Ethnic Residential Segregation and Immigrants Perceptions of Discrimination in West Germany

Transcription

1 Ethnc Resdental Segregaton and Immgrants Perceptons of Dscrmnaton n West Germany Verena Dll Uwe Jrjahn Research Papers n Economcs No. 10/11

2 Ethnc Resdental Segregaton and Immgrants Perceptons of Dscrmnaton n West Germany Verena Dll and Uwe Jrjahn Department of Economcs, Unversty of Trer, Germany Abstract: Usng survey data from the German Soco-Economc Panel, ths study shows that mmgrants lvng n segregated resdental areas are more lkely to report dscrmnaton because of ther ethnc background. Ths apples to both segregated areas where most neghbors are mmgrants from the same country of orgn as the surveyed person and segregated areas where most neghbors are mmgrants from other countres of orgn. The results suggest that housng dscrmnaton rather than self-selecton plays an mportant role n mmgrant resdental segregaton. JEL: J15, J61, R23, R30. Keywords: Segregaton, Immgrants, Housng Dscrmnaton, Self-Selecton. Correspondng Author: Prof. Dr. Uwe Jrjahn, Unverstät Trer, Fachberech IV, Lehrstuhl für Arbetsmarktökonomk, Unverstätsrng 15, Trer, Germany, Emal: jrjahn@un-trer.de.

3 1. Introducton Immgrant resdental segregaton has been a long-standng concern n many developed countres. Ths also holds true for Germany where concerns about the lack of mmgrant ntegraton and fears of parallel socetes play an mportant role. In the polcy debate, t s wdely assumed that mmgrants prefer to lve n segregated resdental areas (Münch 2009). Ths vew accords wth theores suggestng that mmgrants sort themselves nto ethnc enclaves as those enclaves enable the consumpton of ethnc goods (Chswck and Mller 2005) and reduce the need to assmlate to the host country (Lazear 1999). Yet, t s an open queston whether or not self-selecton s ndeed the drvng force of mmgrant resdental segregaton. A contrastng vew s that housng dscrmnaton plays an mportant role (Galster 1992, Ynger 1998). Ths vew emphaszes that mmgrants lve n segregated neghborhoods not because they prefer to lve there but because natves restrct mmgrant locaton choces to specfc areas. Ths vew s supported by feld studes. 1 However, whle feld studes provde evdence of dscrmnaton n the housng market, they usually do not provde evdence of a relatonshp between housng dscrmnaton and resdental segregaton. 2 Usng survey data from the German Soco-Economc Panel (SOEP), our study examnes the assocaton between resdental segregaton and mmgrants perceptons of dscrmnaton. 3 Ths allows nvestgatng whether self-selecton or housng dscrmnaton plays the crucal role n mmgrant resdental segregaton. If mmgrants voluntarly sort themselves nto ethnc enclaves, there should be no assocaton between segregaton and perceved dscrmnaton. Yet, f housng dscrmnaton plays the crucal role n resdental segregaton, there should be a sgnfcant assocaton. 1

4 Moreover, we can examne the potental channels of dscrmnaton n more detal. On the one hand, mmgrants may experence ncreased dscrmnaton and outrght hostlty from prejudced natve neghbors f they lve n resdental areas where natves are most present. As a consequence, they are more or less forced to settle n areas wth other mmgrants to escape everyday dscrmnaton n ther neghborhood. Hence, mmgrants lvng n segregated areas should be less lkely to report dscrmnaton than mmgrants lvng n neghborhoods wth a hgh share of natves. 4 On the other hand, dscrmnaton by landlords rather than outrght hostlty by natve neghbors may drve resdental segregaton. Natves may, to a greater or lesser extent, tolerate the mmgrants lvng n ther neghborhood. In ths case, mmgrants may prefer to lve n natve-domnated areas as these areas allow them to partcpate n the socal and cultural lfe of the host country. Landlords may nonetheless exclude mmgrants from natve-domnated resdental areas f the ntroducton of mmgrants reduces the wllngness of natve tenants to pay hgh rents. 5 Moreover, landlords may restrct mmgraton locaton choces f they are themselves prejudced or ther experence ndcates that mmgrants are on average tenants wth unstable rent payments and less dlgence n mantanng the dwellng n approprate condton. In ths stuaton, mmgrants who are forced to lve n segregated areas should be more lkely to report dscrmnaton than those who are able to avod such areas. Our estmates for West Germany show that lvng n a hghly segregated area s postvely assocated wth perceved dscrmnaton. Ths suggests that dscrmnatory treatment by landlords s an mportant factor drvng ethnc resdental segregaton. Importantly, the postve lnk between segregaton and perceved dscrmnaton apples 2

5 to both segregated areas where most neghbors are mmgrants from the same country of orgn as the surveyed person and segregated areas where most neghbors are mmgrants from other countres of orgn. Partcularly, the postve lnk between the frst type of segregated area and perceved dscrmnaton corroborates the nterpretaton that housng dscrmnaton rather than self-selecton plays an mportant role n mmgrant resdental segregaton. If self-selecton would be the drvng force behnd mmgrant segregaton, areas wth neghbors from the same country of orgn should be specfcally attractve as mmgrants can share the same culture and language. Yet, even mmgrants lvng n these areas are more lkely to report dscrmnaton. Our results hold true even when controllng for other nfluences such as household ncome, rent payment, and qualty of the dwellng. The rest of the paper s organzed as follows. Secton 2 descrbes the data. Secton 3 dscusses the varables. Secton 4 presents the results. Secton 5 concludes. 2. Data Our emprcal analyss uses data from the SOEP (Wagner et al. 1993). The SOEP s a large representatve longtudnal survey of prvate households n Germany. The survey s admnstered by the German Economc Insttute (DIW Berln). Infratest Sozalforschung, a professonal survey and opnon nsttute, conducts the ntervews. Based on face-toface ntervews, a nucleus of soco-economc and demographc questons s asked annually. Dfferent specal topcs are sampled n specfc waves. The frst wave of ntervews started n 1984 wth the collecton of data n the former West Germany. Whle the survey has been extended to East Germany after German reunfcaton, the number of foregners n the East German subsample s too small to allow a separate analyss. Hence, our examnaton s restrcted to West Germany. 3

6 Immgrants are oversampled n the SOEP. The ntal cohort of mmgrants ncluded persons from the former guest worker countres Italy, Greece, Span, Turkey, and Yugoslava. Durng the latter half of the 1950s the German government started actvely recrutng guest workers n response to a labor shortage prompted by very hgh economc growth rates. In 1973 the government stopped the recrutment of further guest workers as Germany entered a perod of economc recesson. In the subsequent years, the nflow of mmgrants from the former guest worker countres conssted manly of famly members of those guest workers who remaned n Germany (famly reunfcaton). We focus on frst generaton mmgrants from Italy, Greece, Span and Turkey. Immgrants from the former Yugoslava are excluded from the analyss because of Yugoslava s dverse ethnc and relgous groups. The 1996 wave of the SOEP ncludes nformaton on the mmgrants perceptons of dscrmnaton. Informaton on the ethnc composton of the neghborhood s avalable n the 1994 wave. Hence, we regress perceved dscrmnaton n 1996 on factors observed n Immgrants who changed ther place of resdence n the perod 1994 to 1996 are excluded from the analyss. 3. Varables Table 1 provdes defntons of the varables and descrptve statstcs. In our basc regressons, the dependent varable s a dummy equal to 1 f the mmgrant answers that he or she s very often dscrmnated aganst because of hs or her ethnc background. The dummy s equal to 0 f the mmgrant answers that he or she s never or only sometmes dscrmnated aganst. 13 percent of the ntervewees answer that they are very often dscrmnated aganst. As the dependent varable s a dchotomous varable, we use the 4

7 probt procedure to estmate the determnants of perceved dscrmnaton. An advantage of the probt model s that the results are relatvely easy to nterpret. However, as a check of robustness we also estmate a generalzed ordered probt wth a categoral varable that comprses three outcomes: 0 never dscrmnated aganst, 1 sometmes dscrmnated aganst, 2 very often dscrmnated aganst. The key explanatory varables are constructed usng two peces of ordered nformaton. Frst, ntervewees are asked f foregners lve n ther neghborhood. Second, those who lve n resdental areas wth foregn neghbors are asked f they share the same country of orgn wth ther foregn neghbors. Combnng the two peces of nformaton yelds a classfcaton of fve dfferent types of resdental areas. The frst type s a resdental area where most or all of the neghbors are foregners and most or all of them are from the same country of orgn as the respondent. 8 percent of respondents lve n such a neghborhood. The second type s a resdental area where most or all of the neghbors are foregners and most or all of them are from other countres of orgn as the respondent. 40 percent of respondents lve n ths type of neghborhood. Consderng the two types of resdental areas together, 48 percent of mmgrants n our sample lve n a hghly segregated neghborhood. The thrd and the fourth type are resdental areas wth some foregn neghbors. In the thrd type of resdental area, most or all of the foregn neghbors are from the same country of orgn as the respondent. In the fourth type of resdental area, most or all of the foregn neghbors are from other countres of orgn as the respondent. The ffth type s the reference category. In ths type of resdental area, all of the neghbors are Germans. As emphaszed, f self-selecton plays the domnant role n mmgrant resdental 5

8 segregaton, there should be no sgnfcant lnk between lvng n a segregated resdental area and perceved dscrmnaton. It s the mmgrant s choce to lve n such area. By contrast, f dscrmnaton s the drvng force of segregaton, we should observe a sgnfcant assocaton between lvng n a segregated area and perceved dscrmnaton. The sgn of the assocaton depends on the type of dscrmnaton. On the one hand, outrght hostlty by natve neghbors n natve domnated areas may force mmgrants to settle n areas wth other mmgrants to escape everyday dscrmnaton. In ths case, mmgrants lvng n segregated areas should be less lkely to report dscrmnaton than those lvng n natve domnated areas. On the other hand, dscrmnaton by landlords rather than hostlty by natve neghbors may play an mportant role n resdental segregaton. If natves to a greater or lesser extent tolerate foregn neghbors, mmgrants may prefer to lve n natve-domnated areas as ths allows them to ntegrate nto the host country. Yet, landlords may restrct mmgrant locaton choces f they are themselves prejudced or suspect that the ntroducton of mmgrants results n lower rental ncome. In that case, mmgrants who are forced to lve n a segregated resdental area should be more lkely to report dscrmnaton than those who are able to fnd housng n a natvedomnated area. 6 The survey provdes a rch set of control varables. In our ntal specfcaton, we control for federal states and demographc characterstcs of the ntervewee. The mmgrant s educaton may play a role n perceved dscrmnaton. On the one hand, negatve atttudes toward foregners are more prevalent among low-educated Germans (Gang and Rvera-Batz 1994, Cornelssen and Jrjahn 2011). To the extent hghereducated mmgrants are less lkely to nteract wth low-educated Germans, they face a 6

9 lower rsk of dscrmnaton. On the other hand, hgher-educated mmgrants are lkely to have hgher expectatons and requrements. Hence, they may tend to be more senstve wth respect to dscrmnaton. Smlarly, expectatons may rse wth the years the mmgrant lves n Germany. The mmgrant s ncome s also lkely to play a role. Immgrants wth a hgher ncome have more fnancal opportuntes to avod dscrmnaton. Furthermore, we account for gender, presence of chldren and country of orgn. We expand the specfcaton by ncludng varables for neghborhood characterstcs. The expanded specfcaton accounts for urban areas, contacts to neghbors, satsfacton wth envronmental condtons n the neghborhood, and satsfacton wth the avalablty of goods and servces n the neghborhood. Ths allows examnng whether or not the ethnc composton of the neghborhood s just a proxy for other neghborhood characterstcs (Swaroop and Krysan 2011). In a fnal step, we nclude varables for housng characterstcs. Immgrants may be concentrated n resdental areas wth poor qualty housng. As housng satsfacton can nfluence perceved dscrmnaton, t s mportant to control for the characterstcs of the dwellng n order to check whether or not a possble lnk between segregaton and perceved dscrmnaton s drven by the qualty of the dwellng. We account for sze of dwellng, year of constructon, and rent pad for the dwellng. 4. Emprcal Analyss 4.1 Basc Results Table 2 provdes the probt estmates of the determnants of hgh perceved dscrmnaton. In regresson (1), several of the varables for demographc characterstcs 7

10 emerge wth statstcally sgnfcant coeffcents. Income s a negatve covarate of the probablty that an mmgrant reports beng hghly dscrmnated aganst. Educaton, years snce mmgraton and a Turksh orgn are postve covarates. Most mportantly n our context, the regresson shows that mmgrants lvng n hghly segregated areas are more lkely to report hgh dscrmnaton. Ths conforms to the hypothess that dscrmnaton by landlords plays an mportant role n mmgrant resdental segregaton. The assocaton between resdental segregaton and hgh perceved dscrmnaton apples to both hghly segregated areas where most of the foregn neghbors are from the same country of orgn as the mmgrant and hghly segregated areas where most of the foregn neghbors are from other countres of orgn. Partcularly the assocaton between the frst type of segregated area and perceved dscrmnaton corroborates the nterpretaton that housng dscrmnaton rather than selfselecton s the drvng force behnd mmgrant resdental segregaton. If self-selecton would be the drvng force, areas wth neghbors from the same country of orgn should be most attractve as mmgrants can share the same culture and language. Immgrants who sort themselves nto these areas would not report dscrmnaton. Yet, even mmgrants lvng n these areas have an ncreased probablty of reportng hgh dscrmnaton. In column (2), we expand the specfcaton by ncludng varables for other neghborhood characterstcs. Three of the four neghborhood varables take statstcally sgnfcant coeffcents. Satsfacton wth envronmental condtons, lvng n an urban area, and contacts wth neghbors are negatve covarates of perceved dscrmnaton. Nonetheless, even when ncludng varables for other neghborhood characterstcs, we 8

11 stll fnd that both types of hghly segregated areas are postvely assocated wth perceved dscrmnaton. Ths suggests that the postve lnk between lvng n a segregated area and perceved dscrmnaton s not smply due to other neghborhood characterstcs such as envronmental condtons. The concentraton of foregners tself appears to drve the lnk. In column (3), we addtonally nclude varables for the qualty of dwellng. A small sze of dwellng and lvng n a recently constructed home are postvely assocated wth perceved dscrmnaton. Most mportantly, even when accountng for the qualty of dwellng, lvng n a segregated area s a postve covarate of perceved dscrmnaton. Ths suggests that the lnk between segregaton and perceved dscrmnaton s not smply drven by poor qualty of dwellngs n segregated areas. The coeffcents on the segregaton varables are not only statstcally but also quanttatvely sgnfcant. The probablty of hgh perceved dscrmnaton ncreases by 10 percentage ponts f an mmgrant lves n a segregated area where most neghbors are from the same country of orgn. Compared to the mean of 13 percent, ths mples an ncrease by 77 percent. The probablty of hgh perceved dscrmnaton ncreases by 8 percentage ponts f the mmgrant lves n a segregated area where most neghbors are from other countres of orgn. Compared to the mean, ths s an ncrease by 62 percent. 4.2 Robustness Check So far we used a dummy dependent varable for hgh perceved dscrmnaton. The reference group comprsed both mmgrants wth no and mmgrants wth moderate perceved dscrmnaton. As a check of robustness, we now consder an ordered varable whch dfferentates between three outcomes. Let y denote the extent of mmgrant s 9

12 perceved dscrmnaton (0 = never feels dscrmnated aganst; 1 = sometmes feels dscrmnated aganst; 2 = very often feels dscrmnated aganst). An ordered probt model would be the standard procedure to estmate the determnants of the extent of perceved dscrmnaton. In ths model, y depends on a latent varable y *: y * β' x, (1) where x s the vector of explanatory varables, and β the vector of coeffcents. The error term has a normal dstrbuton wth zero mean and a varance equal to one. Takng the latent varable nto account, the extent of perceved dscrmnaton s 0 f y* 1, y 1f 1 y* 2, (2) 2 f y* 2. The threshold values 1 and 2 are estmated jontly wth β. However, a shortcomng of the smple ordered probt approach s that t has constant threshold values and only a sngle coeffcent vector for all outcome categores of the dependent varable. A generalzed ordered probt model stands as an alternatve to the rather restrctve ordered probt model (Boes and Wnkelmann 2010). It provdes more flexblty as t does not treat the thresholds as constant but makes them dependent on the explanatory varables. As t allows for dfferent coeffcent vectors, the generalzed model takes nto account that the effects of the explanatory varables may vary wth the categores of the dependent varable. The model makes the thresholds lnear functons of the explanatory varables: where ~ j s a constant term, ~ γ ' x ( j 1, 2), (3) j j j x agan the vector of establshment characterstcs, and γ j 10

13 a vector of coeffcents. Substtutng j for dstrbuton (.) nto account, we obtan the probabltes: j n (2) and takng the standard normal Pr( y 0) ( ~ 1 β 1' x ), (4a) Pr( y 1) ( ~ ) ( ~ 2 β2' x 1 β 1' x ), (4b) Pr( y 2) 1 ( ~ 2 β2' x ), (4c) where β j β γ j as we cannot dentfy β and j γ separately. The coeffcent vectors are now allowed to vary across outcomes and we have a vector β j for each threshold. Table 3 provdes the estmaton results of the generalzed ordered probt model. All control varables lsted n column (3) of Table 2 are ncluded n the estmaton. In order to save space, we only report the coeffcents on our key explanatory varables. Equaton (4c) helps nterpret the results. It shows that the probablty of hgh perceved dscrmnaton s only nfluenced by vector β 2. Aganst ths background, t can be seen that the generalzed ordered probt model confrms our basc fndng. Both types of hghly segregated areas are assocated wth an ncreased probablty of hgh perceved dscrmnaton. 5. Conclusons In the polcy debate, t s often beleved that a lack of mmgrant assmlaton s due to self-selecton of mmgrants nto segregated resdental areas. The results of ths study suggest that housng dscrmnaton rather than self-selecton plays an mportant role n mmgrant resdental segregaton. Immgrants lvng n hghly segregated areas are much more lkely to report hgh dscrmnaton than those lvng n non-segregated areas. The postve lnk between segregaton and perceved dscrmnaton also ndcates 11

14 a specfc mechansm of dscrmnaton. It suggests that dscrmnatory restrctons of mmgrant locaton choces rather than outrght hostlty of natve neghbors are the drvng force of resdental segregaton. In the latter case we would have found that mmgrants lvng n segregated areas are less lkely to report dscrmnaton as those areas provde protecton aganst everyday dscrmnaton by natve neghbors. Yet, our estmates suggest the opposte relatonshp. Of course, our fndng does not mean that mmgrants lvng n natve-domnated areas experence no dscrmnaton at all n ther neghborhood. It rather means that from the mmgrants vewpont the advantages of lvng n a natve-domnated neghborhood outwegh the dsadvantage of dscrmnatory treatment by prejudced natve neghbors. As a consequence, mmgrants perceve the restrctons that force them to lve n segregated areas as dscrmnaton. We note that our analyss apples to West Germany. It would be nterestng to extend the analyss to East Germany as xenophobc tendences appear to be partcularly hgh n East Germany (Krueger and Pschke 1997). Ths requres that suffcent nformaton wll be avalable n future waves of the SOEP. 12

15 Table 1: Varable Defntons and Descrptve Statstcs (N = 771) Varable Descrpton (Mean, Std. dev.) Hgh perceved dscrmnaton Dummy = 1 f the person feels very often dscrmnated aganst because of hs or her ethnc background (.1258,.3319). Modest perceved dscrmnaton Dummy = 1 f the person feels sometmes dscrmnated aganst because of hs or her ethnc background (.4643,.4990). Hgh share of foregners & Dummy = 1 f most or all neghbors are foregners and most or same country of orgn all of them are from the same country of orgn as the mmgrant Hgh share of foregners & other countres of orgn Some foregners & same country of orgn Some foregners & other countres of orgn (.0778,.2681). Dummy = 1 f most or all neghbors are foregners and most or all of them are from other countres of orgn as the mmgrant (.4008,.4904). Dummy = 1 f some neghbors are foregners and most or all of them are from the same country of orgn as the mmgrant (.0558,.2300). Dummy = 1 f some neghbors are foregners and most or all of them are from other countres of orgn as the mmgrant (.3554,.4789). Year of constructon Dummy = 1 f the property was constructed between 1949 and 1971 (.4100,.4921). Year of constructon Dummy = 1 f the property was constructed between 1972 and 1980 (.1076,.3101). Year of constructon Dummy = 1 f the property was constructed between 1981 and 1990 (.0337,.1806). Educaton Years of schoolng rangng from 7 to 18 years (9.074, 1.829). Greece Dummy = 1 f the mmgrant s from Greece (.1764,.3814). Italy Dummy = 1 f the mmgrant s from Italy (.2180,.4131). Turkey Dummy = 1 f the mmgrant s from Turkey (.5253,.5000). Female Dummy = 1 f the mmgrant s a woman (.4578,.4985). Equvalence ncome/100 Real equvalence net ncome of the household n Euro. The Household ncome s dvded by the weghted sum of ndvduals lvng n the household (7.39, 2.77). Sze of dwellng Dwellng area n square meter dvded by the number of people lvng n household (22.70, 11.67). Rent/100 Rent pad for the dwellng n Euro (3.24, 1.56). Urban area Dummy = 1 f the mmgrant lves n an urban area (.1414,.3486). Avalablty of goods and Satsfacton wth the avalablty of goods and servces n the servces n neghborhood neghborhood coded from 0 lowest to 10 hghest (7.224, 2.057). Envronmental condtons Satsfacton wth the envronmental condtons n the neghborhood coded from 0 lowest to 10 hghest (6.545, 1.967). 13

16 Contacts wth neghbors Ordered varable for the mmgrant s contacts wth neghbors coded from 1 almost no contact to 5 very close (2.966,.8506). Chldren Dummy = 1 f chldren lve n the household (.5460,.4982). Years snce mmgraton Years snce the person mmgrated to Germany (20.92, 7.384). Federal state dummes Dummy varables to account for the eleven federal states n West Germany. 14

17 Table 2: Determnants of Hgh Perceved Dscrmnaton; Probt Estmates (1) (2) (3) Hgh share of foregners & same country of orgn.5985 [.0891] (1.99)**.6543 [.0952] (2.09)**.8035 [.1009] (2.56)** Hgh share of foregners & other countres of orgn.5199 [.0730] (2.06)**.5304 [.0702] (2.05)**.6759 [.0763] (2.57)** Some foregners & same country of orgn.4007 [.0513] (1.14).3965 [.0471] (1.07).5128 [.0501] (1.39) Some foregners & other countres of orgn.2740 [.0316] (1.04).2718 [.0291] (0.99).4243 [.0383] (1.56) Female [-.0243] [-.0210] [-.0206] (1.09) (0.99) (1.04) Educaton.0577 [.0094].0531 [.0082].0583 [.0083] (1.70)* (1.54) (1.65)* Years snce mmgraton.0180 [.0029].0170 [.0026].0194 [.0028] (1.71)* (1.61) (1.84)* Equvalence ncome/ [-.0149] [-.0147] [-.0143] (3.22)*** (3.26)*** (3.28)*** Turkey.6581 [.0765].6526 [.0717].6610 [.0574] (1.90)* (1.89)* (1.78)* Italy.2672 [.0218].2377 [.0179].3361 [.0211] (0.74) (0.65) (0.86) Greece.4341 [.0415].4534 [.0417].5349 [.0411] (1.21) (1.27) (1.39) Chldren.0796 [.0130].0705 [.0109] [-.0097] (0.59) (0.52) (0.41) Urban area [-.0497] [-.0357] (1.72)* (1.35) Avalablty of goods and servces [-.0024] [-.0037] (0.50) (0.84) Envronmental condtons [-.0130] [-.0121] (2.31)** (2.25)** Contacts wth neghbors [-.0223] [-.0216] (2.02)** (1.97)** Sze of dwellng [-.0020] (1.83)* Rent/ [-.0025] (0.32) Year of constructon [-.0215] (1.07) Year of constructon [-.0047] (0.13) Year of constructon [.2981] (3.33)*** 15

18 Constant (3.52)*** (1.79)* (1.41) Federal state dummes Yes Yes Yes Observatons Pseudo R The table shows the estmated coeffcents. Robust z-statstcs are n parentheses and margnal effects are n square brackets. Margnal effects of dummy varables are evaluated for a dscrete change from 0 to 1. Margnal effects of the dummes for neghborhood segregaton (dummes for country of orgn) are changes n probablty compared to the reference group of persons lvng n a neghborhood wthout foregners (reference group of people from Span). Margnal effects of varables other than the dummy varables are evaluated at the mean values. *** Statstcally sgnfcant at the 1% level; ** at the 5% level; * at the 10% level. 16

19 Table 3: Determnants of the Extent of Perceved Dscrmnaton; Generalzed Ordered Probt Estmaton 1 2 Hgh share of foregners & same country of orgn (1.86)*.7970 (2.59)*** Hgh share of foregners & other countres of orgn.1958 (1.15).6166 (2.29)** Some foregners & same country of orgn (0.36).5257 (1.41) Some foregners & other countres of orgn.2313 (1.37).2597 (0.93) Log Lkelhood Observatons 771 The table shows the estmated coeffcents. Robust z-statstcs are n parentheses. *** Statstcally sgnfcant at the 1% level; ** at the 5% level; * at the 10% level. The regresson ncludes all control varables lsted n column (3) of Table 2. 17

20 References Ahmed, Al M., Lna Andersson and Mats Hammarstedt Can Dscrmnaton n the Housng Market Be Reduced by Increasng the Informaton about Applcants? Land Economcs 86: Ahmed, Al M. and Mats Hammarstedt Dscrmnaton n the Rental Housng Market: A Feld Experment on the Internet, Journal of Urban Economcs 64: Boes, Stefan and Raner Wnkelmann Income and Happness: New Results from Generalzed Threshold and Sequental Models, Socal Indcators Research 95: Bosch, Marano, M. Angeles Carnero and Lda Farre Informaton and Dscrmnaton n the Rental Housng Market: Evdence from a Feld Experment, Regonal Scence & Urban Economcs 40: Bosch, Marano, M. Angeles Carnero and Lda Farre Rental Housng Dscrmnaton and the Persstence of Ethnc Enclaves, IZA Dscusson Paper No. 5583, Bonn. Card, Davd, Alexandre Mas and Jesse Rothsten Tppng and the Dynamcs of Segregaton, Quarterly Journal of Economcs 123: Carpusor, Adran G. And Wllam E. Loges Rental Dscrmnaton and Ethncty n Names, Journal of Appled Socal Psychology 36: Chswck, Barry R. and Paul W. Mller Do Enclaves Matter n Immgrant Adjustment? Cty and Communty 4: Cornelssen, Thomas and Uwe Jrjahn September 11 th and the Earnngs of Muslms n Germany The Moderatng Role of Educaton and Frm Sze, Journal of Economc Behavor and Organzaton, Forthcomng. Daley, Amy B., Stanslav V. Kasl, Theodore R. Holford, Tene T. Lews and Beth A. Jones Neghborhood- and Indvdual-level Socoeconomc Varaton n Perceptons of Racal Dscrmnaton, Ethncty and Health 15: Dll, Verena, Uwe Jrjahn and Georg Tsertsvadze Resdental Segregaton and Immgrants Satsfacton wth the Neghborhood n Germany, Research Papers n Economcs 8/11, Unversty of Trer. Galster, George C Research on Dscrmnaton n Housng and Mortage Markets: Assessment and Future Drectons, Housng Polcy Debate 3: Gang, Ira N. and Francsco L. Rvera-Batz Unemployment and Atttudes 18

21 Towards Foregners n Germany, n Economc Consequences of Immgraton to Germany, edted by Gunter Stenmann and Ralf Ulrch, Physca Verlag: Hunt, Matthew O., Laren A. Wse, Mare-Claude Jpguep, Yvette C. Cozer and Lynn Rosenberg Neghborhood Racal Composton and Perceptons of Racal Dscrmnaton: Evdence from the Black Women s Health Study, Socal Psychology Quarterly 70: Krueger, Alan B. and Pschke, Joern-Steffen A Statstcal Analyss of Crme Aganst Foregners n Unfed Germany, Journal of Human Resources 32, pp Münch, Syblle It s All n the Mx : Constructng Ethnc Segregaton as a Socal Problem n Germany, Journal of Housng and the Bult Envronment 24: Ondrch, Jan, Alex Strcker and John Ynger Do Landlords Dscrmnate? The Incdence of Racal Dscrmnaton n Rental Housng Markets, Journal of Housng Economcs 8: Page, M Racal and Ethnc Dscrmnaton n Urban Housng Markets: Evdence from a Recent Audt Study, Journal of Urban Economcs 38: Saz, Albert and Susan Wachter Immgraton and the Neghborhood, Amercan Economc Journal: Economc Polcy 3: Swaroop, Sapna and Mara Krysan The Determnants of Neghborhood Satsfacton: Racal Proxy Revsted, Demography 48: Wagner, G., R. Burkhauser and F. Behrnger The Englsh Publc Use Fle of the German Soco-Economc Panel, Journal of Human Resources 28: Ynger, John Housng Dscrmnaton Is Stll Worth Worryng About, Housng Polcy Debate 9: Ynger, John Measurng Racal Dscrmnaton wth Far Housng Audts: Caught n the Act, Amercan Economc Revew 76:

22 Endnotes 1 Specfcally n the US, a seres of audt studes has been conducted (Ondrch et al. 1999, Page 1995, Ynger 1999). Testers from two dfferent ethnc groups are matched and traned so that they make equvalent enqures when speakng to prospectve landlords. Moreover, there are recent feld studes from several countres whch use wrtten applcatons (Ahmed and Hammarstedt 2008 and Ahmed et al for Sweden, Bosch et al. 2010, 2011 for Span, Carpusor and Loges for the U.S.). Fcttous persons wth dstnctve soundng ethnc names apply for vacant rental apartments va the Internet. The results of the varous feld studes pont n the same drecton: Members of ethnc mnorty groups are shown and offered fewer housng unts. 2 An excepton s the feld study by Bosch et al. (2011). They fnd that the dscrmnaton of applcants wth foregn soundng names s more severe n areas wth a low presence of mmgrants. 3 Dll et al. (2011) examne the relatonshp between resdental segregaton and mmgrants neghborhood satsfacton n Germany. Ths study complements ther examnaton by usng a more drect measure of dscrmnaton. 4 Studes by Hunt et al. (2007) and Daley et al. (2010) suggest that ths holds true for Afrcan Amercans n the US. 5 Prejudced natves may try avodng areas wth a hgh share of mmgrant neghbors nstead of showng outrght hostlty. Saz and Wachter (2011) show that the growth of a neghborhood s mmgrant share s assocated wth slower housng value apprecaton. Card et al. (2008) provde evdence of a flght of whtes once the mnorty share exceeds a crtcal level. 20

23 6 Note that fndng housng s usually also a stochastc process. Ths holds true for both housng dscrmnaton and self-segregaton. In case of housng dscrmnaton, an mmgrant s subject to dscrmnaton wth some exogenous probablty (less than 1). The mmgrant can avod housng dscrmnaton wth the complementary probablty. In case of self-segregaton, an mmgrant has an exogenous probablty (less than 1) of fndng vacant housng n a preferred segregated neghborhood. He or she fals to fnd vacant housng n such a neghborhood wth the complementary probablty. 21