Does Being Welcome Matter to Foreigners?

Transcription

1 Does Being Welcome Matter to Foreigners? Regional Unemployment, Migrant-Open Climate and Immigration in Germany Vera Guill This version: July 11, Work in Progress, Comments Welcome. Please Do Not Cite or Circulate Without My Permission. As immigration increases globally, people become less fond of immigrants in their host countries. This concerns reseaerchers but does it also matter to immigrants? While labour economists analyse if foreign migrant inflow affects domestic unemployment (and wage rates), migrant researchers are concerned with the local unemployment rates as pull factors for future foreign immigration. Little attention has so far been given to the political climate towards foreigners as an important pull factor. My work contributes to filling this gap. I argue that potential migrants prefer areas that are politically more migrationfriendly. This is because a more migrant-open climate indicates to foreigners a greater potential for co-existing and cooperating with non-foreigners as well. Thus, migrants do not only base their location decision on economic grounds and along network ties. They also consider how welcome they are in a specific region. Ceteris paribus, foreigners prefer more migrant-open regions. Accounting for both unemployment and immigration, I discuss how migrant-open climate mediates their inter-relation. I test my argument with a cross-lagged panel design for 402 German counties between 1995 and Preliminary results suggest that for fixed levels of unemployment and economic growth, a more migrant-open climate significantly increases foreign immigration by a small extent. Unemployment however increases migrant-open climate. This falsifies that unemployment increases fear of foreign labour competition induced by increasing foreign immigration. 1 Introduction Immigration rates in OECD countries differ. While some economically less succesful countries experience rather high immigration rates, some more succesful countries experience less inflow. This is an empirical puzzle that concerns immigration researchers. Despite higher unemployment rates, foreign immigration into Poland is nowadays nearly twice as high as foreign migrant 5 inflow into the Czech Republic (cf. OECD, 2014a,b). Apparently, to explain immigration flows, we require more than just unemployment levels and economic wealth. The differences between Czech Republic and Poland allow for different interpretations. Czech migration policies could Affiliation: Berlin Graduate School of Social Sciences, Humboldt Universität zu Berlin, Institut für Sozialwissenschaften, Unter den Linden 6, Berlin, Germany address: vera.guill[at]hu-berlin.de 1

2 allow for less inflow than Polish ones. Also, it could be that the Polish population is more migrant-open. To test if migrant-open climate matters, an analysis that controls for migration policies is necessary. Ex-ante, these are controlled by focusing on one country. Germany serves well for studying this phenomenon. While unemployment is generally high, regional disparities are large. Unemployment is higher in the Northern counties, and immigration is lower than in 5 the Southern parts of Germany. Foreign unemployment is highest in the middle of Germany. Over time, regional unemployment varies as well as economic recessions hit counties to different extents. Globally, unemployment in the former GDR states is high and foreign influx is low, manifesting economic differences between East and West German counties. Within Germany, differences in immigration levels exist across and within counties. In 1990, five new federal 10 states joint the old West German states to found the Federal Republic of Germany. More than twenty years later, economists still consider East Germany a different economy than West Germany. The East has experienced higher aggregate unemployment levels, and less economic growth. These aggregate figures are used to explain why significantly less foreign migrants live in and move to East Germany. However, on the disaggregated level, we nowadays find 15 East German counties that experience similar economic conditions to structurally weak Western counties. Still, the Eastern counties experience far less foreigners, and far less foreign immigration. Standard theories of migration as well as economic theories of the labour market fall short with respect to explaining this. The research question is: Why do foreign migrants move into a specific region? I argue that 20 foreign migrants on average rather move to areas where they are more welcome because this enables them to better co-exist and co-operate with the domestic population. While choosing their new location, foreign migrants are not only pulled by economic conditions. Instead, they also consider how welcome they are. The political climate towards immigrants is a neglected factor. 25 This paper analyses immigration and unemployment on the county level. This has not been done so far. Most studies either compare single countries over time, or a cross-section of countries. The county level lends itself to an analysis as certain effects are already controlled for. Changes in labour market regulations as well as changes in migration policies are regulated at the federal level. As these policies affect all counties simultaneously, they are ex-ante controlled 30 for. In what follows, I shall first discuss the literature on economic pull factors of migration, network effects and xenophobia. Also, I consider the labour economic literature on whether foreign influx increases local unemployment. The relationship between unemployment and migrant influx is endogenous. In section 2.2, I postulate my own argument. The empirical model is very close to the theoretical one. I describe this link, the data structure, and the main variables 35 in section 3. The data set consists of county election outcomes, economic statistics, and demographics for 402 German counties between 1995 and I estimate a cross-lagged panel model to account for the endogenous relationship between unemployment and immigration (section 4). In a second model, I add migrant-open climate and control variables in line with network theories and neoclassical migration theories. In the penultimate section 5, I discuss 40 limitations and next steps. The final section concludes. 2

3 2 Theory, and Previous Research In this section, I firstly discuss what factors determine foreigners immigration into a receiving region. Mainly, these are economic considerations that increase migrants potential income as well as information about regions that travels along network lines. Secondly, I describe the endogenous impact of immigration and unemployment. While immigration theoretically may 5 increase unemployment, the unemployment impact on immigration is stronger and stable. I then present my own theoretical contribution: I argue that while deciding on their new location, potential migrants consider how welcome they are. The migrant-open climate influences their choice of region. Migrant-open climate however is not exogenous: it depends on unemployment and on previous immigration influx. I discuss the perceived threat of immigration for domestic 10 workers. Migrant-open climate is thus also of interest to researchers who study the relationship between unemployment and immigration. Figure 1 depicts the theoretical model with the three main variables. I describe each step in the following. employment prospectus as pull factor Unemployment Immigration substitute or complement to domestic labour Political Climate Figure 1: Theoretical model 2.1 What Drives Immigration? The literature on immigration flows and settlement patterns consists of two non-competing main 15 strands. Network theories of immigration consider sociodemographic factors that enable information flows among immigrants along network lines. Neoclassical migration theories on the other hand distinguish mainly economic conditions in (sending and) receiving regions as (push and) pull factors. I focus on economic pull factors. In sum, both network effects and economic conditions play a role when migrants decide on their new area of residence. 20 3

4 Network theories of migration Social network theories consider social ties between foreigners that already live in a receiving country and their relatives and friends in their countries of origin. Along these individual links, information about regions such as available housing, school quality, and the general infrastructure as well as potential open job positions are communicated. Such, foreigners are more likely to move to regions where other migrants with the 5 same cultural background live (cf. Spittel, 1998). These social networks also increase support for new immigrants. Knowing someone who migrates increases one s own chance of migrating to that area because the economic costs are reduced (Greenwood, 1969, 1970). These costs consist of economic search costs for identifying and getting potential jobs, but also of the monetary costs for actually moving, psychological costs due to leaving a known environment and 10 living in a new one, and opportunity costs (cf Massey, 1990, 8-9). Depending on the traveled distance between origin and receiving country, these costs increase. Social networks of migration, i.e. personal ties to experienced previous immigrants, in turn reduce all these costs, thus making immigration cheaper. (Cornelius, 1975; Lomnitz, Lomnitz and Wolf, 1977; Massey and Espana, 1987). 15 Even if one seeks no current personal ties nor future ones, presence of other foreigners provides cheap information on how simple it may be to live in that area. Thus, we would expect higher immigration into regions with higher shares of foreigners. Neoclassical migration theory Neoclassical migration theories discuss immigration in terms of factors that push immigrants out of their domestic regions (for instance economic 20 poverty, high unemployment rates, political instabilities), and those that pull migrants to specific regions. Language barriers (or lack thereof), economic prosperity, receiving country s income, etc. are potential attracting conditions discussed in the literature. Increasing their own potential income, immigrants move to regions where unemployment is low and pcgdp is high. (Boeri and Brucker, 2005; Constant and Massey, 2002; Borjas, 1999; Massey et al., 1993) Im- 25 migration is thus partly determined by unemployment rates and local wage rates. Endogeneity: Immigration Aects Unemployment While unemployment decreases immigration, immigration in turn may also increase unemployment. This is because immigration increases labour supply in a receiving region. The higher labour supply may not be directly matched by an increase in labour demand, which leads to either wage decreases or an 30 increase in unemployment. (Since Germany has relatively high job protection rules, an increase in unemployment is the more likely consequence.) Okkerse (2008) and Longhi, Nijkamp and Poot (2008) provide a detailed discussion. Since an increase in the workforce could lead to an excessive supply, immigration would increase local unemployment rates. Against this stands the argument that immigration also increases domestic consumption, need for housing, increase 35 potential for innovations, and even create new jobs. Thus, immigrants would not only increase labour supply, but also labour demand. The empirical findings are mixed. For an overview, see Okkerse (2008). Many studies, of which three have explicitly modeled endogeneity, find insignificant effects for moderate immigration levels but cannot rule out intermediate rise in unemployment due to mass inflows. 40 4

5 2.2 Migrant-Open Climate Matters To Immigrants In sum, migrants are not only pulled to specific potential new regions of residence by economic conditions, but they rely on information provided by social networks as well. The impact of unemployment on immigration is a stable finding. I argue that migrant-open climate also matters. People differ in their attitudes towards foreigners. Comparing individual survey results on 5 own positions and perceived ones from national parties from the European Election Study 1994, van der Brug, Fennema and Tillie (2000) show that the motives for voting for right-winged parties are similar to those for established parties: Close positions, and pragmatic consideration (i.e. considering a prefered party s size and importance in parliament may lead to voting for a less preferred but bigger party). van der Brug, Fennema and Tillie (2000) show that votes 10 cast for any extreme right party actually grasp anti-immigrant sentiments within the electorate. Especially, these votes are not motivated by protesting against more established parties (cf. van der Brug, Fennema and Tillie, 2000, 82) While it is a stable finding that certain general information about specific regions may travel along network lines, the same could plausibly hold for information about the local climate 15 towards migrants. This may range from rather xenophobic to migrant-friendly. Why migrants care about migrant-open climate Migrants prefer more migrant-friendly areas. Personal well-being and feeling of personal security in everyday life is not only affected by absence of violence, for example attacks against other foreigners. There are also everyday subtle dislikes. Their presence may feed personal fear of discrimination. In areas with 20 less migrant-open climate, immigrants have a higher probability of being discriminated against. More importantly, even if one attributes a low probability of being attacked, getting help by others seems more likely when those others do not discount your value based on unchangeable characteristic (being foreign). A more migrant-open climate increases the chance for immigrants to peacefully co-exist and maybe even cooperate with the domestic population. These 25 individual migrant decisions aggregate to less inflow into regions with less migrant-open climate. 2.3 Migrant-Open Climate is Endogenous Partly, migrant-open climate depends on external factors. Despite the mostly non-significant impact of immigrant influx on domestic unemployment, there is empirically a persistent fear 30 that immigrants take away jobs from domestic workers. (cf Center, 2006) In this subsection, I briefly describe Campbell (1965) s theory of realistic group conflict. I relate to this concept to understand available jobs as a resource on which immigrants pose a threat. I then present research on immigration and anti-immigrant voting. This shows that voters respond to increasing immigration by voting less migrant-open. 35 Perceived Competition results in Realistic Group Conict Campbell (1965) argues that for groups that compete over a certain good, the perceived threat that the others removes it is sufficient for conflict to arise. These conflicts are always zero-sum games. Esses, Jackson and Armstrong (1998) adapt this theory of realistic group conflict to determine attitudes towards 5

6 foreigners. Dependent on how a government responds to remove these conflicting interests over resources, people become more xenophobic. Esses, Jackson and Armstrong (1998) find that increasing migration rates are met with increasing migration aversion in host countries. While people migrate as individuals or in families, they may be perceived as group members because of the network effects of migration. This way, labour does not need to be a scarce resource 5 for conflict and less migrant-open attitudes to arise. Perceived competition is sufficient to drive attitudes. Local unemployment rates affect voters well-being. Unemployment has many individual negative consequences. Higher unemployment rates increases their perceived risk of becoming unemployed as well due to increased competition, and of having to bear welfare constraints due to others increasing unemployment risks. Job positions are a scarcer resource in 10 counties with higher unemployment rates. This increases the perceived threat posed by immigration. Anti-Immigrant Voting depends on Immigration and Unemployment Voting for parties that welcome or oppose migration is partly exogenous as it rests on voter heterogeneity. Partly, it is endogenous to the regional and individual (perceived) economic situation of current 15 and future income. Since immigration may increase competition for domestic workers, as found by Winter-Ebmer and Zweimüller (1999), we would expect workers to move away from areas with increasing competition. However, contrary to theoretical expectations, neither natives nor older migrants (in the United States) are more likely to leave states with higher new migrant rates than those with lower ones (Borjas and Hilton, 1996; Borjas, 2001). Additionally, Euro- 20 peans tend to be rather immobile (Boeri and Brucker, 2005, 631). This should hold especially for unemployed workers that have generally less means to move and more incentives to save money. This means that voters do not physically escape from an increase in migration, should they oppose it (in general, or because of competitive risks). Without this de facto exit option, voters have an even stronger incentive to oppose immigration, should they not want it. 25 Hatton and Williamson (2006) discuss median voter change in attitude towards migration. They find that for 14 countries the income gap between source and destination country matters. The bigger that gap, the less voters are in favour of migration (in the receiving country). Since in OECD countries, the average income increases faster than in less-developed countries, people should become more and more anti-immigrant, at least concerning low-skilled labour 30 inflow. That is for two reasons: for the median voter and every one above her income level, there is mainly the fiscal threat of increasing welfare transfers. For everyone below, labour market competition increases. Analysing survey data that links individual voter attitudes towards foreigners, nationalism, chauvinism and own experience abroad to attitudes toward migrants, and migration policies, O Rourke and Sinnott (2006) find a strong link between chauvinism, 35 racism and reluctance towards migrants. However, economic factors still play the major role. Overall, O Rourke and Sinnott (2006) find that voters want their governments to reduce migration inflows, but (except for Japanese, Philippines, and Slovaks) they favour increasing refugee inflows. In sum, both network effects and economic conditions play a role when migrants decide on 40 their new area of residence. While studies on the endogenous effects of immigration and unemployment find little support for an increase in unemployment due to foreign migrant influx, this perception is still present (see, for example Esses, Jackson and Armstrong, 1998; O Rourke and 6

7 Sinnott, 2006; Hatton and Williamson, 2006). In line with theories of realistic group conflict, job positions are a resource that are at least in the short run a zero-sum one. If a foreigner fills a vacancy, this does not directly create new jobs (for either foreigners or natives). Besides, even a perceived competition is sufficient for hostility towards out-group members.this may explain why native voters become less migrant-open. The link between unemployment and immigra- 5 tion is stable. From a political-science perspective, the political climate towards migrants is also relevant for foreigners. 2.4 Working Hypotheses Hypothesis 1 The political climate towards foreign immigrants ceteris paribus increases foreign influx into any county. 10 Hypothesis 1a Alternatively, foreign migrants do not respond to the county mood towards them. For example, if the group of foreigners is relatively big, they will less likely share points of contact with Germans. In that case, migrant-open climate would be irrelevant. Hypothesis 2 Unemployment ceteris paribus decreases migrant-open climate. 15 Hypothesis 2a Alternatively, migrant-open climate is independent of unemployment levels. This may be due to the relatively high aggregate welfare level in Germany. In the next section, I develop the research design to test my working hypotheses. 3 Data The sample population consists of 295 German counties, and 107 county-free municipalities 20 from 1995 to Observations thus exist for a cross-section of 402 units that are repeated yearly 17 times. I employ German register data (BBSR, 2013; GENESIS, 2014). 1 Values for party migration positions are generated with the Comparatitve Manifesto Database (Volkens et al., 2013). These are reported for each federal election year. Due to territorial reforms, some county areas have changed over time. All collected data is thus converted to the territorial status 25 of 31 Dec All count variables are collected on 31 Dec of the reported year. In the first part, I operationalise migrant-open climate as the weighted sum of party vote shares that grasp different attitudes towards immigration. This is the (party-weighted) migration position for each county. Subsequently, I describe the main variables, their mean values and distribution, and I provide a cross-county comparison of unemployment, foreign migrant-influx, 30 and migrant-open climate in To account for possible endogeneity bias, I estimate unemployment and immigration in a cross-lagged structural equation model. In a second model, I include the party-weighted migration position as dependent on unemployment and immigration, 1 The shapefiles to combine my data with a map of German counties are provided by Bundesamt für Kartographie und Geodäsie (2014). 7

8 and influencing immigration. I control for variables based on network theories and neoclassical immigration theories. Preliminary results suggests that a more migrant-friendly climate ceteris paribus increases immigration into a county. 3.1 Migrant-Open Climate: Operationalisation as Party-Weighted Migration Position 5 Migrant-open climate consists of the number of violent xenophobic acts, discrimination against foreigners, voting for extreme-right parties, migration policies passed by parliaments. These attitudes may be captured by police statistics, survey results on worries about immigrants, and vote shares for parties with different immigration positions. I focus on the latter. Party-weighted migration position is a constructed measure for the aggregate county attitude towards foreign 10 migrants in any county. I measure German federal party positions with the the Comparative Manifesto Project database (Budge, 1982; Volkens et al., 2013). Parties formulate diverging statements about immigration. In line with salience theory of party competition Budge et al. (2001); Budge and Farlie (1981), I assume that these different positions are not postulated as direct conflicts. Rather, parties offer competing formulations of similar topics to their potential 15 voters. Immigration Position = Positive Statements - Negative Statements. I add favourable manifesto statements about multiculturalism (per607) and minority groups (per705), and substract negative statements about multiculturalism (per608) and positive ones about a national way of life (per601), and law and order (per605). The latter includes references to illegal 20 immigration, especially when addressed by an extreme-right party. Findings are not sensitive to exclusion of law and order, see table 7. (cf. Arzheimer and Carter, 2006; Arzheimer, 2009; Alonso and Fonseca, 2012; Schmidt and Spies, 2013) Since the German extreme right parties NPD, DVU and REP have never gained seats in a federal parliament since 1949, their manifesto values are missing. I replace these missing 25 values with the value for Danish People s Party. This is another European extreme right party with similarly strong anti-immigrant positions. I do not expect German extreme right party values to vary over time similar to DPP. Hence, I substitute the missings with the most negative value DPP has reached in 2001: -40. All other small parties that have missing values are measured as having a neutral opinion on immigration. 30 The resulting party-values are multiplied by the local party vote shares in county elections. That way, I account for larger parties gaining more attention in media coverage, and the proxy for migrant-open climate is weighted accordingly. (cf. Schmidt and Spies, 2013) Depending on the vote shares any party wins, they receive more financial means to distribute their values. Moreover, once they enter county council, local politicians are subject to more local newspaper 35 coverage. Indirectly so, when council decisions are communicated. Directly, when they take part in public discussions, or run for the office of county mayor. Vote shares for county councils capture the mood towards foreigners (which then travel along network lines). Migration position is the weighted sum of election vote shares for each county depending on the parties migration positions. I standardise the values. Migration positions ranges from

9 (least migrant open county) to 10. County elections in Germany take place on average every 4 to 6 years. Missing county-year values are carried forward until the next election year. 3.2 Main variables Foreign (and German) Inux Influx generally refers to foreign migrant inflow. Foreign influx is the overall entrance of foreigners into any county in a specific year. I do not differentiate 5 between foreigners coming from abroad and foreigners arriving from other counties. To control for population size, the total number is multiplied by 1000 over the sum of all county inhabitants in the same year. Thus, foreign influx measures foreigners inflow per 1000 inhabitants (foreign or not). German influx is quantified similarly to foreign influx. Unemployment Unemployment is measured as the number of persons without employment 10 by the working-age population, i.e. inhabitants between 15 and 65 years. To describe percentages, this share is multiplied by In line with neoclassical theories of migration, we expect less influx into counties with higher unemployment rates. Migration Position Party-weighted migration position ranges from 0 for counties with the least migrant-open climate to 10 for the most migrant-open counties. Missing county-year 15 values are carried forward until the next election year. Foreign share Foreign share is the number of foreigners in a county divided by the overall county population in the same year in percent. Since higher numbers of foreigners in a region decrease transaction costs for potential new immigrants, we expect higher influx for higher foreign shares. 20 Tax revenue Tax revenue is a per capita measure in Euros: tax revenue by population as counted on 31 Dec in the same year. 3 Counties with higher tax revenues are richer. I assume that their inhabitants are wealthier on average as well. In line with neoclassical migration theories, we expect a higher influx into counties with higher tax revenues. 3.3 Descriptives for Germany: Aggregate, East and West 25 Germany consists of sixteen federal states. Five East German states (plus Berlin) are on the former German Democratic Republic territory. 4 2 For 1995 to 1997, the observations are based on my own calculations with the same factors, unemployment level and working-age population with BBSR-provided data. 3 I substitute 8 missing values for tax revenue in 1997 with the arithmetic mean of the previous and following year. The affected counties are Dahme-Spreewald and Prignitz in Brandenburg, Leipzig in Saxony, as well as Unstrut-Hainich-Kreis, Kyffhäuserkreis, Ilm-Kreis, Saalfeld-Rudolstadt, and Saale-Orla-Kreis in Thuringia. 4 These are, from North to South, Mecklenburg-West Pomerania, Brandenburg, Saxony-Anhalt, Saxony, and Thuringia. While parts of Berlin belonged to West Germany before re-unificiation, the provided data does not allow for a differentiation. Thus, Berlin also counts as an East German state. The ten West German states are, from North to South: Schleswig-Holstein, Hamburg, Lower Saxony, Bremen, North Rhine-Westphalia, 9

10 Table 1 summarises the variable distribution for all German counties. Overall, foreign influx varies highly between counties. The minimum influx is 1.7 foreigners per 1000 inhabitants, the maximum is The mean foreign inflow is This is above the median value (15.94). The standard deviation is The variable is not normally distributed, but skewed to the right. There are more counties with less than mean influx. Outliers are Berlin and Munich. 5 For most years, these counties experience very high foreign influx. Average foreign influx into West German counties (21.18) is more than twice as high as into East German counties (9.4). Foreign share is higher in West German counties (mean 8.6 percent, median 7.5 percent), and significantly lower in the East (mean 2.3 percent, median 1.8 percent). German influx is similar across both sub-regions: 48 Germans per 1000 county inhabitants move into West 10 German regions, while 41 per mill move into Eastern counties. The aggregate average is 47 per mill. Berlin in 1995 has experienced lowest influx of Germans (17 per mill), and it is highest in the university county-free municipality of Göttingen in 2011 (701 per mill). The average unemployment rate is 6.9 percent with a standard deviation of 3.3 percent. Unemployment is lowest in the Bavarian county of Eichstätt in 2011 (1.2 percent), and it is highest in the 15 Uckermark county in Brandenburg in 2004 (19 percent). Overall, unemployment is higher in East Germany (mean 11.9 versus 5.7 in West Germany). The average migration position is higher in East Germany than in the West. Especially in counties near Berlin, the migration position is high. On the disaggregate level, more counties with moderate migration positions exist in West Germany. 20 Table 1: Descriptive Statistics N Mean Median SD Min Max Skewness Kurtosis foreign influx unemployment political climate erp share foreign share tax revenue German influx Figure 2 displays the relative densities of unemployment, immigration, party-weighted migration positions, and tax revenue for all German counties in For this, the variable values are divided into sextiles, see table 2. The darker the shade, the higher the corresponding variable value. While unemployment is generally moderately high, regional disparities are large. Unemployment is higher in the Northern and Eastern counties, and immigration is lower than in 25 the Southern parts of Germany. Globally, unemployment in the former GDR states is high and Hesse, Rhineland-Palatinate, Saarland, Baden-Wuerttemberg, and Bavaria. Hamburg, Bremen and Berlin are city states. Their administrative unit is not only the federal state, but also the county: All three federal city states are county-free municipalities. 10

11 foreign influx is low, manifesting economic differences between East and West German counties. Still, there are counties that are very similar with respect to their economic development. Despite corresponding unemployment rates and tax revenue, these counties differ in the foreign immigration rates they experience. Foreign migrant influx is high in the harbour counties of Bremen and Hamburg, and generally higher in West Germany. The party-weighted migration 5 position is high in the North, middle West, and in some East German counties. I report bivariate distributions in the appendix, see table 6. Table 2: Sextiles for Density Area Maps, 2011 Sextile (a) Unemp. n (b) Influx n (c) Mig. Pos. n (d) Tax rev. n In the next section, I employ a structural equation model with cross-lags between unemployment and influx in German counties from 1995 to This way, I analyse how unemployment in one year correlates with influx in the year after while simultaneously accounting for the effect 10 of previous annual inflow on unemployment in the following year. Also, I control for stability effects (autocorrelation). In a second model, I estimate the effect of party-weighted migration position on immigration while controlling for foreign share, German influx, tax revenue, cross lagged effects, and autocorrelation. 4 Cross-Lagged Panel Analysis 15 Cross-lagged panel designs are a specific type of structural equation models. The latter fit coefficients with maximum-likelihood estimation: Which coefficients are most likely to represent the relationship between variables in the given data? Cross-lagged panel designs specifically model the cross-lagged effects between endogenous variables while also accounting for their autocorrelations. (cf. Kenny, 1975, 2005; Gould, Pitblado and Poi, 2010; Acock, 2013) 20 Cross-sectional data analysis is problematic for several reasons. This data represents only a snapshot at one specific timepoint, and more importantly, it does not allow for analysing changes over time. This is possible with panel data. The units are 402 counties which are significantly more than while comparing OECD countries. The data covers a period of 17 years. This relatively long time-series allows me to estimate changes within units. Estimating 25 the effect of a change in unemployment (and migrant-open climate) on a change in immigration is a stricter conceptionalisation of a causal effect. Unemployment and immigration, as well as immigration and migrant-open climate, are endogenous. This endogeneity cannot be captured with OLS estimation. 11

12 Sextiles (n) (66) (62) (74) (70) (62) (68) Sextiles (n) (67) (67) (67) (67) (67) (67) (a) Unemployment, 2011 (b) Foreign Influx, 2011 Sextiles (n) (67) (67) (67) (67) (67) (67) Sextiles (n) (67) (67) (66) (68) (67) (67) (c) Party-Weighted Migration Position, 2011 (d) Tax Revenue, 2011 Figure 2: Unemployment, Immigration, Migration Position, and Tax Revenue in Germany, Density area maps: Darker shades depict higher densities. 12

13 Figure 3: Cross-Lagged Panel Model of Unemployment and Foreign Influx To control for endogenous effects of unemployment and foreign migrant influx (see section 2.1, I model the path effects of unemployment on foreign influx and vice versa in model 1. Figure 3 depicts this. Since the data spans over no more than 17 years, I can only account for short-term effects. I estimate a one-year lag structure, and autocorrelations. The resulting coefficients that relate unemployment and influx are thus independent of the variables auto- 5 influence, and they are independent of each other. I constrain all time-variant effects. All results are thus average effects. The second model adds the effect of migrant-open voting. The empirical model is very close to my theoretical model. Since the variables are not jointly normally distributed and I cannot employ regional dummies to account for possible spatial correlations, I also estimate the models with the Huber-White 10 sandwich estimator for the variance-covariance matrix. In the second model, I also report significance levels with these robust standard errors. 4.1 Results For the first structural equation model (figure 3 and regression table 3), I estimate the crosslagged effects of unemployment and foreign influx within and across German boarders for all German counties between 1995 and The MLE model fits the data very well, overall R-squared is 99.5 percent. This high goodness of fit foremost reflects the high persistency in the autocorrelated data: Generally, unemployment in the previous year predicts unemployment in the current year with 98 percent. Foreign migrant inflow in the current year is largely determined by foreign migrant inflow in the previous year, the coefficient is For each year group 20 between 1995 and 2011, the stability coefficient for unemployment is significantly different from the stability coefficient for foreign inflow within Germany and across German boarders. The results indicate that unemployment and foreign migrant influx are stable over time. Since both variables are not normally distributed and I cannot employ regional dummies to account for possible spatial correlations in this model, I also estimate the model with the Huber- 25 White sandwich estimator for the variance-covariance matrix. The results do not change. Table 3 summarises the results. 13

14 Model 1 shows that the effect of unemployment on immigration is bigger than the reverse effect, the standardised beta coefficient is compared to Moreover, the latter effect of immigration on unemployment is not significantly different from zero. A one percent increase in unemployment decreases immigration in the following year by , i.e foreigners by county inhabitants less move into that county. A one percent increase in 5 immigration on the other hand decreases unemployment insignificantly: As one foreigner per 100 (per 1000) county inhabitants moves into the county, unemployment in the following year decreases by 1 unemployed person of ( ) inhabitants between 15 and 65 years of age, or it does not decrease at all as the effect is insignificant. Table 3: Model 1 Foreign Influx and Local Unemployment x y, all x vars: L1. coeff. beta se z p Unemp Influx Influx Influx Influx Unemp Unemp Unemp N*T 402*17 R-sqr cross-lagged coefficients differ, chi sqr(1) stability coefficients differ, chi sqr(1) The cross-lagged effects differ significantly at the 10 percent level. This means that unem- 10 ployment has a significantly bigger effect on foreign migrant influx than vice versa. This finding is in line with previous research (see section 2.1). Model 2, table 4, estimates the cross-lagged effect of unemployment and immigration, immigration and migration position, and the lagged effect of unemployment on migration position while controlling for foreign share, German influx, and tax revenue. The results are robust to 15 estimating a cross-lagged effect of migration position and unemployment. As in the first model, all effects are constrained to be time-invariant. The coefficients thus display average results over 17 years. 14

15 15 Table 4: Model 2 - Foreign Influx, Unemployment, and Migration Position Robust Standard Errors Non-Robust Standard Errors x y, all x vars: L1.** coeff. beta se z p se z p Unemp Influx Mig. Pos. Influx For. Share Influx Ger. Influx** Influx Tax Rev. Influx Influx Influx Influx Unemp Ger. Influx Unemp Unemp Unemp Unemp Mig. Pos Influx Mig. Pos For. Share Mig. Pos Ger. Influx Mig. Pos Tax Rev. Mig. Pos Mig. Pos. Mig. Pos N*T 402*17 R-sqr ** German Influx Influx without lag as control for co-movement standardised coefficients (betas) are reported for the effects from 1995 to 1996, slight variance over time due to standardisation with covariance matrix

16 The party-weighted migration position has a positive effect on foreign influx in the following year. An increase by one index unit significantly in the previous year increases influx by (1.87) immigrants per (10 000) county inhabitants, ceteris paribus. This lends support to my hypothesis that migrant-open climate increases immigration. The results are significant at the 90 percent level. 5 Foreign share increases next-year influx (0.1414) in line with network theories of migration. Compared to model 1, the effect of unemployment on immigration is insignificant. This may be because tax revenue and unemployment are partly co-determined. Tax revenue consists to a certain extent of income tax generated by the employed. For higher unemployment levels, tax revenue is lower. An increase in per capita tax revenue by 1 Euro increases influx in the 10 following year by (2.2) immigrants per ( ) county inhabitants, ceteris paribus. This is in line with neoclassical migration theories. With robust standard errors, tax revenue is insignificant. A change in German influx co-moves positively with a change in foreign influx. For each German that moves to a county, foreigners move there, ceteris paribus. This hints to- 15 wards foreigners being attracted by similar regional characteristics as Germans. Unemployment increases the party-weighted migration position in the following year by , ceteris paribus. This falsifies my hypothesis that unemployment decreases migrantopen climate. If county inhabitants become less migrant-open, it is not because of perceived labour competition. 20 Influx ceteris paribus increases the party-weighted migration-position in the following year by Thus, county inhabitants may respond positively to immigration. However, foreign share decreases the migration position by It may either be that new immigrants are different from those that already live in the county, e.g. German guest workers that immigrated in the 1970s. Alternatively, Germans may respond positively to the concept of immigration, 25 but their response turns negatively the longer foreign immigrants stay in a county. Further research is necessary to test this. The combined standardised effect of influx and foreign share on migration position is negative. This lends, albeit weak, support to domestic populations becoming less migrant-open as foreign immigration increases. Comparing both models, the lagged effect of unemployment on influx is stronger while ac- 30 counting for the county migration position. Thus, migration position is not a mediator as I hypothesised. 5 Limitations, and Next Steps 5.1 Measurement Proxy for Migrant-Open Climate Migrant-open climate is only measured as the party- 35 weighted migration position. The effect of unemployment on party-weighted migration position in the following year demonstrates a potential short-coming of the current proxy: Unemployment increases the party-weighted migration position in the following year by , ceteris paribus. Possibly, migrant-open parties are the same as those that favour high welfare spending, especially for the unemployed. Two left parties in Germany are most often identified as the 40 16

17 most migrant-open ones. In that case, the proxy for migrant-open climate captures that unemployment increases the propensity for people to vote for a left county council. Votes for the Left party are generally high in East Germany. If people vote traditionally and independent of unemployment, the same correlation could arise for East Germany. I shall improve on this index by incorporating survey values on worries about immigration. Policies on immigration 5 are decided by German Federal Parliament and on the European Union level. I shall include the corresponding county vote shares as well and test if the main results still hold. Moreover, data on xenophobic crimes is available on the state level, and could be included as well. Replacing German extreme right party positions with the most negative value for the Danish People Party is rather blunt. It would be much better to improve this value by either coding these party posi- 10 tions according to Comparative Manifesto Project criteria. Alternatively, I may use the variable as only capturing the migration position of council inhabitants that vote for moderate parties. Then, I shall include either a dummy or extrem-right party vote shares (see next paragraph) that captures if people in a county have voted for a right-winged party, and include this red-flag dummy in my estimations. Finally, this proxy variable assumes that local party s mirror their 15 federal party positions with respect to immigration. This may be wrong, and it would be great if further research tested this. Extreme-right party vote shares Extreme-right party (ERP) share measures the vote shares reached by the three German extreme-right parties NPD, DVU, and REP in all county elections. Missing year values are carried forward until the next election year. I expect that 20 higher vote shares for these party deter immigrants from moving into a county. An increase in unemployment is associated with a decrease in ERP share. This runs contrary to theoretical expectations. Further analysis is necessary to test if this correlation holds in multivariate estimations. In general, if these parties gain at least three percent of all valid votes, they enter county council. That way, their visibility is even higher. ERP vote share is zero in most counties, the 25 three lowest sextiles combined have a value of 0. The darkest sextile depicts all counties that have ERP council members as their vote share is above three percent. Fig. 4 depicts this for 2011, see table 5 for the sextile range. This variable is not yet included in the cross-lagged structural equation model because of technical limitations. It is difficult to include a variable which has most often the value 0. Moreover, in most West German counties, ERP share is 30 measured as 0. I doubt that the true value is always 0 as some statistical bureaus report values below 3 percent as missing. However, I do not have the means to test this as I have only this data provided by state statistical bureaus. While in some counties right-extreme party voting is a re-occuring event, it is very seldom for the whole time and units under study. Table 5: Sextiles, ERP, 2011 Sextile ERP Share n

18 Figure 4: Extreme-Right Party County Vote Shares in Germany, Density Area Map. Foreign share, and Foreign Inux Foreign share only captures the amount of non- Germans over the whole population. I do not have data to differentiate between non-german nationalities. Similarly, foreign influx captures immigration by non-germans, not their na- 5 tionalities. I can only assume that foreign nationals are rather attracted to areas where samenationality foreigners live. It may as well be that foreigners generally prefer areas with higher foreign shares, thus not relying on networks. Likewise, I cannot differentiate to which extent immigrants are driven to similar regions not only based on increasing their income but because they are more likely to find affordable housing and living conditions. This is because I do not 10 have information on their disposable income. Tax Revenue Tax revenue is a per capita measure. I assume that higher per capita tax revenue on average depicts higher per capita income. However, I do not have regional GINI measure to confirm this. The alternative measure for individual and reginal wealth, pcgdp, is not available on the county level before Regional distance German counties are small and they are contiguous to other counties. I expect that unemployment rates, foreign share and political climate in adjacent counties correlate spatially. While spatial effects play a role, I cannot include any measure of distance. The available migrant outflow data does not include the destination. Likewise, inflow data does not survey the original county. An alternative measure to account for spatial effects is the distance 20 between a county s capital and the nearest international port of entry. However, I do not have the data to test this. 18

19 CLPD Model All coefficients are constrained to be equal across time. The lag structure is one year. Since I have observations for 17 years, I cannot model long-term effects but I shortterm effects only. In a next step, I shall analyse the effects on German migrant influx as well. 5 Foreign influx and German influx co-move. An increase in German influx increases foreign influx by , ceteris paribus. Further analyses is required to study this. Inter-group comparisons are necessary to test if both groups are attracted by similar regional features, and if Germans respond to migrant-open climate as well. Moreover, the models lack control for foreign unemployment. Foreigners may not be as affected by general employment markets. Since 10 they are on average younger, accept lower wage rates than Germans, and may be discriminated against, foreign unmployment rates may be more important for foreigners while deciding on their new area of settlement. (cf. Kuhlenkasper and Steinhardt, 2011, 5) If foreign immigrants are employed in secondary labour markets, they do not compete with Germans and older immigrants that are employed in the first labour market. That way, we would expect a weaker effect 15 of unemployment on foreign influx, but expect less influx for higher foreign unemployment levels. Secondly, we would expect less impact of unemployment on migrant-open climate. Before testing this, the proxy for migrant-open climate needs to be improved (see first paragraph). Migrant-open climate has a significant positive effect on foreign influx even while accouting for the standard controls regional wealth and foreign share. It would be interesting to see if German 20 influx to specific regions is driven by similar factors, and if Germans are affected differently by the political climate towards foreigners. It may be that migrant-open climate matters less to Germans because they are not directly affected. On the other hand, they may as well prefer areas where foreigners are welcome because this could indicate that in general, the local receiving population deals well with strangers. 25 Since the variables are not normally distributed and I cannot employ regional dummies to account for possible spatial correlations in this model, I estimate the models with the Huber-White sandwich estimator for the variance-covariance matrix. Further diagnostics include bootstrapping and jackknife to test if outliers drive the results. 6 Conclusion 30 I argue that unemployment decreases foreign migrant influx not only directly, but mediated by the political climate towards migrants. This is because while deciding on their new settlement location, foreign migrants ceteris paribus prefer areas that are more friendly towards strangers. Information about the local migrant-open climate travels along network lines as well. Modeling cross-lagged structural equations for 402 German counties between 1995 and 2011, 35 the empirical model is very close to the theoretical one. Preliminary results suggest that in line with previous research unemployment decreases foreign immigration in the following year. Immigrants on average move to counties with a more migrant-open climate. This lends support to my first hypothesis. However, unemployment decreases migrant-open climate in the following year. This falsifies my second hypothesis. Either, perceived job competition does not depend on 40 unemployment levels, or unemployment levels do not influence migrant-open climate.the effect of unemployment on immigration is even stronger when I include migration position. Thus, the 19