Immigration, Regional Conditions, and Crime: Evidence from an Allocation Policy in Germany

Transcription

1 Immigration, Regional Conditions, and Crime: Evidence from an Allocation Policy in Germany Marc Piopiunik Jens Ruhose PRELIMINARY VERSION. Please do not cite without permission. Abstract After the collapse of the Soviet Union, more than 3 million people with German ancestors immigrated to Germany under a special law allowing immediate naturalization. Exploiting the exogenous allocation of ethnic German immigrants across regions, we find that immigration significantly increases the overall crime rate. Effects are particularly strong for burglary, property damage, and battery. Most importantly, the effects of immigration on crime depend on the socioeconomic conditions of the receiving region. Immigration increases the overall crime rate in regions with bad labor market conditions, but not in regions with good labor market conditions. Immigration s impact on crime is also stronger in regions having a high preexisting crime level. JEL-Code: F22, J15, K42, R10 Keywords: Immigration, crime, allocation policy We are grateful to George J. Borjas, Oliver Falck, Eric A. Hanushek, Ludger Woessmann, and seminar participants at the Ifo Institute in Munich for helpful comments and discussion. Ifo Institute - Leibniz Institute for Economic Research at the University of Munich, Poschingerstr. 5, Munich, Germany; piopiunik@ifo.de; Phone: Ifo Institute - Leibniz Institute for Economic Research at the University of Munich, Poschingerstr. 5, Munich, Germany; ruhose@ifo.de; Phone:

2 1 Introduction Criminal activity by immigrants is an important concern in many industrialized countries. Indeed, residents in most OECD countries are more worried about immigrants increasing crime than about them taking jobs away from natives (Bianchi et al., 2012). Thus, this widespread concern about crime might play a greater role in immigration restrictions than do labor market concerns (Card et al., 2012). While the literature on the labor market effects of immigration is huge (e.g., Friedberg and Hunt, 1995; Borjas, 1994; Card, 2005), the crime effects of immigration have remained largely unexplored. 1 This paper aims at increasing our knowledge about the impact of immigration on crime and about the role that a region s socioeconomic conditions plays in this relationship. We investigate the crime impact of a particular group of immigrants to Germany that was allocated across regions by German authorities upon arrival. Because this allocation was based on the presence of family members and noncompliance with allocation was severely sanctioned, the possibility of self-selecting into regions was highly restricted for this group of immigrants. Therefore, inflows into a particular region were not driven by local crime or labor market conditions. Since the place of residence is likely exogenous to these conditions, this allocation policy provides a unique setting for studying the effects of immigration on crime. 2 Furthermore, the exogenous allocation arguably led to homogeneous inflows of ethnic German immigrants across regions, allowing us to investigate the importance of local labor market conditions for the crime impact. There are several reasons why immigration may increase crime rates. First, there might be a direct effect on crime. In their economic theory of crime, Becker (1968) and Ehrlich (1973) argue that propensity for crime decreases with legitimate earning opportunities, the probability of being convicted, and the cost of conviction. This implies that immigrants should have higher crime propensities than natives if they face worse labor market conditions than natives. 3 if natives change their behavior due to immigration. Second, immigration may affect crime rates indirectly For example, studies show that immigrants may affect the employment or wages of natives, especially for groups with similar skill levels (e.g., Card, 1990; Kugler and Yuksel, 2008; Glitz, 2012), which in turn may affect the criminal activity of natives. Borjas et al. (2010) present evidence that black males in the United States became more criminal when faced with stronger labor market competition due to immigrant inflows. Identifying the causal effects of immigration on crime is challenging. Controlling for observable (time-varying) crime determinants and region and year fixed effects is 1 Exceptions are studies by Butcher and Piehl (1998, 2007) for the United States, Bianchi et al. (2012) for Italy, and Bell et al. (2013) for the United Kingdom. 2 Glitz (2012) exploits this allocation policy to estimate the impact of immigrants on skill-specific employment rates and wages of natives, finding a displacement effect but no effect on wages. 3 As is true in most industrialized countries, immigrants to Germany perform worse than natives in the German labor market (see Table 1). 1

3 not sufficient because the inflow of immigrants across regions might still be correlated with unobserved time-varying determinants of crime. Most importantly, changing labor market conditions might not be completely captured by unemployment rates or income measures. For example, improving labor market conditions might attract immigrants and reduce crime, which would bias OLS estimates downward. On the other hand, economic downturns might lead to increasing crime rates as well as attract immigrants due to lower housing costs, which would bias OLS estimates upward. Finally, reverse causality could be an issue if crime levels affect the location choice of immigrants. Previous studies on the effects of immigration mainly use two approaches to overcome these issues. One strand of the literature combines two factors to instrument for immigrant inflows across regions: supply-push factors events in origin countries that increase the propensity to emigrate combined with the preexisting distribution of immigrants by origin country. The latter factor is a good predictor of new location decisions since new immigrants tend to settle in areas where immigrants from their home country have settled (Card, 2001; Kerr and Lincoln, 2010; Peri, 2012). Other studies exploit dispersal policies that allocated arriving immigrants (typically refugees) across regions in the host country (Edin et al., 2003; Damm, 2009; Glitz, 2012). The two studies most similar to ours also use these identification strategies. Bianchi et al. (2012) exploit the increase in the immigrant population in Italy in the 1990s that was mainly driven by the collapse of the Soviet Union and the Balkan Wars. In particular, they use immigration flows toward the rest of Europe that are arguably exogenous to demand-pull factors in Italian regions. Variation across the Italian provinces results from preexisting regional differences in the distribution of immigrants by origin country. Bell et al. (2013) study the impact on crime of two immigration waves to the United Kingdom. They use the supply-push instrument for migrant laborers from eight European countries that joined the European Union in Furthermore, the authors exploit a dispersal policy applied to asylum seekers and instrument the number of asylum seekers in a region with the number of asylum seekers in assigned dispersal accommodations. Our identification strategy is similar to the dispersal policy applied to asylum seekers in the United Kingdom. The particular group of immigrants we focus on in this study is ethnic German immigrants. This group lived in Eastern Europe and the former Soviet Union before migrating to Germany after the fall of the Berlin Wall. Between 1988 and 2005, more than 3 million ethnic Germans immigrated, increasing the population in Germany by about 5%. Ethnic German immigrants were granted German citizenship upon arrival, in contrast to other immigrants to Germany who typically can receive German citizenship only after eight years of permanent residence in Germany without having committed a severe crime and having sufficient German-language proficiency. We combine annual county-specific inflows of ethnic German immigrants with annual county-specific crime rates. Furthermore, we add county-specific information on demographic composition and 2

4 labor market conditions to study whether local socioeconomic conditions play a role in the impact of immigration on crime. We focus on West Germany (excluding Berlin) from 1996 until 2005, the period during which newly arriving ethnic German immigrants were allocated across regions by German authorities. Importantly, this particular group of immigrants has characteristics with respect to education level and labor market attachment that are similar to the general immigrant population in Germany (see Table 1). Therefore, the economic model of crime predicts similar crime levels for ethnic German immigrants and the general immigrant population in Germany. However, crime levels might differ between these immigrant groups because ethnic German immigrants receive German citizenship upon arrival (and other immigrants do not), which protects them from being expelled from the country in the event they are convicted of a crime. Since this lowers the cost of conviction for ethnic German immigrants, theory predicts higher crime levels for this immigrant group. We find that ethnic German immigrants increase crime rates considerably. An immigrant inflow of one ethnic German per 1,000 inhabitants increases total crime by 0.88%, which amounts to an elasticity of approximately Effects vary by type of crime, with immigration having a strong impact on burglary, property damage, and battery, but no effect on street crime (a category that contains several street-related types of crime). More importantly, we find that the socioeconomic conditions of the region matter: in regions with good labor market conditions (as measured by the unemployment rate), immigration does not appear to affect crime. On the other hand, crime rates increase considerably in the presence of immigration to regions that suffer from high unemployment. The crime-inhibiting effect of good labor market conditions is consistent both with the economic theory of crime and with empirical studies finding that improvements in local labor market conditions reduce crime rates (Gould et al., 2002; Öster and Agell, 2007; Fougère et al., 2009). Furthermore, we find a similar pattern with respect to preexisting crime levels: the impact of immigration on crime is strongest in regions with preexisting high crime levels which might be due to social interactions (Zenou, 2003). 4 Our study differs from existing work in several dimensions. This is the first study to investigate whether local labor market conditions matter for the impact of immigration on crime. Previous studies investigate either the impact of local labor market conditions on crime in general (without focusing on immigrants) or the effect of immigrants on crime without considering the potentially important role of local labor market conditions. Another difference is that the immigrants we study receive citizenship of the host county upon arrival, while this is not the case for the immigrant groups studied in previous papers. 4 We also find suggestive evidence that the crime impact is larger in cities and in more densely populated counties. This pattern is consistent with lower probabilities of arrest and recognition in urban areas (Glaeser and Sacerdote, 1999). 3

5 Since citizenship reduces the cost of conviction by eliminating the possibility of being expelled from the host country, this could possibly explain why the crime effects discovered in this study are much larger than those found by Bianchi et al. (2012) and Bell et al. (2013). Another way our study differs from previous work is that selection into regions was severely limited by the allocation policy in Germany, whereas the immigrant groups studied by Bianchi et al. (2012) and Bell et al. (2013) were free to choose their region of residence (with the exception of refugees in the United Kingdom). This policy arguably led to rather homogeneous inflows of ethnic German immigrants across regions, enabling us to investigate the importance of local labor market conditions for crime impact. Finally, we estimate short-run effects of immigration on crime by exploiting annual immigration inflows, whereas previous studies estimate medium- to long-run effects based on changes in immigrant stocks across regions. If immigrants are more likely to commit crimes in the first years after arrival, we are more likely to find stronger crime effects. The remainder of the paper proceeds as follows. Section 2 describes the ethnic German immigrants and compares their socioeconomic characteristics to those of the general immigrant population in Germany; furthermore, the allocation policy is described. Section 3 presents the datasets and Section 4 the empirical model. Section 5 reports the determinants of ethnic German immigrant inflows across counties and the main results on the effect of immigration on total crime and by type of crime. Finally, the role of local socioeconomic conditions is investigated, especially the importance of local labor market conditions. Section 6 concludes. 2 Ethnic German Immigrants and the Allocation Policy Ethnic German immigrants are the descendants of German colonists who had migrated to Russia and other East European countries in the 18th and 19th century where they mostly founded closed settlements (Bade, 1990). After the Nazi attacks during World War II in June 1941, many ethnic Germans considered potential collaborators with the Nazi regime were forced to leave their settlements and to move eastward to Siberia, Kazakhstan, and several regions in Middle Asia where they were kept in special settlements. During the Cold War, only very few ethnic Germans were able to emigrate due to political restrictions (see Figure 1). Toward the end of communist rule in the Soviet Union, emigration restrictions were alleviated, leading to a massive emigration wave driven by bad economic conditions and unstable political conditions (Bade and Oltmer, 2003). Since the end of the 1980s, more than 3 million ethnic German immigrants have arrived in Germany. There was a heavy inflow of ethnic German immigrants in the late 1980s and early 1990s, with a peak of 397,000 individuals in Due to these 4

6 large immigration waves, a yearly quota of about 225,000 individuals was introduced in 1993, which was reduced to about 100,000 individuals per year in Due to these quotas, and because the stock of German descendents was becoming smaller, the number of ethnic German immigrants decreased over time (Figure 1). 5 The breakdown of the Soviet Union also led to a significant change in the source countries of ethnic German immigrants. Between 1950 and 1987, only about 7% of the ethnic German immigrants came from countries of the former Soviet Union. In contrast, since the mid-1990s, almost all ethnic German immigrants have come from former Soviet Union countries, especially Kazakhstan, Kirgisistan, Usbekistan, Ukraine, and the Russian Federation (see Figure 1). 6 The immigration of ethnic Germans to Germany was administered centrally. ethnic German immigrants had to apply for a visa at the German embassy in their country of residence which was conducted by the Federal Office of Administration (Bundesverwaltungsamt). To be recognized as an ethnic German, an applicant had to fulfill all requirements of the Federal Refugees Act. Legally, ethnic Germans are members of German minorities who meet the following three requirements (Peters, 2003): (1) must be descendants of at least one parent or grandparent with German nationality; (2) must have declared confession to German ethnicity before leaving the settlement area (typically by possessing the German nationality, sending their children to a German school, or through being a member in a German association); and (3) being able to conduct a simple conversation in German. Due to assimilation pressures in their settlement areas, ethnic German immigrants had rather low language proficiency, however. All Therefore, language requirements were kept at a relatively low level. To maintain the unity of the family, ethnic Germans were allowed to migrate with their spouses and all their offsprings, given that the marriage had existed for at least three years. All ethnic Germans and their relatives migrating with them were granted German nationality immediately upon arrival. All arriving ethnic German immigrants had to pass through a central admission center. If ethnic German immigrants did not have a job or other source of income, which was the case for the vast majority, they were then allocated to one of the 16 German states according to predefined quotas (Königsteiner Schlüssel). These quotas were marginally adjusted every year and were based on the tax revenues (weight of two thirds) and the population size (one third) of each state. Based on the Place of Residence Allocation Act (Wohnortzuweisungsgesetz), established in 1989 due to the large inflows, most states subsequently allocated ethnic German immigrants across counties. The goal of this allocation was a socially acceptable 5 Ethnic German immigrants do not include East Germans who moved to West Germany after unification. 6 The major reason for this development is that since 1993 individuals from countries other than the former Soviet Union were only recognized as ethnic Germans if they could individually prove that they had been discriminated permanently in their country of origin because of their German ethnicity. In contrast, it was assumed that individuals from countries of the former Soviet Union were still being discriminated because of their German ethnicity. 5

7 integration of these immigrants. Because ethnic German immigrants preferred to live close to their relatives and other ethnic German immigrants from their country of origin, and because non-compliance with the assignment did not have any consequences, many left their assigned county and moved to another county. This led to a strong inflow of ethnic German immigrants into some regions (Swiaczny and Mammey, 2001). Because municipalities were responsible for the housing of ethnic German immigrants and the payment of social assistance, this implied great financial burdens for municipalities with large ethnic German immigrant inflows. Furthermore, high concentrations of ethnic German immigrants increased the necessity of integration measures and lowered the acceptance among native Germans (Haug and Sauer, 2006). Seven counties 7 which were particularly hit by massive inflows of ethnic German immigrants signed the so-called Gifhorn Declaration for the Integration of Ethnic German immigrants in March 1995, demanding a more solidary distribution of the burdens among the federation, the states, and the municipalities (Niedersächsische Landeszentrale für Politische Bildung, 2002, p. 10). 8 This declaration caused a crucial amendment to the Place of Residence Allocation Act that mandated that ethnic German immigrants were bound to their assigned county during the first three years after immigration. Especially, it introduced severe sanctions: Ethnic German immigrants would lose all social benefits and any type of public assistance in case of non-compliance with the allocation decision. Only individuals who could prove both a job and housing in their preferred county were exempt from this rule. However, survey evidence (presented below) indicates that only a minority of newly arriving ethnic German immigrants were able to freely choose their county of residence. Most West German states implemented the new law in March 1996, whereas Hesse adopted the law in January 2002, and two states (Bavaria and Rhineland-Palatine) did not adopt it at all. 9 In a later trial, the Federal Constitutional Court decided that the restriction of freedom of movement was not against the constitution. The allocation of ethnic German immigrants within states across counties was not handled uniformly by the states. Typically, the number of ethnic German immigrants allocated to counties was based on state-specific quotas that mainly depended on the size of the resident population. Because quotas were not only based on population size, the ethnic German inflow rates the number of allocated ethnic German immigrants divided by the population in the county varied across counties. The allocation of ethnic German immigrants to counties also accounted for intake capacities or for the county area in some 7 Wolfsburg, Salzgitter, Gifhorn, Nienburg/Weser, Cloppenburg, Emsland, and Osnabrück. 8 Dropping these counties from the analysis does not affect the results. 9 The amendment to the Place of Residence Allocation Act was indeed effective. For example, the counties of the Gifhorn Declaration which experienced a massive inflow until 1995 received only few ethnic German immigrants after the law was adapted in March 1996 (Wenzel, 1999, Figures 1 and 2). Glitz (2012) furthermore notes that the perception at the Ministry of the Interior and the Association of German Cities and Towns was that the sanctions were effective and ensured high compliance with the initial allocation decision. 6

8 states (e.g. in North Rhine-Westphalia). And some states did not adjust their quotas to changes in the counties population size. 10 Furthermore, state quotas for allocating ethnic German immigrants were also based on economic strength, such that richer states were assigned more ethnic German immigrants. The county of residence of newly arriving ethnic German immigrants was largely determined by family ties. The Ministry of the Interior estimates that the presence of family members was the decisive factor for the allocation decision in about 90% of all cases. The presence of health and care facilities and the infrastructure for single parents were additional factors. Most importantly, the skill level of the ethnic German immigrants did not play any role in the allocation process (Glitz, 2012, p. 180). Even though official data are lacking 11, there is strong evidence that a vast majority of ethnic German immigrants complied with their assigned county of residence during the entire period of retention (three years after immigrating). First, ethnic German immigrants were much more likely to depend on social welfare than native Germans. Therefore, leaving the assigned county would be quite costly. In a survey among 1,554 ethnic German immigrants, only 11% reported that they did not receive any type of social benefit or state assistance during the first three years after arrival (Haug and Sauer, 2007, p. 101). And among those who immigrated after 1996, only 36% worked at some point during the first three years after arrival (Haug and Sauer, 2007, p. 120). Second, ethnic German immigrants were likely to comply with the assignment because they could not afford their own housing. After arriving in Germany, about 80% of ethnic German immigrants lived in temporary residential homes or admission centers (Seifert, 1996). Furthermore, they stayed on average about two years in these temporary residential homes (Mammey, 1999). Adolescent ethnic German immigrants are also very likely (about 51%) to live in residential establishments or in social housing, compared to 10% of native Germans (Dietz, 1999, Table 12). Again, this further indicates a strong reliance on housing provided by the state. 12 The socioeconomic characteristics of ethnic German immigrants who arrived in the late 1990s and early 2000s (the immigrants in our analysis) are worse than those of native Germans. This can be explained by the fact that while ethnic German immigrants who emigrated early (in the late 1980s and early 1990s) were predominantly individuals of own German descent according to the Federal Act on Expellees (Bundesvertriebenengesetz), 10 Controlling for county and year state fixed effects, we find that the county-specific ethnic German immigrant inflow rates are uncorrelated with both the share of foreigners and with the inflow rates in the previous three years. 11 Because ethnic German immigrants cannot be identified in the registration offices of the municipalities due to their German citizenship, the number of ethnic German immigrants residing in a county is unknown. Therefore, we do not directly know how many ethnic German immigrants were able to choose their place of residence because they had both job and housing, nor how many of them changed their assigned county within the three-year period of retention. 12 Furthermore, it is likely that ethnic German immigrants predominantly search for a job in the vicinity of their place of residence (Glitz, 2012). 7

9 but after 1995 the spouses and offsprings of ethnic German descendants constituted the majority. For example, the share of ethnic German immigrants of own German descent was only 47.7% in 1996 and declined further to 21.5% in 2005 (Haug and Sauer, 2006, p. 417). The proficiency in the German language of ethnic German immigrants is surprisingly low. One reason is certainly that ethnic Germans had to assimilate in their countries of origin, thereby losing German language skills. Even among ethnic German immigrants who immigrated in the early 1990s when the majority was of own German descent self-assessed German language proficiency at the end of the 1990s is not better than the language proficiency of other immigrants who immigrated in the same years, and is even worse than the language proficiency of second-generation immigrants in Germany (Haug, 2005). German language proficiency is presumably even worse for the ethnic German immigrants who arrived in the late 1990s and early 2000s because the majority were non-german family members. The language spoken at home clearly points in this direction: The majority of adolescent ethnic German immigrants spoke either Russian (46%) or Russian and German (46%) at home (Dietz, 1999, p. 159). Only a minority of adolescents spoke only German at home (8%). The prevalence of Russian as main language at home is even more severe in binational families: The majority of adolescent ethnic German immigrants (67%) assess their German language proficiency as being mediocre of bad (Dietz, 1999, p. 159). And only 13% of adolescent ethnic German immigrants who migrated to Germany between 1990 and 1994 report that they had spoken German in their country of origin (Dietz, 1999, p. 161). 13 Furthermore, low German language proficiency might also arise because the social network of adolescent ethnic German immigrants mainly consists of other ethnic German immigrants: 54% of adolescents reported that their friends are predominantly other ethnic German immigrants, and 23% report that their friends are other ethnic German immigrants and native Germans. Only 4.3% of ethnic German immigrants reported that their friends are mainly native Germans (Dietz, 1999). Apart from language deficiencies, young ethnic German immigrants also belong to the lower end of the education distribution. Administrative school data from the largest German state, North Rhine-Westphalia, for the school year 1996/1997 show that ethnic German children are highly overrepresented in the least academic secondary school track (Hauptschule) and highly underrepresented in the most academic track (Gymnasium). Strikingly, this over-/underrepresentation is even more extreme than that of foreigners in Germany (Dietz, 1999, Table 9). 13 Ethnic German immigrants were supported more than other immigrants by the German government. They received allowances (Eingliederungshilfe) and were offered German language courses. However, due to increasing numbers of ethnic German immigrants and the worsening macroeconomic conditions in Germany, integration allowances such as language courses and vocational qualification measures were cut to a maximum duration of only six months. 8

10 The occupation composition of newly arriving ethnic German immigrants also indicates a rather low-educated immigrant group. When registering in Germany, ethnic German immigrants reported the last occupation in their home country upon arrival. It turns out that ethnic German immigrants were highly overrepresented in lower skill occupations such as farmers, laborers, transport workers, operatives, and craft workers (see Glitz 2012, p. 188f.). Based on the Micro Census 2008, an administrative dataset that is representative of the resident population in Germany, we find that ethnic German immigrants have much worse socioeconomic outcomes than native Germans, but similar outcomes than other immigrants in Germany. 14 Table 1 compares native Germans with ethnic German immigrants and other immigrants who arrived in Germany between 1996 and Because the number of ethnic German immigrants declined over time, ethnic German immigrants have been living in Germany about one year longer than other immigrants. According to the assimilation literature (e.g. Lubotsky, 2007), this should tend to improve their labor market outcomes. Both ethnic German immigrants and other immigrants are heavily overrepresented in the two lowest education categories (ISCED 1 and 2; see Panel B). While other immigrants have similar levels of post-secondary and tertiary education (ISCED 4-6) than native German, ethnic German immigrants are heavily underrepresented in the highest education groups. While ethnic German immigrants have somewhat better employment outcomes than other immigrants, both groups have substantially worse outcomes than native Germans, in particular much higher unemployment rates and higher incidences of temporary employment (Panel C). Concerning individual net income, ethnic German immigrants and other immigrants have lower incomes than native Germans (Panel D). Finally, ethnic German immigrants and other immigrants are much more likely to receive social assistance (Hartz IV) and unemployment benefits (ALG I) than natives. In sum, although ethnic German immigrants are Germans from a legal point of view, this group has much poorer socioeconomic characteristics and employment outcomes than native Germans. Interestingly, employment outcomes of ethnic German immigrants and other immigrants are quite similar. 3 Data We have collected administrative data on the number of reported crimes at the county level from the State Offices of Criminal Investigation (Landeskriminalämter) for the years and from the Federal Criminal Police Office (Bundeskriminalamt) for the years 14 We use the Micro Census from 2008 since this is the first large micro dataset that allows to identify ethnic German immigrants unambiguously. All samples are restricted to individuals residing in those West German states we use in our analysis. 9

11 We were able to collect data on total crime and on some categories of crime: battery, drug offense, property damage, burglary, and street crime. Total crime contains any type of crime that is reported to the police. Battery includes, among others, bodily harm with fatal consequences, grievous and serious bodily harm, and bodily injury caused by negligence. Drug offense contains all crimes that are associated with general violations of the narcotics law, drugs trafficking, and illegal importation of drugs. Property damage covers all crimes that destroy, damage or make useless another person s property. Burglary contains all crimes that include housebreaking for committing theft. Street crime is a mixed category that includes various street-related crimes, for example, theft of, in, and out of cars and robberies. 16 Table 2 reports the West German states we use in our analysis along with the number of observation. We include all West German states that allocated ethnic German immigrants to counties. 17 We focus on ethnic German immigrant inflows in the period because the allocation to counties was not binding before 1996 and because inflows were rather small after Note that the time period in Table 2 ( ) refers to the years when crimes were reported (that is one year after immigration takes place; see the discussion of the empirical model in Section 4). 18 number of counties in our sample by crime category. 19 The remaining columns report the Summary statistics at the county level are reported in Table 3. Panel A reports crime rates by the six crime categories. Crime rates are defined as the number of crimes reported to the police per 100,000 inhabitants (see Equation 1). The average total crime rate in our sample is 6,883 crimes per 100,000 inhabitants. Compared to the mean, the standard deviation of 2,661 crimes is rather large, which is particularly due to large differences between counties. After taking out permanent differences between counties and changes 15 German counties correspond to the NUTS-3 level in the Eurostat classification. They are comparable in size to U.S. counties. 16 Crime categories can be reviewed in the List of Crimes (Straftatenkatalog) from the Federal Criminal Police Office. Street Crime includes the following crime categories: rape and gang rape in hit-and-run style; exhibitionistic and indecent behavior; robberies out of cash and valuable transports; predaceous assaults of car drivers; robberies by not paying the bill; purse snatching; other robberies on streets, ways or places; grievous and serious bodily harm on streets, ways or places; kidnapping or hostage-taking together with robberies out of cash and valuable transports; burglary in/out of kiosks; burglaries in/out of shop windows, showcases, or vitrines; burglaries in/out of cars, burglaries of cars, mopeds or motorcycles; pickpocketing; burglaries in/out of vending machines; violations of the public peace; property damages on cars; other property damages on streets, ways or places. 17 We exclude Bavaria and Rhineland-Palatinate because these states did not allocate ethnic German immigrants to counties. We focus on West Germany (excluding East Germany and Berlin) because crime data at the county level are not available for East German states for the years and because data on ethnic German immigrant inflows are very fragmentary for East Germany. 18 Hamburg starts only in 1999 due to missing information on ethnic German immigrant inflows and Hesse starts only in 2003 because the allocation of ethnic German immigrants to counties started only in Observation numbers vary by crime category due to missing data. Furthermore, crime data for North Rhine-Westphalia were only provided at the level of police administration areas. While in most cases police administration areas are identical with the borders of counties, we had to aggregate 14 counties (8.3%) into 6 police administration areas. 10

12 over years within states by regressing the outcomes on county and year state fixed effects, the variation in crime rates naturally decreases, but still amounts to about 10% of the mean value (Adj. SD). Crime rate t = reported crimes t 100, 000 population t 1 (1) The relative importance of each crime category can be expressed by the number of crimes in the respective category as a percentage of total crime. The largest crime category is street crime covering 27.1% of all reported crimes. The second biggest category is property damage with 11.0%, followed by battery (7.2%), drug offense (4.2%), and burglary (1.9%). reported crimes. 20 Taken together, these five crime categories account for 51.4% of all Annual data on the inflows of ethnic German immigrants into counties is obtained from two different sources. Inflow data from 1996 to 2001 is taken from Glitz (2012) which come from the admission centers in each state. Inflow data from 2002 to 2006 come from the Bundesarbeitsgemeinschaft Ev. Jugendsozialarbeit e.v., Jugendmigrationsdienste which is a church initiative that takes care of underprivileged youths. This initiative collected the data directly from the state Ministries of the Interior. We checked the comparability of the two different data sources by comparing data from the overlapping years , finding a very high degree of consistency. 21 Summary statistics of the ethnic German immigrant inflow at the county level is reported in Panel B. The average yearly inflow between 1996 and 2005 was 293 ethnic German immigrants per county, with a minimum of 0 and a maximum of 2,393 ethnic German immigrants. There is substantial variation across counties, with a standard deviation of 270 ethnic German immigrants. Netting out time-consistent county differences and yearly within-state changes, the adjusted standard deviation (Adj. SD) is 93 ethnic German immigrants. The main explanatory variable of interest is the county-specific ethnic German immigrant inflow rate. The inflow rate is defined as the number of ethnic German immigrants assigned to a county in year t divided by the population of that county at the end of year t 1 (see Equation 2). The average inflow rate was about one ethnic German immigrant per 1,000 inhabitants in each year. Over the entire sample period, the average inflow rate was 12 ethnic German immigrants per 1,000 inhabitants, i.e. slightly more than one percent of the population. This inflow rate implies that the average county received 3,151 ethnic German immigrants between 1996 and The total inflow rate 20 Using the nationwide crimes rates for Germany between yields very similar results. 21 Furthermore, at the national level, information on the immigrants countries of origin, last occupation in the country of origin, and last labor force participation status is available. At the state level, the absolute numbers of immigrants allocated, the age structure, and religious affiliation is available. 11

13 varies between 2 ethnic German immigrants per 1,000 inhabitants and 22 ethnic German immigrants per 1,000 inhabitants. Ethnic German inflow rate t = Number of allocated ethnic German immigrants t P opulation t 1 (2) Summary statistics of additional county-specific control variables are reported in Panels C and D. County characteristics (Panel C) include demographic information such as population, the share of foreigners, and the share of young male adults aged 15 to 39. Furthermore, we use the unemployment rate and GDP per capita to capture the economic conditions in a county. 22 Finally we use the clear-up rates by crime category (Panel D) as a proxy for the probability of being arrested. The clear-up rates are constructed by multiplying the number of crimes that are cleared up by 100 and dividing by the number of reported crimes, crimes cleared upt 100 reported crimes t. 4 Empirical Model To investigate the effect of immigration on crime, we estimate the following model: ln crime rate ist = α + β ethnic German immigrant inflow rate ist 1 + X ist 1γ + µ i + (µ t µ s ) + ɛ ist (3) where the logarithm of the crime rate of county i (situated in state s) in year t is regressed on the ethnic German immigrant inflow rate of the previous year. 23 We control for county fixed effects (µ i ) to eliminate any time-invariant determinants of crime across counties. We furthermore control for year state fixed effects (µ t µ s ) to partial out annual changes in crime rates that are common across all counties in the same state (see discussion below). X ist 1 incudes variables on county demographics and economic conditions. It is a well-known problem in the crime literature that reported crimes underestimate the true (but unobserved) number of committed crimes (e.g., Bianchi et al., 2012). This may lead to biased estimates if the extent of underreporting is correlated with the determinants of crime. Typically, previous studies have therefore used the logarithms of crime rates and included region and year fixed effects (see Levitt, 1996; Gould et al., 2002; Öster and Agell, 2007; Fougère et al., 2009; Bianchi et al., 2012). This method eliminates measurement errors that are constant within regions over time and measurement errors that are constant within time periods across regions. This arguably 22 Population data and GDP per capita come from the Federal Statistical Office (Regionalstatistik) and unemployment rates from the Federal Employment Agency (Bundesagentur für Arbeit). 23 As a robustness check, we have also used the absolute ethnic German inflow as the main explanatory variable. These results (not reported) are very similar. 12

14 eliminates several important sources of underreporting, such as cultural differences across regions or changing crime-awareness over time (e.g. transported via national media). If newly arriving ethnic German immigrants were free to choose their place of residence, β might be biased. It might be biased downward, for example, if immigrants would choose to move to counties because the crime rate was falling. In contrast, β would be biased upward if immigrants would settle in counties with increasing crime rates, for example, because of low house prices. We address this problem by using the exogenous allocation of newly arriving ethnic German immigrants to counties where compliance was ensured by severe sanctions. Our preferred specification includes year state fixed effects instead of using only year fixed effects. This model is more restrictive since it nets out crime trends that are common to all counties within a state, and not only common trends across all counties. There are three reasons for including year state fixed effects. First, each state decides how to allocate ethnic German immigrants across its counties. Therefore, changes in the allocation policy would affect all counties within the same state, but would not affect counties in other states. Second, states, and not the federal level, have the authority over police issues. Each state decides, for example, about hiring new police officers and how to allocate police officers across counties. Third, states differ in the demographic composition of their population and in the proximity to neighboring countries which might lead to different crime developments over time. Figure 2 depicts the crime trends for six states in our sample (excluding the city states Hamburg and Bremen which are outliers with high levels of crime), showing that crime rates indeed develop differently over time. These differing crime trends are taken into account by including year state fixed effects. However, even after controlling for county fixed effects, crime rates might be correlated within counties over time if deviations from the long-run county-specific mean are correlated over time. Similarly, standard errors across state-year cells might be correlated across counties. For this reason, we cluster standard errors at the level of the included fixed effects, assuming the following error structure: ɛ ist = θ i + (θ t θ s ) + η ist. We thus allow the error term ɛ ist to be correlated within counties over time (θ i ) and within state-year cells across counties (θ t θ s ). We use two-way clustered standard errors to account for both dimensions at the same time. 24 We also control for several county characteristics, X ist 1, that might influence crime rates. 25 Demographic characteristics include the log of population size 26, the share of 24 See Cameron et al. (2011) for a theoretical derivation of the two-way clustering method and Acemoglu and Pischke (2003) for an application of two-way clustering. We use the program cgmreg provided by Douglas L. Miller for computing two-way clustered standard errors. 25 For a review of the literature on the determinants of crime, see, e.g., Freeman (1999). 26 Since the model includes county fixed effects, population size implicitly also captures population density, which is one determinant of criminal activity (Glaeser and Sacerdote, 1999). 13

15 foreigners, and the share of young male adults (aged 15-39) since this population group is more likely to commit crimes (Freeman, 1999). We also control for log GDP per capita and unemployment rate to capture labor market opportunities (see, e.g., Gould et al. 2002). Finally, we include the clear-up rate for the respective crime category defined as the ratio of crimes cleared up by the police divided by the total number of reported crimes as a proxy for the expected costs of crime (see Ehrlich, 1996). Note that the county characteristics are also lagged by one year since the inflow of ethnic German immigrants might have affected these characteristics, thus rendering the county characteristics in year t bad controls (Angrist and Pischke, 2009). 27 Given that the allocation of ethnic German immigrants was binding for three years, we are able to identify the short-run impact of immigration on crime. Because we exploit annual inflow data and data on the stock of ethnic German immigrants are not available we cannot identify medium- to long-run effects which are identified by Bianchi et al. (2012) and Bell et al. (2013). The empirical model is set up to reflect this short-run perspective: the crime rate in year t is related to the immigration inflow in the previous year (t 1). Using the one-year lag of the inflow ensures that we do not erroneously attribute crimes committed in year t to ethnic German immigrants who arrived at the end of year t. However, this erroneous attribution might happen if we used crime and inflow data from the same year instead. In particular, the model assumes that newly arriving immigrants commit crimes between arrival and two years after arrival. This model therefore implies that we estimate short-run effects of immigration on crime. If immigrants have poor labor market outcomes particularly in the years after arrival, for example due to occupational downgrading (Dustmann et al., 2013), crime impacts might be stronger in the short run than in the medium to long run. The identification of a causal effect on crime depends on the effectiveness of the allocation procedure and in particular on the exogeneity of the allocation with respect to transitory regional crime (and labor market) conditions. Glitz (2012, p.193) points out that the exogeneity requirement is likely satisfied because the main allocation criterion was the proximity of family members and because labor market market skills did not play a role in in the allocation process. He provides evidence that the allocation to states was exogenous with respect to individual characteristics as suggested by the overwhelming importance of family ties for the allocation decision. He compares the age distributions of ethnic German immigrants across the West German states and finds that these distributions are very similar (with standard deviations much smaller than the corresponding standard deviations of the overall population), which should be the case if immigrants were exogenously allocated with respect to individual characteristics. Glitz argues that these results also indicate that the allocation within states to different counties was also exogenous since the allocation to different counties followed similar 27 The results do not change when we use contemporaneous controls. 14

16 administrative processes and decision criteria as the previous allocation to states. In sum, because skills and age two potentially important determinants of crime were no factors for allocating ethnic German immigrants to particular counties, the skill and age distribution of newly arriving ethnic German immigrants should be identical across counties. 28 A potential threat to our identification strategy is that family members or other relatives could have self-selected into particular low crime counties. However, if the selection is due to permanent characteristics of the county, we would control for this by including county fixed effects. And if relatives selected into counties based on temporary features of the county (such as good labor market conditions), this would only lead to biased estimates if the historic local temporary shock is correlated with the current local temporary shock. But there are often many years or even decades between the immigration of the ethnic Germans and their ancestors, rendering a correlation of temporary shocks unlikely. In sum, the allocation procedure and the descriptive statistics across states provided by Glitz (2012) indicate that the allocation of ethnic German immigrants is exogenous to their individual characteristics. Ethnic German immigrants were only allowed to choose their county of residence if they could prove both regular employment and permanent residence. However, this was unlikely for ethnic German immigrants who arrived with low proficiency in German language and strong dependence on social assistance (see Section 2). The exogenous allocation thus ensures a homogeneous inflow of immigrants across regions that allows identifying the impact of immigrants on crime and to investigate how the local socioeconomic (especially labor market) conditions shape this impact. 5 Results In this section, we first investigate the determinants of the ethnic German immigrant inflow rates into counties (Section 5.1). Then we present the main results for the effect of immigrants on total crime and on the various crime categories (Section 5.2). Finally, we investigate whether the crime impact of immigrants depends on the demographic composition or labor-market conditions of the region (Section 5.3). 5.1 The Determinants of the Immigrant Inflows across Regions We first investigate the determinants of the number of ethnic German immigrants allocated to different counties. Counties with larger populations and counties with higher GDP per capita receive more ethnic German immigrants (Columns 1 and 2 in Table 4). 28 Glitz (2012, footnote 12) notes that this assumption may not hold if the skills between immigrants and their relatives are correlated. However, because these families were typically separated a long time ago and attended different education systems, the correlation in skills is likely small. 15