national identity, something ‘typically Dutch’ (Brand & Sijs, 2007). Of the Italian respondents, 11.74 percent has indicated to feel belonging to the region. This corresponds with the notion of Italy being generally recognized as one of the most centralized states in western Europe (Allum & Amyot, 1970). In France, only 12.68 percent of the respondents feels to belong to the region they live in. Also the French central government has always tried to minimize mediation between the individual citizen and the nation, in order to create a strong central state (Wagstaff, 1999). Consequently, in France and the Netherlands, the low percentages for the category ‘region’ are accompanied by a high percentage of respondents answering that they belong first to the country (respectively 36.72 and 37.33 percent). For Italy however, this percentage is lower (28.92 percent), because more respondents have indicated that they belong to the locality or town they live in.
In Belgium, United Kingdom and Spain, regional identity is moderate. In Belgium (22.21 percent), this is most probably related to the substantial differences (e.g. in language) between the Flemish, Wallonian and Brussels regions, which is also translated in a certain degree of autonomy
region locality country N
19 for the three subnational governments (Albrechts, 2001). The percentage of Belgian respondents answering ‘locality’ is also quite low (29.55 percent). The United Kingdom is a union of different regions by definition. The 17.77 percent respondents answering to belong first to their region are expected to be mostly Scottish, Northern Irish and Welsh people, for who the British identity will be closer towards a second identity than a first. At the same time, the relatively high percentage for ‘country’ (35.69 percent) will probably be the English people, who might see their identity as the national one because they form the majority of the population (Aughey, 2010). Spain is quite an exceptional case. The Spanish constitution is based on the integration of different nationalities and regions, with each a high degree of autonomy. Spain is the only country in the sample in which five to nine regional parties always participate in the national government (Llera, 2009). The score of 16.36 percent indicates that a substantial part of the population does identify themselves with the region, and especially with their locality (46.58 percent). Only 25.27 percent of the respondents feels to belong to the main country.
20
Estimation method
For estimating spatial spillovers in regional unemployment, it is important to choose the right model. Halleck-Vega and Elhorst (2016) provide methodological arguments to simultaneously control for serial dependence, spatial dependence, and common factors. Serial dependence describes the tendency for region-specific unemployment rates to be correlated over time. Spatial dependence is the factor of interest in this research: unemployment within a region is influenced by the level of unemployment in surrounding regions as well, through labour migration. The third factor they emphasize are common factors. All regional unemployment rates tend to move in tandem with the national unemployment rate, but one more closely than another. The extent to which a region responds to changes in the national unemployment rate is called cyclical sensitivity. The model they propose that addresses these three dependencies simultaneously is a Dynamic Spatial Panel Data model. The model specification as it will be used in this paper is specified in formula 1:
𝑌𝑟𝑡 = 𝜏𝑌𝑟,𝑡−1+ 𝜌 ∑𝑅𝑗=1𝑊𝑟𝑗𝑌𝑗𝑡+ 𝜂 ∑𝑅𝑗=1𝑊𝑟𝑗𝑌𝑡−1+ 𝛾1𝑟𝑢𝑁𝑡+ 𝛾2𝑟𝑢𝑁𝑡−1+ 𝜇𝑟+ 𝜀𝑟𝑡 (1)
21
Results
First, the model in equation 1 is estimated for each country separately. This means for each country the panel exists of its own regions and the spatial weight matrix is defined accordingly. Therefore, bordering regions that belong to another country are not included. The model is estimated as a Spatial Durbin model using maximum likelihood, with the xsmle function in Stata. Table 4 presents the estimation results for all seven countries. The coefficients of serial and spatial dependence are reported in separate columns, which will be discussed in the next sections. To provide a clear overview of the results needed to evaluate the main hypothesis, the statistic for regional identity is reported in column 7. In the last column the countries are ranked based on this statistic, where number 1 has the highest score on regional identity and number 7 the lowest. Model diagnostics are reported in table 5.
Spatial effects
The coefficients of serial and spatial dependence are presented in columns 1-4 of table 4. All estimated coefficients are significant, except for the lagged spatial coefficient in the Netherlands. The first column shows the coefficient for serial dependence τ. All serial coefficients are positive and significant, which indicates a positive correlation between regional unemployment rates over time for the whole sample. The coefficients differ in magnitude, where the highest value 0.704 is found for Germany, and the lowest 0.422 for the Netherlands. The next three columns present the coefficients of spatial dependence. In all countries a significant spatial (ρ) and spatial lagged effect (η) is found. ρ + η gives the total spatial effect, which is also positive in all countries. This is in line with theory; when one region experiences an increase in unemployment, neighbouring regions will get higher unemployment rates as well. Common factors are not reported for reasons of space, they can be found in the appendix.
22 Netherlands are ranked 5th and 7th respectively based on their regional identity, so regional identity is relatively low. Italy has the third biggest spatial effect and equally low regional identity, being ranked number 6. The lowest spatial effect is found in Germany, where regional identity is strongest. In Belgium, Spain, and United Kingdom intermediate values are found for both spatial effect and regional identity, although last is somewhat higher in Belgium.
To compare spatial effects across countries, a second step might be to compare the ratio of spatial and serial dependence between countries. Absolute values of coefficients may differ between countries because unemployment levels are of different size. The ratio between serial and spatial effects however gives an indication of the relative importance of spatial effects in each country. Total spatial effect is divided by serial effect, which gives the ratio (ρ + η)/τ, reported in column 5. The countries are ranked based on this ratio in column 6, where number 1 has the highest ratio and number 7 the lowest.
23
Table 4 Dynamic Spatial Panel Data model, results
* indicates significance at 1% level
Table 5 Model diagnostics
Serial effect Spatial effect Ratio spatial/serial effect Regional identity
τ ρ Η ρ+ η (ρ + η)/τ rank % rank (1) (2) (3) (4) (5) (6) (7) (8) Belgium 0.494* 0.708* -0.422* 0.286 0.58 #3 22 #2 France 0.452* 0.289* 0.184* 0.473 1.05 #1 13 #5 Germany 0.704* 0.678* -0.467* 0.211 0.30 #7 38 #1 Italy 0.602* 0.450* -0.119* 0.311 0.52 #4 12 #6 Netherlands 0.422* 0.459* -0.027 0.432 1.02 #2 9 #7 Spain 0.659* 0.869* -0.631* 0.238 0.36 #6 16 #4 United Kingdom 0.623* 0.711* 0.238 0.271 0.43 #5 18 #3
Belgium France Germany Italy Netherlands Spain United
Kingdom
N 11 21 38 20 12 16 36
T 31 31 31 31 31 31 31
R2 0.766 0.683 0.698 0.918 0.894 0.689 0.707
Direct and Indirect effects
To properly interpret the impact of a change in unemployment in region i on the unemployment rate in neighbouring regions, the direct and indirect effects are reported in table 6. For this purpose, LeSage and Pace (2009) suggest to use the reduced form of the regular spatial Durbin model
y=(I-ρW)-1βX + (I-ρW)-1δWX + (I-ρW)-1e. The matrix (I-ρW)-1 includes both types of effects. The rj-th element equals the effect of region j on region r. Likewise, direct effects can be calculated by the diagonal element of this matrix (I-ρW)-1 and indirect effects by the average non-diagonal row sum
(being the total indirect effect of all other regions on region r). Since an extended version of the spatial Durbin model is used, the dynamic spatial panel data model, the matrix of interest changes into [(1-τ)I-(ρ+η)W]-1. Also the serial effect τ and the spatial lagged effect η, which make the model
dynamic, are then included. The direct effects presented here can be interpreted as the long-term impact of a regional unemployment shock in the own region i (including feedback effect from other regions). The indirect effect represents the long-term impact on the unemployment rate in region i of a shock in other regions.
Table 6 Direct and Indirect effects
Direct effect Indirect effect Total effect
Belgium 0.470*** -0.226** 0.245*** France 0.475*** 0.421*** 0.896*** Germany 0.708*** 0.033 0.740*** Italy 0.624*** 0.226*** 0.850*** the Netherlands 0.452*** 0.280*** 0.732*** Spain 0.626*** -0.393** 0.233 United Kingdom 0.625 *** 0.016 0.641***
*indicates significance at 10% level;**indicates significance at 5% level;***indicates significance at 1% level
25 countries had the highest total spatial effect and were among the four countries with the highest ratio of spatial to serial effects. One difference stands out: where the Netherlands and France had similar results for spatial effects in table 4, spillover effects in France are significantly higher. These three countries with positive and significant spillover effects are also the three countries with the lowest score on regional identity.
In Germany and the United Kingdom, the results show no significant spillover effects. This means that regions in the long term are not affected by an unemployment shock in a neighbouring region. For Germany, this corresponds to the low spatial effects reported in table 4. With the highest score on regional identity, Germany repeatedly confirms the hypothesis that stronger regional identity leads to lower spatial spillovers. The results for the United Kingdom show a somewhat more mixed picture, as spatial effects in table 4 were not remarkably low. So only with respect to indirect effects, the United Kingdom is similar to Germany. Regional identity is also much stronger in Germany than in the United Kingdom.
In Belgium and Spain, spillover effects are negative. This means that an unemployment shock in one region will lead to lower unemployment rates in connected regions. This spillover effect seems to contradict with standard theory but might be illustrated with an example. Normally, when in one region a large company goes bankrupt, one would expect people to search for jobs in neighbouring regions. The increased unemployment rate in region i then also leads to a higher unemployment rate in neighbouring region j. In the case of negative spillover effects, an increased unemployment rate in region i will lead to a lower unemployment rate in region j. After the large company going bankrupt, other companies might decide to relocate to the other region j, or companies already located in the region j might see possibilities to fill the gap in the market created by the bankruptcy of the other company. These effects apparently dominate in Spain and Belgium. When taking a closer look at these countries, this can be well-explained. Both Spain and in Belgium score relatively high on regional identity, and in both countries differences in language exist between regions1. These language differences go together with strong rivalry between the regions or between the regions and the ‘main’ country. Both Belgium and Spain are also characterized by large differentials in regional unemployment rates. This could form an additional incentive for firms to relocate: once a few large companies go bankrupt in a high unemployment region, the
1 In Belgium, French is spoken in the southern regions and Flemish in the northern regions. In Spain, although
26 economic circumstances may have become too bad for other firms to stay in this region. These factors could contribute to the negative spillover effects as found in this analysis: people might not be able to find jobs elsewhere because of the language, and firms might decide to relocate to one of the competing regions as circumstances change, just like big multinational firms choose to build their headquarters in the country with most favourable tax laws.
To sum up, in France the Netherlands and Italy, positive spillover effects are found. France, the Netherlands and Italy are also the three countries that score lowest on regional identity. In the countries with stronger regional identity, these positive spillover effects are not found. In Belgium and Spain we identified significant spillover effects, but negative. Here interregional language differences and rivalry may play a role. In the United Kingdom and Germany, hardly any spillover effects are found. Especially for Germany, where regional identity is strongest, this result is line with the main hypothesis.
Other factors
27
Table 7 Average size of NUTS 2-region
Country km2 Belgium 2775 France 26269 Germany 9404 Italy 15067 the Netherlands 3462 Spain 31624 United Kingdom 6736
The largest regions are found in France and Spain. These countries however differed a lot with respect to spatial effects. In France spatial effects were high, in Spain they were much lower and the spillover effects were even negative. The smallest regions are the Dutch and Belgium regions. The spatial effects of the Netherlands and Belgium however, did not show many similarities either, being high in the Netherlands and much lower in Belgium. The size of Dutch regions could contribute to the positive spatial spillovers because commuting can take place between NUTS-2 regions, but this does not hold for Belgium. Apparently regional identity or language differences also play a role when commuting is possible based on distance. Germany, the country with on average the lowest spatial effects, holds the median position with respect to average region size. To conclude, average size of a region does not seem to influence spatial spillovers as there is no clear relationship found between size of the region and spatial spillover effects for this sample of countries.
28
Table 8 Fiscal decentralization by percentages of total government expenditure
Belgium France Germany Italy Netherlands Spain United Kingdom Central 55.13 80.60 59.91 72.52 68.62 56.59 75.89
State 31.76 - 22.80 - - 32.11 -
Local 13.12 19.40 17.28 27.48 31.38 11.30 24.11
Source: OECD Fiscal Decentralization database (2016)
The percentages of central and decentralized government spending mostly confirm the association between regional identity and decentralization and its influence on the relationship between regional identity and spatial spillovers. In countries with strong regional identity and low spatial effects (Germany, Belgium), centralized government spending accounts for less than 60 percent of total government spending. In the Netherlands, Italy and France, where regional identity is low and spatial effects are high, this share is higher. Spain and the United Kingdom are almost similar in terms of regional identity,but have a very different share of central government spending: in Spain it is only 56.59 percent and in the United Kingdom 75.89 percent. This might be related to the fact that the United Kingdom has one fewer layer of government, nevertheless they do not confirm the expected association between regional identity and decentralization. On average, spatial effects were somewhat higher in the United Kingdom than in Spain (see tables 4 and 6). The difference in decentralization could serve as an explanation for this observation. In the United Kingdom, regional identity is less strongly reflected in regional autonomy, which might explain why spatial effects are higher than in countries with similar regional identity (Spain).
Discussion
29
Countries with low regional identity
Of this sample, regional identity is smallest in France, the Netherlands and Italy. For these three countries, spatial effects were high in comparison to the other countries, but lower for Italy than for France and the Netherlands. We identified positive and significant spatial spillover effects for all three countries. These were highest for France, followed by the Netherlands and Italy. Regarding this subgroup, the hypothesis that stronger regional identity leads to lower spatial spillovers holds, since the results show that in these countries with the lowest regional identity, spatial spillovers are highest.
Within the subgroup, the biggest difference exists between Italy and the other two countries: in Italy spatial dependence is slightly lower on average. Two factors may be driving this finding. First, the Italian respondents indicated that they mostly belong to their ‘locality or town’, whereas the French and Dutch respondents answered more frequently that the ‘country’ was their main territorial unit for identification. Secondly, regional unemployment differentials are much larger in Italy than in France and the Netherlands. Therefore, in Italy it is expected that there is a more continuous stream of interregional migration from the South to the North, not only in response to unemployment shocks. This continuous stream of migration leads to negative spillovers between regions (because it lowers unemployment in the South, vice versa in the North) and may therefore lower the spatial spillovers found in response to an unemployment shock. Although the central government is relatively powerful in all three, the degree of decentralization is still different between the countries, which may serve as an explanation why spatial spillovers are higher in France then in the Netherlands and Italy.
Countries with moderate regional identity
For Belgium, Spain and the United Kingdom, regional identity is found to be stronger and spatial dependence lower than in the first subgroup, but both to a less extent than in the third. Therefore, it can be concluded that also the results of this subgroup confirm the hypothesis that stronger regional identity leads to lower spatial spillovers.
30 the United Kingdom. Second, for the United Kingdom no significant spillovers could be identified, where for Belgium and Spain negative spillover effects were found. The first difference concerning spatial effects in Belgium might be related to the relatively small size of Belgian regions, which makes migration or commuting between regions of the same language area relatively easy. The United Kingdom differs from the other two countries in terms of decentralization and regional unemployment differentials. Belgium and Spain know a more decentralized government and a higher degree of unemployment disparities between their regions. This might be the explanation for the fact that negative spillover effects were found here and not in the United Kingdom.
Country with strongest regional identity
Of this sample, Germany scores highest on regional identity. Very few Germans identify themselves with their country. Besides, the lowest values for spatial effects and no significant spatial spillovers are found. Hence, also this subgroup consisting of one country confirms the hypothesis that stronger regional identity leads to lower spatial spillovers.
Regarding the other factors discussed, Germany has a strongly decentralized government, as the strength of regional identity would suggest. Interregional disparities in Germany are relatively low, which might explain why there are no spillover effects found for Germany instead of negative values.
To summarize, the results of this paper confirm the hypothesis that stronger regional identity leads to lower spatial spillovers because of lower interregional migration. Furthermore, it seems that three other factors influence this relationship. The degree of decentralization strengthens the relationship as it reduces the need for interregional migration. The degree of unemployment disparities causes negative spillovers, therefore strengthening the relationship for countries with strong regional identity, and weakening the relationship for countries with low regional identity. To a small extent the average size of a region influences the relationship, because migration is easier and there are more commuting possibilities in countries with smaller regions.
Conclusion
31 spatial spillover effects between regions are lower. Using a sample of seven western European countries, it can be confirmed that in countries with stronger regional identity, spatial spillovers are lower or of a different nature than in countries with less influence of regional identity. This means that in countries with low regional identity, regional unemployment shocks will be spread out to other regions, while in countries with stronger regional identity a regional unemployment shock will have a longer-lasting effect on unemployment in the region itself. The main explanation for this relationship is related to labour mobility. There might be a difference in resistance of people to move to another region in response to unemployment shocks. People with a strong regional identity are hesitant to move to another region because they don’t feel connected to those people. In countries where regional identity is low, people might have less cultural and social barriers to move, which increases spatial spillovers between regions.
This relationship is influenced by the amount to which regional identity is translated into regional autonomy. A stronger decentralized character of the government implies that regional governments, who can better adapt to the needs and preferences of citizens, have more power. This increases welfare of the citizens in the region, thus makes them less willing to migrate. Furthermore, the regional government might be more responsive to a regional unemployment shock, which reduces the need to migrate to another region to find work. Secondly, in countries where language differences and large unemployment differentials exist, regions can even benefit from economic downturn in other regions, because firms might relocate while people are unable to migrate. At the same time, large unemployment differentials might cause a negative spillover in a country with low regional identity when this leads to a continuous stream of interregional migration.
32 in most of the countries with strong regional identity, or implementing region-specific policies. The second option might be more difficult as cultural factors form the barrier for people to move. However, stimulating the use of the national language across all regions might be an example of an effective policy to increase mobility and thus spatial spillovers within the country. Also encouraging youngers to go to university in other parts of the country can reduce the attachment to the region where they were born. However, encouraging migration is not without consequences. Policy makers who are afraid of too much migration from certain regions to others, sometimes resulting in ‘brain drains’, should focus on the first option and try to improve the regional economic situation, i.e. reduce the need for interregional migration.
The main conclusion of this paper is that there is a relationship between the strength of regional identity and spatial spillovers. This conclusion is drawn using a sample of NUTS-2 regions of seven western European countries. In further research, it would be interesting to increase the current sample with more countries, and find out how this influences the results. Furthermore, to learn more about the relationship between regional identity and spatial spillovers, different samples and different scales should be studied. As indicated, the labour force in the United States is more mobile in general, but does regional identity influence the willingness to migrate between states? Also migration between European countries in the European Union may be affected by people’s attachment to their home country, and therefore increase the vulnerability of certain European countries to an unemployment shock. Finally, there might be other factors influencing the willingness or the ability to migrate to another region. The availability of affordable housing is important in people’s migration decision. Moreover, the necessity to move from a high unemployment region could be less when generous unemployment benefits are provided. Thus, the housing market and social security generosity would be interesting factors to include in following studies as well.
33
References
Albrechts, L. (2001). Devolution, regional governance and planning systems in Belgium. International planning studies, 6(2), 167-182.
Allum, P. A., & Amyot, G. (1970). Regionalism in Italy: old wine in new bottles?. Parliamentary
affairs, 24(1), 53-78.
Aughey, A. (2010). National identity, allegiance and constitutional change in the United Kingdom. Nations and Nationalism, 16(2), 335-353.
Blanchard, O.J., Katz, L.F. (1992). Regional evolutions. Brookings Papers on Economic Activity,
1, 1-75.
Brancati, D. (2008). The origins and strengths of regional parties. British Journal of Political
Science, 38(1), 135-159.
Brand, C. J. M., & Sijs, N. (2007). Geen taal, geen natie: parlementaire debatten over de relatie tussen de Nederlandse taal en de nationale identiteit.
Cappellin, R. (1997). Federalism and the network paradigm: guidelines for a new approach in national regional policy. Regional Governance and Economic Development. London:
Pion, 47-67.
Cornips, L., & Stengs, I. (2010). Regionale identiteit. idee.
Cracolici, M. F., Cuffaro, M., & Nijkamp, P. (2009). A spatial analysis on Italian unemployment differences. Statistical Methods and Applications, 18(2), 275-291.
De Grauwe, P., & Vanhaverbeke, W. (1993). Is Europe an optimum currency area? Evidence from regional data. Policy issues in the operation of currency unions, 111-29.
Decressin, J., & Fatás, A. (1995). Regional labor market dynamics in Europe. European
Economic Review, 39(9): 1627-1655.
34 September 1, 2000, Barcelona, Spain.
European Commission. NUTS Nomenclature of territorial units for statistics. Retrieved 25-03-2018 from http://ec.europa.eu/eurostat/web/nuts/background
European Values Survey. Data and Documentation - Survey 2008. Retrieved from http://www.europeanvaluesstudy.eu/page/data-and-documentation-survey-2008.html European Union (2017). Europe 2020 indicators – employment. Retrieved 24-03-2018 from
http://ec.europa.eu/eurostat/statistics-explained/index.php/Europe_2020_indicators_-_employment
Green, A., Hogarth, T., & Shackleton, R. (1999). Longer distance commuting as a substitute for migration in Britain: A review of trends, issues and implications. International Journal of
Population Geography, 5(1), 49-67.
Hogwood, P., (2000) Citizenship controversies in Germany: The twin legacy of Völkisch nationalism and the Alleinvertretungsanspruch , German Politics, 9:3, 125-144.
Johnson, N. (1973). Government in the Federal Republic of Germany: The executive at work. Oxford/New York: Pergamon Press.
Layos, L. A. D. L. F., & António, L. (2007). Passenger mobility in Europe. Eurostat. http://epp. eurostat. ec. europa. eu/cache/ITY_OFFPUB/KS-SF-07-087/EN/KS-SF-07-087-EN. PDF (last access 29/02/2009).
Lessmann, C. (2009). Fiscal decentralization and regional disparity: evidence from cross-section and panel data. Environment and Planning A, 41(10), 2455-2473.
Llera, F.J., (2009) Spain: Identity Boundaries and Political Reconstruction. Nationalism and
Ethnic Politics, 15:3-4, 305-335.
Marston, S. T. (1985). Two views of the geographic distribution of unemployment. The
35 Martinez‐Vazquez, J., Lago‐Peñas, S., & Sacchi, A. (2017). The impact of fiscal decentralization:
A survey. Journal of Economic Surveys, 31(4), 1095-1129.
McArthur, D. P., Thorsen, I., & Ubøe, J. (2010). A Micro‐Simulation Approach to Modelling Spatial Unemployment Disparities. Growth and Change, 41(3), 374-402.
Molho, I., 1995. Spatial autocorrelation in British unemployment. Journal of Regional Science,
35 (4), 641–658.
Oates, W. E. (1993). Fiscal decentralization and economic development. National tax
journal, 46(2), 237-243.
Obstfeld, M., & Peri, G. (1998). Regional non-adjustment and fiscal policy. Economic
Policy, 13(26), 206-259.
OECD, Fiscal Decentralization Database (2016). Retrieved 22-04-2018 from: http://www.oecd.org/tax/federalism/fiscal-decentralisation-database.htm#C_3
Overman, H.G., & Puga, D. (2002). Unemployment clusters across Europe’s regions and countries. Economic Policy, 34, 115-147.
Paasi, A. (1986). The institutionalization of regions: a theoretical framework for understanding the emergence of regions and the constitution of regional identity. Fennia-International
Journal of Geography, 164(1), 105-146.
Paasi, A. (2011). The region, identity, and power. Procedia-Social and Behavioral Sciences, 14, 9-16.
Paasi, A. (2013). Regional planning and the mobilization of ‘regional identity’: from bounded spaces to relational complexity. Regional Studies, 47(8), 1206-1219.
36 Pissarides, C. A., & McMaster, I. (1990). Regional migration, wages and unemployment: empirical evidence and implications for policy. Oxford economic papers, 42(4), 812-831. Raagmaa, G. (2002). Regional identity in regional development and Planning1. European
planning studies, 10(1), 55-76.
Renkow, M. (2003). Employment growth, worker mobility, and rural economic development. American Journal of Agricultural Economics, 85(2), 503-513.
Renkow, M., & Hoover, D. (2000). Commuting, Migration, and Rural‐Urban Population Dynamics. Journal of regional science, 40(2), 261-287.
Romani, J., Suriñach, J., & Artiís, M. (2003). Are commuting and residential mobility decisions simultaneous?: The case of Catalonia, Spain. Regional Studies, 37(8), 813-826.
Schafer, A., & Victor, D. (2000). The future mobility of the world population. Transportation
Research Part A, 34(3), 171-205.
Schmalenbach, H., (1922), Die Socziologische Kategorie des Bundes, Die Dioskuren, Vol. I, München: 35-105.
Tönnies, F. (1955). Community and association:(Gemeinschaft und gesellschaft). Routledge & Paul.
Treyz, G. I., Rickman, D. S., Hunt, G. L., & Greenwood, M. J. (1993). The dynamics of US internal migration. The review of Economics and Statistics, 209-214.
Vega, S. H., & Elhorst, J. P. (2016). A regional unemployment model simultaneously accounting for serial dynamics, spatial dependence and common factors. Regional Science and Urban
Economics, 60, 85-95.
37 Appendix Belgium sigma2_e .872425 .0695132 12.55 0.000 .7361816 1.008668 Variance rho .7083808 .0319156 22.20 0.000 .6458274 .7709343 Spatial Ureglag -.4222899 .0452451 -9.33 0.000 -.5109686 -.3336111 Wx Unatlag11 -.096078 .068644 -1.40 0.162 -.2306176 .0384617 Unatlag10 .0246027 .0682974 0.36 0.719 -.1092578 .1584632 Unatlag9 .0362543 .0671201 0.54 0.589 -.0952986 .1678073 Unatlag8 -.0993152 .070185 -1.42 0.157 -.2368752 .0382449 Unatlag7 .0350898 .067475 0.52 0.603 -.0971589 .1673384 Unatlag6 -.1371542 .0685588 -2.00 0.045 -.2715271 -.0027814 Unatlag5 -.1686616 .0390195 -4.32 0.000 -.2451384 -.0921848 Unatlag4 -.127997 .0442448 -2.89 0.004 -.2147152 -.0412789 Unatlag3 -.0926186 .0439019 -2.11 0.035 -.1786646 -.0065725 Unatlag2 -.0659416 .0388278 -1.70 0.089 -.1420427 .0101596 Unatlag1 -.0340915 .0423079 -0.81 0.420 -.1170134 .0488304 Unat11 -.0410403 .0561912 -0.73 0.465 -.151173 .0690923 Unat10 .0463756 .0592151 0.78 0.434 -.0696839 .1624351 Unat9 -.0898638 .060409 -1.49 0.137 -.2082633 .0285357 Unat8 -.1731814 .0559746 -3.09 0.002 -.2828895 -.0634733 Unat7 -.0777463 .0600947 -1.29 0.196 -.1955296 .0400371 Unat6 .1908129 .0642255 2.97 0.003 .0649333 .3166925 Unat5 .1716877 .0627828 2.73 0.006 .0486357 .2947396 Unat4 .1001828 .0608244 1.65 0.100 -.0190307 .2193964 Unat3 .0907452 .0579081 1.57 0.117 -.0227526 .2042429 Unat2 .0466232 .064574 0.72 0.470 -.0799396 .173186 Unat1 -.0252666 .0779319 -0.32 0.746 -.1780103 .1274771 Ureglag .4939997 .0371738 13.29 0.000 .4211404 .566859 Main Ureg Coef. Std. Err. z P>|z| [95% Conf. Interval] Log-likelihood = -490.9881 Mean of fixed-effects = 2.5539 overall = 0.7664 between = 0.9051 R-sq: within = 0.3863
38 Germany Unat38 .0918577 .2181364 0.42 0.674 -.3356817 .5193971 Unat37 .1276928 .2181888 0.59 0.558 -.2999493 .5553349 Unat36 .1028399 .2181633 0.47 0.637 -.3247522 .5304321 Unat35 .0370133 .2181576 0.17 0.865 -.3905678 .4645944 Unat34 .0090765 .2184769 0.04 0.967 -.4191304 .4372834 Unat33 .0985394 .2181665 0.45 0.652 -.329059 .5261378 Unat32 -.0215938 .2181749 -0.10 0.921 -.4492088 .4060211 Unat31 .0336122 .2182097 0.15 0.878 -.3940709 .4612953 Unat30 -.1051183 .2184387 -0.48 0.630 -.5332503 .3230138 Unat29 -.0639325 .2182283 -0.29 0.770 -.4916521 .3637871 Unat28 .1686685 .218173 0.77 0.439 -.2589427 .5962796 Unat27 -.0368263 .2181937 -0.17 0.866 -.4644781 .3908255 Unat26 .1999737 .2185639 0.91 0.360 -.2284037 .6283511 Unat25 .0276158 .2181468 0.13 0.899 -.3999442 .4551757 Unat24 .0890886 .2181298 0.41 0.683 -.3384379 .5166152 Unat23 .0117273 .2181403 0.05 0.957 -.4158197 .4392744 Unat22 .0945429 .218157 0.43 0.665 -.3330369 .5221227 Unat21 .0391019 .2185008 0.18 0.858 -.3891518 .4673555 Unat20 .0414154 .2182528 0.19 0.849 -.3863522 .4691829 Unat19 -.0391463 .2210159 -0.18 0.859 -.4723295 .3940369 Unat18 -.0177198 .2181608 -0.08 0.935 -.4453072 .4098676 Unat17 .2075627 .2181599 0.95 0.341 -.2200229 .6351483 Unat16 -.0308119 .2181957 -0.14 0.888 -.4584677 .3968438 Unat15 .0478759 .2181789 0.22 0.826 -.3797469 .4754987 Unat14 -.0310459 .2181536 -0.14 0.887 -.4586191 .3965273 Unat13 -.052083 .2181219 -0.24 0.811 -.4795941 .3754282 Unat12 .0816499 .2181533 0.37 0.708 -.3459228 .5092226 Unat11 .056787 .2181871 0.26 0.795 -.3708517 .4844258 Unat10 .0459188 .2181331 0.21 0.833 -.3816142 .4734517 Unat9 -.1092434 .2217557 -0.49 0.622 -.5438765 .3253898 Unat8 .0275065 .2181683 0.13 0.900 -.4000956 .4551086 Unat7 -.0303353 .2181928 -0.14 0.889 -.4579853 .3973148 Unat6 .0157333 .2201997 0.07 0.943 -.4158502 .4473168 Unat5 .0863154 .2181357 0.40 0.692 -.3412227 .5138535 Unat4 .1371418 .2184383 0.63 0.530 -.2909895 .5652731 Unat3 .0851229 .2191472 0.39 0.698 -.3443977 .5146436 Unat2 -.1058574 .2186864 -0.48 0.628 -.5344747 .32276 Unat1 .0403645 .2181427 0.19 0.853 -.3871873 .4679163 Ureglag .7041534 .0213254 33.02 0.000 .6623563 .7459505 Main Ureg Coef. Std. Err. z P>|z| [95% Conf. Interval] Log-likelihood = -2151.4740 Mean of fixed-effects = 0.3965 overall = 0.6977 between = 0.6237 R-sq: within = 0.7200
40 France sigma2_e .6838558 .0381712 17.92 0.000 .6090417 .75867 Variance rho .2893127 .0457149 6.33 0.000 .1997131 .3789123 Spatial Ureglag .1843829 .0545091 3.38 0.001 .0775471 .2912187 Wx Unatlag21 -.6160888 .1040903 -5.92 0.000 -.8201021 -.4120754 Unatlag20 -.7703923 .105547 -7.30 0.000 -.9772605 -.563524 Unatlag19 -.7667588 .1016958 -7.54 0.000 -.9660789 -.5674386 Unatlag18 -.7664224 .1055724 -7.26 0.000 -.9733406 -.5595043 Unatlag17 -.8028746 .1016554 -7.90 0.000 -1.002116 -.6036337 Unatlag16 -.8466308 .109309 -7.75 0.000 -1.060872 -.632389 Unatlag15 -.531539 .1012978 -5.25 0.000 -.7300791 -.332999 Unatlag14 -.7749965 .1004897 -7.71 0.000 -.9719527 -.5780403 Unatlag13 -.7255034 .1017942 -7.13 0.000 -.9250164 -.5259903 Unatlag12 -.8540268 .1025932 -8.32 0.000 -1.055106 -.6529478 Unatlag11 -.545177 .1035376 -5.27 0.000 -.748107 -.3422469 Unatlag10 -.8833435 .1015643 -8.70 0.000 -1.082406 -.684281 Unatlag9 -.7380581 .1001121 -7.37 0.000 -.9342742 -.5418419 Unatlag8 -.7941098 .103515 -7.67 0.000 -.9969954 -.5912242 Unatlag7 -.7157891 .1015633 -7.05 0.000 -.9148496 -.5167286 Unatlag6 -.808298 .1017274 -7.95 0.000 -1.00768 -.6089159 Unatlag5 -.6995053 .1005937 -6.95 0.000 -.8966654 -.5023453 Unatlag4 -.6643585 .1033671 -6.43 0.000 -.8669543 -.4617628 Unatlag3 -.7925058 .1011209 -7.84 0.000 -.9906991 -.5943126 Unatlag2 -.7034154 .1009126 -6.97 0.000 -.9012004 -.5056304 Unatlag1 -.5033242 .1018057 -4.94 0.000 -.7028597 -.3037888 Unat21 .5178933 .1769031 2.93 0.003 .1711696 .8646171 Unat20 1.082585 .1756526 6.16 0.000 .7383121 1.426858 Unat19 .4941603 .175715 2.81 0.005 .1497652 .8385553 Unat18 .8705087 .1780208 4.89 0.000 .5215944 1.219423 Unat17 .8223972 .1745837 4.71 0.000 .4802194 1.164575 Unat16 1.099477 .1790278 6.14 0.000 .7485893 1.450365 Unat15 .5956101 .1748898 3.41 0.001 .2528324 .9383878 Unat14 .8245361 .1756524 4.69 0.000 .4802637 1.168809 Unat13 .6225519 .1754931 3.55 0.000 .2785918 .966512 Unat12 1.091033 .1737542 6.28 0.000 .7504807 1.431585 Unat11 .4552969 .1775342 2.56 0.010 .1073362 .8032575 Unat10 .8991335 .1753257 5.13 0.000 .5555014 1.242766 Unat9 .5517879 .1761174 3.13 0.002 .2066041 .8969716 Unat8 1.012046 .1766998 5.73 0.000 .6657204 1.358371 Unat7 .8746663 .1749803 5.00 0.000 .5317111 1.217621 Unat6 .5169823 .1761971 2.93 0.003 .1716424 .8623222 Unat5 .7202767 .1751342 4.11 0.000 .37702 1.063533 Unat4 .6517683 .1763569 3.70 0.000 .306115 .9974216 Unat3 .6481035 .1757561 3.69 0.000 .3036279 .9925791 Unat2 .6717107 .174309 3.85 0.000 .3300713 1.01335 Unat1 .6431868 .1759608 3.66 0.000 .29831 .9880637 Ureglag .4515535 .035311 12.79 0.000 .3823452 .5207618 Main Ureg Coef. Std. Err. z P>|z| [95% Conf. Interval] Log-likelihood = -806.4610 Mean of fixed-effects = 0.6002 overall = 0.6829 between = 0.7009 R-sq: within = 0.7447
42 Italy Unatlag18 -.6634632 .1241543 -5.34 0.000 -.9068012 -.4201251 Unatlag17 -.3080158 .1218809 -2.53 0.011 -.5468979 -.0691336 Unatlag16 -.4501779 .1241969 -3.62 0.000 -.6935993 -.2067565 Unatlag15 -.7466951 .1345891 -5.55 0.000 -1.010485 -.4829053 Unatlag14 -1.371494 .1309934 -10.47 0.000 -1.628236 -1.114751 Unatlag13 -.7325013 .1406791 -5.21 0.000 -1.008227 -.4567752 Unatlag12 -.5073331 .121815 -4.16 0.000 -.7460862 -.2685801 Unatlag11 -.1052412 .1355259 -0.78 0.437 -.3708671 .1603847 Unatlag10 -.3072109 .1258774 -2.44 0.015 -.5539261 -.0604957 Unatlag9 -.3457513 .1221186 -2.83 0.005 -.5850994 -.1064032 Unatlag8 -.3605467 .1241697 -2.90 0.004 -.6039148 -.1171786 Unatlag7 -.5585816 .1301005 -4.29 0.000 -.8135738 -.3035894 Unatlag6 -.5141599 .1215645 -4.23 0.000 -.7524219 -.2758979 Unatlag5 -.4850116 .1226337 -3.95 0.000 -.7253693 -.2446539 Unatlag4 -.7487853 .1387191 -5.40 0.000 -1.02067 -.4769009 Unatlag3 -.5526971 .1491349 -3.71 0.000 -.8449962 -.260398 Unatlag2 -1.359752 .1467493 -9.27 0.000 -1.647376 -1.072129 Unatlag1 -.8226687 .1249683 -6.58 0.000 -1.067602 -.5777353 Unat20 .3554718 .1652019 2.15 0.031 .031682 .6792615 Unat19 -.0025467 .1669456 -0.02 0.988 -.3297541 .3246608 Unat18 .646288 .1660993 3.89 0.000 .3207394 .9718366 Unat17 .2296329 .1652177 1.39 0.165 -.0941879 .5534537 Unat16 .4596054 .1665426 2.76 0.006 .1331878 .7860229 Unat15 1.501068 .1632902 9.19 0.000 1.181025 1.821111 Unat14 1.872445 .1630603 11.48 0.000 1.552853 2.192038 Unat13 .9398106 .1734145 5.42 0.000 .5999244 1.279697 Unat12 .5324064 .1653425 3.22 0.001 .2083411 .8564716 Unat11 .2528876 .1719299 1.47 0.141 -.0840888 .5898641 Unat10 .154001 .1675027 0.92 0.358 -.1742982 .4823002 Unat9 .3837808 .1651152 2.32 0.020 .060161 .7074006 Unat8 .5336356 .1662087 3.21 0.001 .2078725 .8593987 Unat7 .5681876 .1689642 3.36 0.001 .2370238 .8993513 Unat6 .5490488 .1645513 3.34 0.001 .2265342 .8715634 Unat5 .327258 .1656489 1.98 0.048 .0025921 .6519238 Unat4 1.160857 .1713666 6.77 0.000 .8249844 1.496729 Unat3 1.312205 .1776382 7.39 0.000 .9640406 1.66037 Unat2 .860617 .1794839 4.79 0.000 .508835 1.212399 Unat1 .7454362 .1659788 4.49 0.000 .4201236 1.070749 Ureglag .6015361 .0305424 19.70 0.000 .5416741 .6613981 Main Ureg Coef. Std. Err. z P>|z| [95% Conf. Interval] Log-likelihood = -980.3140 Mean of fixed-effects = 0.3183 overall = 0.9183 between = 0.9548 R-sq: within = 0.7756
