Available via license: CC BY-NC-SA 4.0
Content may be subject to copyright.
|
8
The Second Generation in the German Labor Market
Frank Kalter
In recent years a number of large-scale studies have addressed the integration
of former labor migrants’ children into the German labor market (Granato
2004; Granato and Kalter 2001; Kalter 2005; Kalter and Granato 2002, 2007;
Kalter, Granato, and Kristen 2007; Konietzka and Seibert 2003; Seibert and
Solga 2005). Despite the use of very dierent indicators of labor success, the
ndings are rather consistent and lead to a series of stable common insights.
First, although doing noticeably better than the rst generation, the second
generation is still clearly disadvantaged compared to native-born Germans.
is holds true at least for Greeks, Italians, (ex-)Yugoslavs, and Turks, while
only second-generation Spaniards seem to have caught up with their Ger-
man peers in many respects. Second, studies also agree that the second-
generation disadvantage in the labor market is mainly due to schooling and
vocational training. Upon controlling for formal qualications, dierences
to the reference population decrease considerably and are no longer signi-
cant for most of the groups in most of the analyses. In other words, so-called
ethnic penalties, a term suggested by Heath and Ridge (1983) for disadvan-
tages that are not mediated by educational attainment, seem to play only a
minor role in the labor-market integration of the second generation. ird,
immigrant youth of Turkish heritage play an exceptional role within this pat-
tern. In all analyses they face considerable and, as a rule, highly signicant
ethnic penalties. Even if, occasionally, ethnic penalties can also be observed
for other groups, these are always, and always by far, overshadowed by the
respective gures for the Turks.
In this chapter I want to continue this line of research by more deeply
examining the last of these points, the exceptional role of the Turks. More
precisely, I ask how their specic disadvantage, even aer controlling for edu-
cation, might be explained. Is there a particular discrimination against Turk-
e Second Generation in the German Labor Market |
ish youth in Germany, as many authors tend to assume (Seibert and Solga
2005), or does the particular Turkish penalty result from other processes?
Although a number of rival hypotheses have been suggested to account for
the ethnic penalties of immigrant youth in the labor market (Kalter et al.
2007; Heath et al. 2008), most data sets do not contain measures that would
allow for direct empirical tests. is also holds true for the German Micro-
census, which is the data source in all the studies cited earlier. erefore, I
rely here on an alternative data set, the German Socio-Economic Panel Study
(GSOEP). e GSOEP contains some helpful indicators for important theo-
retical concepts, above all country-specic resources, which, besides discrim-
ination, are seen as major potential causes of ethnic penalties. In addition,
being a panel study, the GSOEP allows tracking of the early career paths of
second-generation immigrants in a longitudinal design. is enables stricter
tests of the assumed causal relationships between labor-market success and
other factors than would be possible in a mere cross-sectional perspective.
In the next section I start with a brief review of potential mechanisms
accounting for ethnic penalties in the labor market and discuss whether their
underlying assumptions are met in the case of second-generation Turks in
Germany. Aerward, I sketch the data structure and relevant variables of my
analyses. I then present the major results: aer replicating the three general
ndings noted earlier, I show that a lack of host-country-specic capital, most
notably language prociency, and the ethnic composition of network struc-
tures are critical to explaining the exceptional Turkish case. e latter nding
holds also when using longitudinal techniques. Finally, in the concluding sec-
tion I discuss the major implications of the results for the situation of Turks in
Germany and for migration research in general.
Explaining Ethnic Penalties of Turkish Youth
in the German Labor Market
To account for variations in ethnic penalties, meaning residual eects of eth-
nicity net of education, at least four main classes of argument have been pro-
posed (Kalter et al. 2007; Heath et al. 2008). To begin with, ethnic penalties
are seen to result from dierential treatment, above all from direct and indi-
rect forms of discrimination and social exclusion. In the German case there
might be a specic discrimination against Turkish youth in the labor market.
Supercially, this hypothesis seems quite plausible, given the fact that many
studies have shown that negative stereotypes and social distance on the part
of Germans are more pronounced toward Turks than toward any other group
|
of labor migrants (Ganter 2003; Steinbach 2004). Although it is tempting to
directly connect these results with the exceptional pattern of second-genera-
tion Turks in the labor market, one must bear in mind that there is no auto-
matic relationship between attitudes and behavior—least of all for actors in
the labor market. Further, it has been questioned whether the structural con-
ditions fostering discrimination are especially pronounced in the case of the
German labor market (Kalter and Granato 2007). For example, comparative
research has shown that in Germany the link between educational and voca-
tional qualications and the labor market is especially close (Müller, Stein-
man, and Ell 1998). is means, in other terms, that the signaling power of
educational qualications is relatively strong, leaving less room for the occur-
rence of processes of statistical discrimination based on ascribed characteris-
tics. us, before drawing an overhasty conclusion from the aforementioned
ndings that specic discrimination exists, it seems worth asking what other
factors could be responsible for the pronounced Turkish penalties.
A second obvious explanation would be that ethnic penalties arise from
skills and abilities that are relevant for an employee’s productivity but are not
captured by formal qualications. In other words, unmeasured aspects of
human capital may account for ethnic dierences, above all aspects that are
culturally specic, such as language prociency or other cultural knowledge.
Basically, this argument is the most obvious explanation for why residual
eects of ethnicity (controlling for education) may be observed for the rst
generation (Chiswick 1978, 1991; Friedberg 2000). However, although immi-
grants’ children will probably do much better than their parents with respect
to such culturally specic skills, a considerable gap between them and indig-
enous youth might still exist. And this would be especially reasonable in the
case of the Turks: although all labor migrants had to bridge some type of
cultural gap on arrival—for example, no knowledge of German, since it is
not spoken in any of the six former recruitment countries—there is no doubt
that cultural distance is greatest for Turks. us, factors related to cultural
distance would also be plausible explanations for their exceptional role.
A third possible explanation rests on the notion that one’s own human
capital is not the only resource relevant to achieving a good labor-market
position. Most notably, there might be a direct impact of parental resources
on children’s success, which is not mediated by children’s educational attain-
ment. Parents, for example, might invest money in their children’s search for
an adequate position or, because of their own socioeconomic position, have
better access to job opportunities. Given that the rst generation of immi-
grants, for whatever reasons, occupies lower labor-market positions, this
e Second Generation in the German Labor Market |
might lead to an ethnic penalty for their children too. Again, this kind of
reasoning would apply especially to Turks: among rst-generation groups,
Turks have the lowest educational and occupational attainment (Granato
2004; Granato and Kalter 2001; Kalter and Granato 2007), and that would
reasonably account for the second generation experiencing a specic penalty
in the labor market.
Finally, besides using parental resources, young job seekers might also
draw on the resources of other persons; that is, they might use their social
capital. It is well known in the economic literature that social networks play
an important part in the labor market, as many jobs are found with the help
of friends and relatives (Granovetter 1995). e theoretical reason for this lies
in the fact that for the job seeker, network information on job oers is inex-
pensive and promises a comparatively high probability of success, while for
the rms, referrals by third persons might be important as a comparatively
valid and likewise inexpensive screening device (Montgomery 1991, 1408).
erefore, relevant characteristics of a person’s network—its size, density,
and, most notably, the resources connected to network ties—may make a
dierence for status attainment beyond a person’s human capital (Lin 1999;
Portes 1995b, 9).
But why might the characteristics of social networks result in ethnic dis-
advantages for second-generation immigrants in general and Turks spe-
cically? It is reasonable to assume that the networks of second-generation
youth still tend to consist predominantly of coethnic ties. Coethnic ties, how-
ever, give access only to the information and resources available within the
ethnic community and—given that there is ethnic stratication—might not
be as helpful as ties to the indigenous population (Portes and Rumbaut 2001,
48). For example, in a recent U.S.-based study, missing network informa-
tion turned out to completely explain race disadvantages in hiring processes
(Petersen, Saporta, and Seidel 2000). With respect to the German situation,
once again, this line of reasoning would indeed be able to account for a spe-
cic disadvantage of Turks: they are the largest immigrant group by far and,
as a consequence, have more opportunities to build ethnic ties and to engage
in ethnic communities. In fact, recent studies nd that Turkish youth have
considerably more ethnically homogeneous networks than do other second-
generation groups (Haug 2003, 724). erefore, a comparatively low level of
social assimilation could be a fourth mechanism behind the exceptional role
of second-generation Turks in the labor market.
Note, however, that this typical view of “straight-line assimilation” has
been challenged, and the reasoning could also be turned upside down. As
|
has been argued, for example, in the concept of segmented assimilation,
under specic structural conditions, reliance on ethnic ties and avoidance
of social assimilation may even promise a relative advantage (Portes 1995a,
251; Portes and Rumbaut 2001, 44). Most obviously, if discrimination against
a certain group is severe, the ethnic community may provide job opportuni-
ties that are not accessible within the host country’s labor market. In addi-
tion, the ethnic community may oer relative economic advantages in the
form of self-employment in niches that the mainstream economy does not
include (Portes 1995b, 25). A further position—one could call it a pluralist
or transnationalist view—argues that it might be especially advantageous for
a second-generation immigrant to have both: host-country-specic as well
as ethnic ties. And a similar argument could be made with respect to other
culturally specic kinds of capital, for example, language prociency. e so-
called middlemen minorities (Bonacich 1973) would be examples of ethnic
groups who prot from their position in between two cultures.
To summarize, there are several plausible hypotheses to explain the excep-
tional situation of second-generation Turks. Turkish youth are especially dis-
advantaged not only with respect to prejudices of the indigenous population
but also with respect to socioeconomic background, host-country-specic
cultural capital, and host-country-specic social capital. As all these factors
may reasonably account for labor-market disadvantages, an empirical answer
must be sought to the question of which of them turns out to be more or less
important. With respect to the fourth of the sketched mechanisms, it is a
further empirical question whether ethnically homogeneous social networks
work in a negative direction at all and, if so, whether assimilative or mixed
networks then promise relatively more success. Analyzing the precise eects
of the composition of friendship networks not only contributes to solving
the “Turkish puzzle” but also highlights the importance of concepts such as
segmented assimilation, pluralism, or transnationalism for the second gen-
eration in Germany, thereby contributing to recent discussions in migration
research in general.
Data and Variables
e empirical analyses rest on data from the German Socio-Economic
Panel Study (GSOEP), a yearly longitudinal survey of private households in
Germany conducted since 1984 (see Haisken-DeNew and Frick 2004). Up
to 2003, the GSOEP had gathered information on 55,439 persons in total,
oversampling nationals of the former recruitment countries.1 Although the
e Second Generation in the German Labor Market |
GSOEP is a sample of households, individual interviews are conducted with
each household member once he or she reaches the age of sixteen. In a rst
step, I select only those individuals (n = 5,179) who had been interviewed at
the age of seventeen, in order to follow their career paths. e advantage of
this design is that for this specic subgroup there is ample information on
family background, and I select only those persons for whom both parents
have been interviewed at least once in the GSOEP (n = 4,653). I then restrict
my analysis to only three groups: respondents whose parents were both born
in Germany, respondents whose parents were both born in Turkey, and
respondents whose parents were both born in one of the other four recruit-
ment countries: Italy, ex-Yugoslavia, Spain, or Greece. I drop all persons who
were born outside Germany and immigrated at age seven or older. To avoid
confounding the analysis with German reunication and its aermath, I
nally select only those persons who had been living in West Germany at
the age of seventeen. Using these rules, I end up with 2,931 individuals, 2,150
of whom belong to the German group, 342 to the Turkish, and 439 to the
groups of other labor migrants. In total, these individuals reveal informa-
tion on 21,298 person years, 2,499 of them stemming from Turkish youth
and 3,125 from youth with backgrounds from other labor-migrant countries.
e following variables are used in the analyses: e basic dependent
variable is a rough measure of occupational attainment and contrasts salaried
employees (1) with workers (0). It is constructed out of the EGP scheme (Erik-
son, Goldthorpe, and Portocarero 1979).2 e choice of this specic indicator
of labor-market success makes the analyses directly comparable to those of
Granato and Kalter (2001), which are based on the German Microcensus of
1996.3 Besides ethnic group membership—dened as described earlier—in all
models I control for gender (time-constant), age, age squared (time-varying),
and I include dummy variables for the year of the survey. Educational quali-
cations (time-varying) are captured by the CASMIN scheme, consisting of
eight categories (Brauns and Steinmann 1999). Socioeconomic background
is measured by father’s years of education (time-constant), on the one hand,
and father’s occupational status (time-constant) in terms of the ISEI score
(Ganzeboom, de Graaf, and Treiman 1992), on the other. If the father’s ISEI
score is missing, this is indicated by a dummy variable.
In addition to delivering information on socioeconomic background, the
GSOEP data have the crucial advantage of containing indicators of culturally
specic skills and resources. A central variable for my analyses is the per-
centage of best friends who are German. Six times within the twenty waves
of the GSOEP information has been gathered on the (up to) three persons
|
a respondent considers to be his or her best friends. My measure expresses
the fraction of these friends having German citizenship. For missing years I
impute the last information available. If the respective information is missing
at the beginning, I impute the rst information available. is measure of
social assimilation can also be constructed for the reference group of Ger-
mans, but other information is available only for respondents with a migra-
tion background. is holds true for the variable language problems in Ger-
man, which is based on a self-reported evaluation of speaking uency (1 =
very good, 5 = very poor). e relevant question is included in the GSOEP
at least every two years. erefore the variable can be built as a time-varying
one, imputing the previous value for those years where the information is
missing. In a similar manner, a variable for problems concerning the language
of the country of origin can be constructed. e general design allows me
to measure social assimilation and language prociencies not only for the
respondent him- or herself but also for the father and the mother. In these
cases as well, the variables are treated as being time-varying. When the father
or the mother drops out of the panel in a certain year, the last information
available is imputed for all subsequent years.
Results
is section analyzes which of the rival, previously oered explanations for the
penalty suered by second-generation Turks turns out to be empirically more
important. e next section reports summary statistics of relevant variables, in
order to check whether the assumed background conditions of several mecha-
nisms do indeed hold according to the data. e subsequent section moves on
to multivariate analyses, allowing me to answer the leading question system-
atically. To present the major nding in advance: the ethnic composition of
friendship networks seems to play the most important part in explaining the
specic role of Turks. Building on this result, the nal empirical section tests
whether this network eect turns out to be indeed a direct and causal one.
Descriptive Statistics
Table 8.1 gives the percentages or means of all variables used in the analy-
ses. Most importantly, the ndings show that the second generation is at a
considerable disadvantage with respect to access to salaried employee posi-
tions, the basic dependent variable: whereas 62 percent of all young Germans
in the data set have occupied such a position at least once within the time
e Second Generation in the German Labor Market |
span under consideration, this holds true for only 43 percent of Turks and 50
percent of the children of other labor migrants. In addition, the second gen-
eration is also clearly disadvantaged with respect to educational attainment.
e percentage of those who have at least an upper secondary education (i.e.,
a credential earned in one of the two higher tracks of the school system)
is considerably lower among Turks (44.3 percent) and the children of other
labor migrants (46.8 percent) than among Germans (66.6 percent). In the
sample, only minor dierences between the three groups exist with respect
to the composition according to gender and age.4
Although these four variables are also contained in many available studies
mentioned at the beginning of this chapter, some additional variables relate
to possible explanations for the specic ethnic penalty of second-generation
Turks. e ndings on father’s socioeconomic status and years of education
indicate that the second generation is indeed underprivileged with respect
to familial socioeconomic background, but no specic Turkish disadvantage
.
Summary statistics of relevant variables
German (1) Tur k is h (2)
other labor
migrant Tot al n (total)
percentages:
salaried employeea62.2% * 42.5% 50.0% 57.9% 2038
at least secondary educationb66.6% * 44.3% 46.8% 60.6% 2472
female 48.8% 45.6% 48.7% 48.4% 2931
meansc
age 20.2 20.2 20.1 20.2 2931
father’s ISEI 47.0 * 31.0 32.3 42.9 2165
father’s years of education 12.0 * 9.3 9.1 11.3 2806
percentage best friends German .97 * .38 * .51 .80 2007
father: % best friends German .98 * .18 * .27 .73 1846
mother: % best friends German .98 * .15 * .27 .73 1925
language problems German – 1.55 * 1.36 1.44 754
language problems country of origin – 2.09 * 1.97 2.02 754
father: language problems German – 2.86 * 2.70 2.77 733
mother: language problems German – 3.33 * 2.83 3.04 746
() *=dierence between German and Turkish signicant on a -level
() *=dierence between Turkish and ‘other labor migrant’ signicant on a -level
a an individual is treated as ‘yes’ if the person was a salaried employee at least once over all years under
consideration
b CASMIN-classication higher than a,b, or c; highest value for all years is chosen for each individual
c for time-varying variables the mean of intra-individual means is reported
|
was found in the GSOEP data. ere is, however, a considerable and specic
gap for Turks with respect to integration into German networks: on average,
the percentage of German friends is only 38 percent for a second-generation
Turk, whereas it is 51 percent for a descendant of one of the other former
labor migrants. It is interesting to note that a similar dierence can also be
observed with respect to parents’ social integration, the dierence being a bit
more pronounced for mothers (15 percent versus 27 percent) than for fathers
(18 percent versus 27 percent). Turkish youth deviate from the residual sec-
ond generation also in other culturally specic resources. eir speaking u-
ency in German is signicantly worse than that of the children of other labor
migrants, and the same—albeit less pronounced—holds true for their speak-
ing uency in the language of the country of origin. Again, considerable dif-
ferences in German-language prociency already exist for the parents.
Testing Rival Hypotheses on the Exceptional Situation of Turks
To analyze the reasons for the exceptional situation of Turkish youth in the
German labor market, several logit models are run using the dichotomous
salaried-employee-versus-worker variable, described earlier, as the depen-
dent variable. In a rst set of models (table 8.2, models 1–5), data are pooled
for all years, and robust standard errors reecting the clustering of individu-
als (Rogers 1993; Stata Corporation 2001, 256) are estimated to account for
the panel structure of the data. Model 1 shows the gross disadvantages of the
second-generation groups, expressed by the log-odds eects of ethnic origin
controlling only for gender, age, age squared, and year of the survey. Model 2
then also includes education in terms of the full CASMIN scheme.
Although the analysis rests on a dierent data set and a somewhat dif-
ferent denition of ethnic origin, models 1 and 2 basically conrm the three
major results stemming from former Microcensus analyses, reported in the
introduction: there is a distinct gross disadvantage of second-generation
Turks in the German labor market. e respective log-odds eect in model
1 is –0.96, indicating that the relative odds to be in a salaried employee posi-
tion versus a worker position as compared to Germans is exp(–0.96) ≈ 0.38.
e gross disadvantage for the other labor-migrant groups is much less pro-
nounced, but dierences to the reference groups are also highly signicant.
As can be seen in model 2, this disadvantage is considerably reduced and no
longer signicant when educational qualications are controlled. In contrast,
the Turkish disadvantage is only reduced, leaving an odds-ratio of exp(–
0.66) ≈ 0.52, which is still highly signicantly dierent from 1. is illustrates
|
.
Log-odds effects (selected coefficients) on salaried employee position () vs. worker ()
log-odds effects using pooled data
(robust standard errors accounting for clustering on ID)
(1) (2) (3) (4) (5)
groups (ref: German)
- Turkish -.96* -.66* -.54* -.49* -.02
(.19) (.19) (.21) (.21) (.27)
- other labor migrant -.39* -.19 -.07 -.05 .29
(.15) (.16) (.17) (.18) (.22)
education in CASMIN categories (ref: 1a)
- CASMIN 1b -.38 -.29 -.35 -.37
(.19) (.20) (.21) (.21)
- CASMIN 1c -.11 -.04 -.03 -.08
(.22) (.23) (.24) (.24)
- CASMIN 2b .67* .69* .66* .60*
(.20) (.21) (.22) (.22)
- CASMIN 2a .92* .98* .95* .90*
(.21) (.22) (.23) (.22)
- CASMIN 2c_gen 1.62* 1.58* 1.55* 1.50*
(.23) (.24) (.25) (.25)
- CASMIN 2c_voc 2.64* 2.65* 2.64* 2.60*
(.31) (.32) (.33) (.33)
- CASMIN 3a, 3b 3.59* 3.52* 3.52* 3.47*
(.42) (.43) (.44) (.44)
father years of education .02 .03 .02
(.03) (.03) (.03)
father ISEI .01* .01 .01*
(.01) (.01) (.01)
father ISEI missing .41 .43 .47
(.25) (.27) (.27)
% best friends German .79*
(.25)
number of persons 2038 1925 1846 1527 1527
person years 11274 10942 10589 9971 9971
Pseudo-R2.16 .27 .27 .27 .27
data: GSOEP ; *=p<.
in brackets: robust standard errors accounting for clustering on ID
gender, age, age squared and dummy variables for years are also controlled for in all models
|
and, once again, underscores the exceptional position of Turks among the
second generation.
What, now, are the reasons for this ethnic penalty borne by Turks? As
stated earlier, one hypothesis assumes that it could result from a dierent
socioeconomic background that impacts labor-market positioning regard-
less of educational qualication. erefore in model 3 the father’s years of
education and the father’s socioeconomic status are included as additional
independent variables. However, although the latter variable has a signicant
impact on the odds of attaining a skilled position, it can hardly explain the
specic situation of Turks. As compared to model 2, the log-odds eect for
Turks is only slightly reduced and still highly signicantly dierent from zero.
But the latter fact changes once we include a measure of social assimila-
tion in the model. Model 5 shows that the percentage of Germans among the
three best friends signicantly raises odds of attaining a skilled labor-market
position, leading to a complete reduction of the disadvantage for Turks. is
suggests that a lack of contacts to native-born German peers substantially
accounts for their ethnic penalties. As one might assume that this reduction
could possibly be due to a selective loss of cases between the two models,
model 3 is reestimated only on the basis of those cases also underlying model
5. e results are given in model 4 and reveal that the estimates are nearly the
same. So indeed, the conclusion lies near at hand that the structure of friend-
ship networks is a main factor in explaining the specic diculties of Turks
in the German labor market.
Confirming the Effect of Ethnic Network Composition
Although the results seem convincing, one might nevertheless object
that the conclusion about the importance of assimilative network contacts
may have been drawn too fast, for at least two reasons. First, the correlation
between friendship network and occupational attainment could be spurious
and the eect therefore biased due to a misspecication of the model. is
is the general problem of unobserved heterogeneity. For example, unmea-
sured aspects of human capital, most notably culturally specic skills such
as language prociency, might be the reason for ethnically endogamous net-
works, on the one hand, and for lower occupational attainment, on the other.
Remember that a lack of culturally specic skills has been proposed as a fur-
ther potential mechanism to explain the specic ethnic penalties of Turks.
To tackle this problem, I included other indicators for culturally specic
skills as independent variables in the model. As these variables are measured
e Second Generation in the German Labor Market |
only for nonindigenous youth, the analyses must be restricted to second-
generation immigrants, now comparing Turks to the group of ospring of all
other labor migrants. Models 1 and 2 in table 8.3 reestimate models 3 and 5
in table 8.2 for the subsample of immigrant youth. It is worth noting that one
nds roughly the same results as before. A disadvantage of Turks remains,
even aer controlling for educational qualication (model 1), but it is
reduced and becomes insignicant upon controlling for the ethnic structure
of the friendship network (model 2). e parameter estimate for the friend-
ship network in model 2, table 8.3, is nearly the same as that in model 5, table
8.2, thus indicating that the German reference group does not dominate the
eect strength in the latter model.
Now, in model 3 of table 8.3, two measures of one’s own language pro-
ciency (in German and in the language of the country of origin), as well as
the father’s German-language prociency and his ethnic network structure,
are included in the model to capture culturally specic skills and resources.
Not speaking German very well is a further cause of the problems immi-
grant children face in the labor market and also contributes to explaining
the specic diculties of Turks, since controlling for language prociency in
German leads to a further reduction of the negative eect for Turks. Never-
theless, aer controlling for this and for the other three additional variables,
the strength of the network eect is only slightly reduced. ere is still a sig-
nicant direct impact of ethnic network structure (log-odds eect: –0.58)
independent of these variables.5
An interesting by-product of these analyses can be found in model 3 of
table 8.4: although language prociency in German still has a direct impact
even in the second generation, language prociency with respect to the lan-
guage of the country of origin does not increase the relative labor-market
success at all—even when controlling for German-language prociency. In
the context of discussions on transnationalism and multiculturalism, this is
an important empirical nding, meaning that there is no positive economic
return to bilingualism—the variable “problems with language of coun-
try of origin” even shows the “wrong” sign. In the same spirit, one may be
interested to know whether ethnically mixed friendship networks—that is,
those having both German and ethnic ties—oer a relative advantage over
networks of only German ties. is can be done by categorizing the friend-
ship network indicator in model 4. e nding here is basically the same as
that in the case of language: compared to a network of only German friends,
an ethnically mixed network does not oer any advantage. As expected, a
completely ethnically endogenous network, however, leads to a clear relative
|
.
Log-odds effects on salaried employee position – immigrants only (selected coefficients)
dependent variable measured at time t time t+1
(1) (2) (3) (4) (5)
salaried employee at time t
2.70*
(.24)
other labor migrant
Ref.: Turkish
.49* .34 .25 .27 .07
(.23) (.23) (.23) (.23) (.19)
% best friends German .77* .58* .52*
(.23) (.24) (.20)
no German friend
Ref.: all friends German
-.57*
(.26)
Germ. and other friends -.07
(.23)
father years of edu. .11* .11 .08 .09 .04
(.06) (.06) (.06) (.06) (.05)
father ISEI .00 .00 .01 .01 -.01
(.01) (.01) (.02) (.02) (.01)
father ISEI missing .19 .22 .27 .27 -.10
(.46) (.47) (.50) (.50) (.42)
language probl. German -.35* -.34* -.22
(.15) (.15) (.12)
lang. problems country of origin .12 .12 -.00
(.12) (.12) (.09)
father: lang. problems German -.17 -.18 -.07
(.11) (.11) (.09)
father % friends German .08 .14 .32
(.25) (.25) (.24)
number of persons 533 499 486 486 476
person years 2785 2724 2700 2700 2358
Pseudo-R2.23 .24 .25 .25 .34
data: GSOEP ; in brackets: robust standard errors accounting for clustering on ID; * = p <.
gender, age, age squared, education (full CASMIN), and dummy variables for years are also controlled for in
all models
|
.
Effects (selected coefficients) on percentage of German friends – immigrants only
dependent variable: at
% best friends German
time t
(1)
time t+1
(2)
% best friends German at time t .503*
(.032)
other labor migrant
Ref.: Turkish
.088* .072*
(.029) (.017)
CASMIN 1b
Ref.: 1a
.012 .010
(.045) (.031)
- CASMIN 1c .070 .030
(.051) (.036)
- CASMIN 2b .072 .047
(.048) (.035)
- CASMIN 2a .093 .044
(.057) (.036)
- CASMIN 2c_gen .076 .023
(.057) (.043)
- CASMIN 2c_voc .053 .024
(.070) (.048)
- CASMIN 3a, 3b .174 .081
(.093) (.046)
father: % of best friends German .256* .032
(.045) (.033)
mother: % of best friend German .161* .059
(.049) (.032)
language problems German -.078* -.050*
(.020) (.013)
lang. problems country of origin .046* .039*
(.013) (.008)
father: language problems German -.043* -.019
(.014) (.010)
mother: language problems German -.017 .007
(.014) (.009)
salaried employee (ref. worker) .048 .027
(.027) (.018)
number of persons 483 473
person years 2723 2444
R2.26 .40
data: GSOEP ; in brackets: robust standard errors accounting for clustering on ID; *=p<.
gender, age, age squared, and dummy variables for years are also controlled for in both models
|
disadvantage. us, it seems that the subtype of “selective acculturation,” as
distinguished in the concept of segmented assimilation (Portes and Rumbaut
2001), and ideas on the importance of transnational or pluralistic networks
do not receive much support in the case of second-generation labor migrants
in Germany.
A second major objection to the conclusion that the ethnic structure of
friendship networks matters for labor-market success arises from the gen-
eral problem of endogeneity, that is, the question about the causal relation-
ship between the two variables involved. In the literature (Esser 1980), social
assimilation is oen interpreted as being primarily a consequence of struc-
tural assimilation rather than a cause thereof. e main mechanism stems
from the fact that workplaces constitute important opportunity structures
for meeting and forming friendship ties (Feld 1984; Mouw 2003). erefore,
a h model (model 5) is estimated, which adds the lagged dependent vari-
able to the model (Wooldridge 2003, 300). More precisely, I dene occupa-
tional status (salaried employee = 1; worker = 0; else = missing) at time t+1
as the dependent variable, including occupational status at time t (salaried
employee = 1; else = 0) plus all other independent variables measured at time
t into the model. Model 5 in table 8.3 shows that ethnic network composi-
tion has a signicant eect (0.52) on occupational status at time t+1 that is
independent of occupational status at time t. is nding delivers strong
evidence for the thesis that networks really matter, because it goes against
the hypothesis that the observed correlation is due only to the inuence of
occupational status on friendship structure, which is in turn inert. Or, put
more simply, of two second-generation immigrants who have the same occu-
pational status at time t (and have the same other covariate pattern), the one
with more German friends is more likely to be a salaried employee a year
later.6 Model 5 shows that gender, age, and educational qualications are also
important. All other variables seem to be less important, and the eect of
German-language prociency is signicant only at a 10 percent level.
To complete the story, a nal analysis addresses the reverse question,
that is, whether there is nevertheless an eect in the opposite direction. e
results are given in table 8.4, in which the percentage of German friends now
serves as the dependent variable and a set of plausible predictor variables
includes occupational status. In a model of pooled cross-sections (model 1),
the eect of being a salaried employee is positive, as expected, but signi-
cant only on a 10 percent level. Likewise, the percentage of German friends
tends to rise with the level of education; however, the inuence is also very
weak. In contrast, other variables are much more important, above all lan-
e Second Generation in the German Labor Market |
guage prociency and the ethnic composition of the parents’ network. e
results thus suggest that rather than being a mere consequence of struc-
tural assimilation, network structures seem to result from specic cultural
skills and traits and the respective transmission processes between genera-
tions, thus basically conrming prior research on this topic (Nauck 2001).
Note that in this model Turks still have signicantly fewer German friends
than do the children of other labor-migrant groups, even controlling for all
these factors. Following the idea stated earlier, model 2 uses the percentage
of German friends at time t+1 as a dependent variable, including as predic-
tors the respective percentage at time t and all other variables from model 1
(measured at time t). When analyzing the problem from this perspective, the
eect of occupational status is further reduced. It is, above all, language pro-
ciency that seems to make the dierence when it comes to changes in the
ethnic composition of the friendship networks.
All in all, therefore, in looking at these kinds of models, the evidence is
much stronger that network composition determines occupational posi-
tion rather than that the inuence runs in the opposite direction. Less social
assimilation thus really does seem to play an important part in explaining
the exceptional role of Turks with respect to structural assimilation, that is,
in the labor market.
Summary and Final Remarks
ere is no doubt that second-generation labor migrants in Germany have
improved their labor-market positions relative to those of their parents. In
other words, a marked trend toward economic assimilation over generations
can be found. Nevertheless, the descendants of the former labor migrants
still face signicant disadvantages compared to the indigenous population.
In this analysis of the causes of limited structural assimilation, the results
strongly support prior ndings that it has mainly to do with the diculties
of immigrants’ children in the educational system. Aer controlling for for-
mal qualications, ethnic penalties are nearly absent in most of the second-
generation groups. However, in contrast, Turkish youth still face a specic
ethnic penalty.
In my analysis of rival explanations for this exceptional disadvantage,
there is strong evidence that existing penalties are related to factors other
than labor discrimination in the narrow sense. In addition to the degree of
language prociency in German, the ethnic structure of friendship networks
seems crucial for the occupational attainment of the second generation. As a
|
matter of fact, social assimilation is far less developed among Turkish youth
than among the children of other groups of labor migrants; and controlling
for ethnic network composition, the ethnic penalties of second-generation
Turks almost completely disappear. Using the longitudinal character of the
data set, I was able to further support the view that the impact of social net-
works is indeed direct and causal.
ese ndings immediately give rise to the question of how the miss-
ing social assimilation of second-generation youth, especially Turks, can be
explained. Although I was able to show that culturally specic skills and their
transmission between generations seem to play an important part, additional
explanations lie near at hand, among them discrimination. It is important
to note that according to my analyses, labor-market discrimination does not
seem responsible for the specic disadvantage of Turks; however, I do not
rule out the possibility that discrimination may occur in relevant processes
preceding entry into the labor market. Here, further theoretical elaboration
of the exact mechanisms and further empirical research are urgently needed
in order to understand the complex processes through which ethnic inequal-
ity in Germany is reproduced.
e important role that the ethnic composition of networks plays in
understanding the processes of occupational attainment in Germany also
challenges traditional views of assimilation as well as newer theoretical con-
cepts and frameworks. One, mostly implicit but sometimes even explicit,
assumption of many assimilation theorists is that there is a denite, albeit
imperfect, causal order among several dimensions of assimilation. While
cognitive assimilation (acculturation) is seen as a necessary precondition
of structural assimilation, social assimilation is primarily seen as its conse-
quence. Assuming that immigrants are interested in the benets of structural
assimilation, the idea is that they will sooner or later (in terms of genera-
tions and birth cohorts) invest in the cognitive requisites, and social assimi-
lation will then only be a matter of time, as structural assimilation provides
the necessary opportunity structures. My results, however, demonstrate that
the feedback eects of social assimilation on the “prior” dimensions may be
more severe than has long been suggested. is is underpinned by recent
parallel research in Germany that has revealed the likewise important eects
of social assimilation on the school-choice behavior of immigrant parents
(Kristen 2004) and even on their children’s success in German soccer (Kalter
2003). Such feedback eects do not imply, however, that there is no base-
line trend toward assimilation or that there is even a trend in the opposite
direction. Nevertheless, given that eects may be cross-generational and that
e Second Generation in the German Labor Market |
there are mechanisms of direct intergenerational transmission of social net-
works (Nauck 2001), the speed of assimilation may be reduced considerably.
On the other hand, the current results clearly indicate that for the second
generation in present-day Germany there seems to be no path to economic
success other than the routes of the mainstream society. Besides the predom-
inant role of educational qualications, it is capital specic to the receiving
society that accounts for residual disadvantages. In contrast, ethnic capi-
tal—that is, capital specic to the country of origin—does not lead to any
increase in labor-market success at all, even when controlling for all other
assets. erefore, for the second generation in Germany there does not seem
to be any promising third alternative of “selective acculturation” or “plural-
ism” between straight-line assimilation, on the one hand, and permanent
economic disadvantage, on the other.
. As I report summary statistics separately for ethnic groups in the descriptive
section and use logistic regression models that deliver unbiased estimates if the sample
is exogenously stratied but the models are otherwise correctly specied (Hosmer and
Lemeshow , ), I do not use design weights in my analyses.
. More precisely, EGP classes I, II, and IIIa+b are recoded to , and classes V, VI, and
VIIa+b to . EGP classes IVa–c are treated as missing values.
. In the meantime, analyses similar to the ones in this chapter have been conducted
using alternative indicators of labor-market success. For example, Kalter () looks
at employment versus unemployment and at qualied labor versus nonqualied labor.
Results and conclusions are rather similar.
. e last nding suggests that the average duration of observation of respondents
in the panel data is rather similar for all the groups. Note that, in general, the mean for
all metric time-varying variables is computed out of individual means over time, without
weighting for number of years observed.
. Given that there is panel information, a more severe test of whether there is indeed
a direct eect of networks on occupational attainment could be obtained by estimating
a xed-eects model. As this model accounts for a xed individual-specic eect, it
controls for all unmeasured variables that do not change over time (see Wooldridge ,
). However, as there is little variance in the binary dependent variable, one can run this
model with only persons. is leaves the coecient (.) to pure chance to attain
signicance (p = .). It is worth noting, however, that a somewhat changed dependent
variable SKILLED (= for EGP I, II, IIIa+b, V, VI; = for EGP VIIa+b; missing else)
leads to a highly signicant eect (., p = .) in a xed-eect model with per-
sons. All in all, my impression, given the available data, is that although the network eect
in table . (model ) and table . (model ) may be positively biased, there seems to be a
direct relationship between the ethnic structure of friendship networks and occupational
attainment.
|
. In addition to the analyses presented here, event-history models were run as an
alternative to test the causality direction. ey also support the view that networks are
a cause of success. For example, in a discrete event-history model analyzing the risk of
gaining salaried-employee status there is also a signicant positive eect of the network
indicator, which reduces the ethnic penalty of Turks considerably. See also Kalter for
similar analyses.
