Content uploaded by Annette Bergemann
Author content
All content in this area was uploaded by Annette Bergemann
Content may be subject to copyright.
DISCUSSION PAPER SERIES
ABCD
www.cepr.org
Available online at: www.cepr.org/pubs/dps/DP8295.asp
www.ssrn.com/xxx/xxx/xxx
No. 8295
THE THREAT EFFECT OF
PARTICIPATION IN ACTIVE LABOR
MARKET PROGRAMS ON JOB
SEARCH BEHAVIOR OF MIGRANTS IN
GERMANY
Annette Bergemann, Marco Caliendo,
Gerard J van den Berg
and Klaus F Zimmermann
LABOUR ECONOMICS
ISSN 0265-8003
THE THREAT EFFECT OF PARTICIPATION IN
ACTIVE LABOR MARKET PROGRAMS ON JOB
SEARCH BEHAVIOR OF MIGRANTS IN GERMANY
Annette Bergemann, Universität Mannheim, IZA, and IFAU-Uppsala
Marco Caliendo, IZA, DIW Berlin and IAB
Gerard J van den Berg, Universität Mannheim, IZA, IFAU-Uppsala, VU
University Amsterdam and CEPR
Klaus F Zimmermann, IZA, Universität Bonn and CEPR
Discussion Paper No. 8295
March 2011
Centre for Economic Policy Research
77 Bastwick Street, London EC1V 3PZ, UK
Tel: (44 20) 7183 8801, Fax: (44 20) 7183 8820
Email: cepr@cepr.org, Website: www.cepr.org
This Discussion Paper is issued under the auspices of the Centre’s research
programme in LABOUR ECONOMICS. Any opinions expressed here are
those of the author(s) and not those of the Centre for Economic Policy
Research. Research disseminated by CEPR may include views on policy, but
the Centre itself takes no institutional policy positions.
The Centre for Economic Policy Research was established in 1983 as an
educational charity, to promote independent analysis and public discussion
of open economies and the relations among them. It is pluralist and non-
partisan, bringing economic research to bear on the analysis of medium- and
long-run policy questions.
These Discussion Papers often represent preliminary or incomplete work,
circulated to encourage discussion and comment. Citation and use of such a
paper should take account of its provisional character.
Copyright: Annette Bergemann, Marco Caliendo, Gerard J van den Berg and
Klaus F Zimmermann
CEPR Discussion Paper No. 8295
March 2011
ABSTRACT
The Threat Effect of Participation in Active Labor Market Programs
on Job Search Behavior of Migrants in Germany*
Labor market programs may affect unemployed individuals’ behavior before
they enroll. Such ex ante effects may differ according to ethnic origin. We
apply a novel method that relates self-reported perceived treatment rates and
job search behavioral outcomes, such as the reservation wage or search
intensity, to each other. We compare German native workers with migrants
with a Turkish origin or Central and Eastern European (including Russian)
background. Job search theory is used to derive theoretical predictions. We
examine the omnibus ex ante effect of the German ALMP system, using the
novel IZA Evaluation Data Set, which includes self-reported assessments of
the variables of interest as well as an unusually detailed amount of information
on behavior, attitudes and past outcomes. We find that the ex ante threat
effect on the reservation wage and search effort varies considerably among
the groups considered.
JEL Classification: C21, D83, D84, J61 and J64
Keywords: active labor market policy, expectations, immigrants, policy
evaluation, program evaluation, reservation wage, search effort and
unemployment duration
Annette Bergemann
Tinbergen Institute Amsterdam
Roeterstraat 31
1018 WB Amsterdam
THE NETHERLANDS
Email: bergemann@tinbergen.nl
For further Discussion Papers by this author see:
www.cepr.org/pubs/new-dps/dplist.asp?authorid=152069
Marco Caliendo
Director of Research
Institute for the Study of Labor - IZA
Schaumburg-Lippe-Str. 5 - 9
D-53113 Bonn
GERMANY
Email: caliendo@iza.org
For further Discussion Papers by this author see:
www.cepr.org/pubs/new-dps/dplist.asp?authorid=157674
Gerard J van den Berg
Alexander von Humboldt Professor
Dept. of Economics L7, 3-5
Mannheim University
68131 Mannheim
Germany
Email: gberg@feweb.vu.nl
For further Discussion Papers by this author see:
www.cepr.org/pubs/new-dps/dplist.asp?authorid=126469
Klaus F Zimmermann
Director
Institute for the Study of Labor - IZA
Schaumburg-Lippe-Str. 5-9
D - 53113 Bonn
GERMANY
Email: Zimmermann@iza.org
For further Discussion Papers by this author see:
www.cepr.org/pubs/new-dps/dplist.asp?authorid=102751
* We thank Olof Åslund, Corrado Giulietti and two anonymous Referees for
their comments and suggestions. The IAB (Nuremberg) kindly gave us
permission to use the administrative data. Caliendo thanks the German
Research Foundation (DFG) for financial support of project CA 829/1-1.
Submitted 5 March 2011
1 Introduction
Migrants are often disadvantaged with respect to their labor market outcomes. In Germany
the unemployment rate of foreign born men was 11.8% in 2008, whereas only 6.8% of
native born men were unemployed. For women the figures are similar at 13.1% vs. 6.8%
(see OECD, 2010). Past research has shown that differences in employment and earnings
between natives and migrants persist, even when controlling for individual characteristics,
such as education and age (Algan et al., 2010).
Active labor market policy programs (ALMP) are common tools to improve the labor
market outcomes of the unemployed. Potentially, these programs could be particularly
helpful for unemployed migrants. Migrants may lack skills specific to the national labor
market, and these skills might be easily transferred with the help of labor market policies
such as training programs.
ALMP can have ex ante and ex post effects. Evaluation studies typically focus on ex
post effects, i.e. on the effect of actual participation. However, ex ante effects, i.e. the effects
that occur before participation, may also exert a large effect on unemployment durations.
If individuals expect large benefits from a treatment then they may postpone their job
search until after the treatment. In this case the ex ante effect is negative, and the average
realized unemployment durations of participants and non-participants may be larger than
in the absence of the program. Knowledge of ex ante effects is thus an important input
for the evaluation of the program and for the assessment of possible modifications of the
program. For example, if the ex post effect of having been trained on the exit rate to work
is positive, whereas the ex ante effect on this rate is negative, then this may suggest that
the program is best offered early on during the unemployment spell.
Ex ante effects require individuals to have some knowledge about the existence of ALMP
and about the process leading to participation. The ALMP participation probability is a
determinant of the optimal job search strategy and will affect the outcome of interest.
Consider, for example, a training program that upgrades skills for a certain profession.
Knowledge of the rate that an individual can participate in such a program can be valuable
for the individual. If an individual finds out that her individual rate is high then it becomes
attractive for the individual to reduce her search effort as participation could lead to higher
wages.
In this paper we investigate whether migrants and natives react similarly to the expec-
tation of participating in an ALMP, and whether different types of migrants (as captured
by the region of origin) react similarly or not. There are a number of reasons why the
migrant status may affect ex ante effects. First, ex ante effects are affected by the extent
to which individuals enjoy the participation experience itself, which may depend on the
composition of the program. Second, the effects depend on the degree of familiarization
2
with the services provided by the employment office and with the baseline expectations
about the extent to which the state is perceived as a helpful or as a threatening institution.
Some types of migrants may not be aware of the existence of a program at all. Finally,
the magnitude of the ex ante effect depends on the magnitude of the ex post effect, so
differences in ex ante effects between different types of unemployed may be due to differ-
ences in ex post effects. Notice that we do not aim to distinguish formally between these
possible explanations. The latter primarily serves as a motivation for why differences may
exist. In order to put the results into perspective, we also examine whether natives and
migrants have the same job search strategies in terms of search effort and reservation wage
values. Any differences in the ex ante effects between the different types of unemployed
could help to fine tune the allocation of ALMP and, thus, could help reduce the labor
market disadvantages of migrants.
Some bodies of empirical work are relevant for the present study. Evaluation studies of
ex post effects by migrant status provide heterogeneous results. ALMP are partly success-
ful and partly ineffective. In rare cases they even seem to be harmful ex post. However, an
important lesson from these studies is that migrants are often affected differently by ALMP
compared to natives, giving additional emphasis to the importance of our approach to in-
vestigate the ex ante effects of ALMP separately. German studies that distinguish between
natives and migrants mainly focus on welfare recipients. Huber et al. (2009) evaluate three
different types of welfare to work programs and find positive effects of these programs for
natives but not for migrants. Aldashev et al. (2010) evaluate short-term training schemes.
They estimate positive effects for aptitude tests that are larger for migrants than for na-
tives, and positive effects for short-term skill provision that are especially large for female
immigrants. Surprisingly, they find that job search training is generally ineffective and even
has a negative effect for female immigrants. Caliendo and K¨unn (2011) analyze the effects
of start-up subsidies for the unemployed and show that both natives and migrants benefit
from participation. However, they find higher effects for natives in terms of employment
probabilities and income.
There are very few studies on ex ante effects, and they only consider averages of treated
individuals rather than specific subgroups like migrants. Black et al. (2003) use locally ran-
domized assignment of treatment status to examine empirically whether this affects the
voluntary inflow into unemployment insurance (UI) benefits. Here, the treatment regime
starts right after entering the UI regime. Abbring and Van den Berg (2005) show that
if the moment of treatment has a random element, if the observed treatment and labor
market outcomes are duration variables, and if there is randomized variation in the treat-
ment intensity, then identification of ex ante effects still requires a semi-parametric model
structure and absence of anticipation of the moment of treatment (that is, no anticipation
beyond what is captured in the treatment assignment equation; see Rosholm and Svarer,
3
2008, for an application). De Giorgi (2005) and Van den Berg, Bozio and Costa Dias (2008)
use a policy discontinuity in time to study the effect of a treatment at a six-month unem-
ployment duration on the probability of finding work before six months. Specifically, they
compare a situation where individuals in the inflow are aware of the policy to a situation
where the policy regime has not yet been introduced. Lalive, Van Ours and Zweim¨uller
(2005) observe whether and when unemployed individuals receive advance warnings about
the timing of future treatments. By viewing such warnings as treatments themselves, they
can apply the semi-parametric timing-of-events framework of Abbring and Van den Berg
(2003) to study their effect.
Our approach builds on the study by Van den Berg, Bergemann and Caliendo (2009)
that develops and applies a novel general method to identify ex ante effects. Specifically,
they identify ex ante effects of the comprehensive German package of active labor market
policy by using self-reported variables of unemployed workers in a panel survey. The unem-
ployed are asked about their perceived probability of being treated in future periods, and
they are also asked about their current optimal job search strategy, notably their current
reservation wage and their current search effort. All things equal, the expectation of a
future event that changes the individual’s expected present value should have an effect on
the reservation wage or, in general, change the current search effort. They find that the ex
ante effect on the reservation wage and search effort are negative and positive, respectively.
This means that individuals try to prevent program participation by accepting worse jobs
and searching harder than they would do if the programs were absent. They conjecture that
this is due to a large extent to individuals disliking the actual participation experience.
Van den Berg, Bergemann and Caliendo (2009) use information from the first survey
wave of the IZA Evaluation Data Set (see Caliendo et al., 2011, for details). This is an ongo-
ing data collection process in which an inflow sample of unemployed in Germany is followed
over time. They use information from the first survey wave. The survey interviews were
held in late 2007 and early 2008 with individuals who had recently become unemployed.
Respondents answered an extensive set of questions inter alia about their search behav-
ior, reservation wages, previous employment experience, and expectations about program
participation.
Similar to Van den Berg, Bergemann and Caliendo (2009), we use the first wave of
the IZA Evaluation Data Set. To some extent we use matching in order to estimate the
effect of different participation expectations on the reservation wage and the search effort.
The matching approach is well-suited to dealing with individual heterogeneity. The data
contain a number of self-reported personality and behavioral assessments and individual
past labor market outcomes, which allow for a rich set of conditioning variables in the
matching procedure. However, migrants with specific countries of origin constitute small
subsamples of the full sample, which is why we frequently resort to regression methods.
4
The paper has the following structure. Section 2 introduces the job search model on
which we build our empirical approach. This model allows individuals to receive utility or
disutility from participation in ALMP for reasons other than their effect on labor market
outcomes. The data source, the variables used and the definition of migrants are described
in Section 3. Section 4 presents the estimation results and Section 5 concludes.
2 The job search model
This section summarizes the job search model developed in Van den Berg, Bergemann and
Caliendo (2009). In this model the unemployed search sequentially for a job. For a given
level of search effort s, job offers arrive at a certain rate λs. Offers are random drawings
from a wage offer distribution F(w). If an offer arrives, the individual must decide whether
to accept the wage offer (and keep it forever) or reject it and continue searching at least
until the next offer arrives. The individual receives benefits bwhile unemployed and incurs
search costs c(s), which depend on the search effort level. The aim of the individual is
to maximize the expected present value of income or utility over an infinite horizon. The
optimal search strategy can then be described by the reservation wage φ, giving the minimal
acceptable wage offer and an optimal level of search effort s.
Program participation is then introduced into this basic job search model.1An individ-
ual that has not been treated enters at a specific rate η≥0 into treatment. What actually
matters is the perception of individuals about this entrance rate. For convenience, how-
ever, we do not distinguish in the text between the perceived rate and the actual treatment
rate. Treatment can have an effect on the job finding rate parameter λand/or the wage
offer distribution F(w) and the individual is aware of these effects. Concerning treatment,
however, the individual does not know the exact moment of treatment, only the rate at
which it occurs.
The expected present value without treatment is R. With treatment, the expected
present value changes to Rp. The total gain Gdue to treatment can be described by
G=Rp−R−γ, where γcaptures the direct costs of treatment. These costs can be
positive or negative, depending on whether the individual dislikes or likes the treatment.
Similarly, the treatment effect Rp−Rdue to changes in the labor market position can also
be positive or negative.
In this new setting the reservation wage φand the optimal search effort level sdepend
on the total gains Gof the treatment and on the rate ηat which the treatment occurs.
In order to make this explicit, we write φ(η, G) and s(η, G). Ex ante effects can then be
described by the difference in the reservation wage φand the optimal search effort level
1The main insights are robust with respect to the model assumptions; see the discussion in Van den
Berg, Bergemann and Caliendo (2009).
5
sthat arise if we compare a world without treatment, (η= 0 and G= 0) with a world
with treatment (η=η0and parameter values λpand Fpleading to G=G0), i.e. by the
difference φ(η0, G0)−φ(0,0) and s(η0, G0)−s(0,0).
Van den Berg, Bergemann and Caliendo (2009) show that if the total gain is positive
and treatment occurs with a positive rate (G > 0 and η > 0), then the ex ante effect on
the reservation wage φis positive, whilst the effect on the optimal search effort level sis
negative. As a results, individuals become more choosy and search less extensively. Both
effects reduce the transition rate from unemployment to employment. Similarly, if the gains
become negative (G < 0) and the treatment rate is positive (η > 0), then the effect on the
reservation wage will be negative (φ(η0, G0)−φ(0,0) <0) and on the search effort level
positive (s(η0, G0)−s(0,0) >0). Consequently, if we compare the reservation wage (search
effort level) of similar individuals, where one group report a certain value η > 0 with the
reservation wage (search effort level) of another group that report a value of η= 0, one is
able to draw conclusion on the sign of the total gain of treatment G. Thus, the empirical
signs of dφ/dη and ds/dη can be used to infer whether G≷0. If the empirical signs of
dφ/dη and ds/dη are zero then there are no ex ante effects, and it can be concluded that
G= 0, so either the program is ineffective, or the program is beneficial but the individual
dislikes the treatment itself.
The ex ante effects may be heterogeneous. The extent to which the treatment entry rate
influences the optimal job search strategy reflects the effect of treatment on the expected
present value. Simultaneously, model determinants that lead to a high rate of moving
from unemployment to employment reduce the ex ante effects. Formally, the derivatives
dφ/dη and ds/dη depend on all the other model determinants, which means that the
effect of the treatment entry rate ηon the reservation wage φand on the optimal search
effort level sinteracts with all other model determinants, leading to heterogeneous ex
ante effects. Consequently, the matching method is particularly well-suited to determining
ex ante effects, as it allows for effect heterogeneity. Moreover, matching does not impose
functional form restrictions and it is explicitly clear about the weighting procedure used to
estimate average treatment effects (see also Section 4). However, adequate application of
matching requires a sufficiently large sample size. Migrants of specific origins may constitute
small samples, so we frequently resort to regression methods where we estimate interaction
effects to capture effect heterogeneity.
3 Data
In the empirical analysis we estimate ex ante effects for recently unemployed workers of the
comprehensive German package of active labor market policies, for both native Germans
and migrants separately. The most prominent ALMP in Germany are short training pro-
6
grams and job search assistance schemes. However, start-up subsidies for the unemployed,
job creation programs, long-term (re-)training programs and wage subsidies for jobs in the
private sector are of quite considerable size as well (see Eichhorst and Zimmermann, 2007,
or Bernhard et al., 2008, for recent overviews). In Germany, as in other European countries,
case workers have a large influence on the (timing of the) participation of an unemployed
worker in ALMP. Recently unemployed individuals are typically assigned to job search as-
sistance programs or training programs. Long-term unemployed individuals are more often
assigned to employment programs, consisting of either wage subsidy programs for jobs in
the private sector or job creation schemes.
The data we use are from the IZA Evaluation Data Set. As explained in Section 1,
this survey data set targets an inflow sample into unemployment from June 2007 to May
2008. The key feature of the data set is that individuals are interviewed shortly after they
become unemployed and are asked a variety of non-standard questions about attitudes and
expectations (see Caliendo et al., 2011, for details). The sampling is restricted to individ-
uals who are 16 to 54 years old and who receive or are eligible to receive unemployment
benefits under the German Social Code III. From the monthly unemployment inflows of
approximately 206,000 individuals in the administrative records2, a 9% random sample is
drawn which constitutes the gross sample. Out of this gross sample representative samples
of approximately 1,450 individuals are interviewed each month, so that after one year 12
monthly cohorts are gathered.
For the first wave 17,396 interviews were conducted and individuals were interviewed
about two months after becoming unemployed. We restrict our analysis to individuals who
are still unemployed and are actively searching for a job. That is, we exclude individuals
who have already found a job, are participating in a program or are not searching for
other reasons.3This leaves us with a preliminary sample of 8,612 individuals, from which
we further exclude the lowest and highest percentile of the reported hourly reservation
wage and the reported benefit level as well as individuals with missing values for any key
variables. The final results is a sample of 7,913 individuals.
Throughout the paper we use a broad definition for migrants and individuals with
migration background (called migrants henceforth). We define an individual as a migrant
if the individual is either born abroad, or not in possession of a German passport, or with
either a father or a mother who was born abroad. With this definition we basically cover
first and second generation migrants. As migrants themselves cannot be expected to be a
homogenous group with respect to their labor market behavior, we differentiate between
2Administrative records are based on the “Integrated Labour Market Biographies” of the Institute for
Employment Research (IAB), containing relevant register data from four sources: employment history,
unemployment support recipience, participation in active labor market programs, and job seeker history.
3Of these three categories, program participation is by far the smallest.
7
major immigration groups. We distinguish between individuals originating from Central
and Eastern Europe/former Yugoslavia, Russia and Turkey. We compare the results found
for these migration groups with those of the native Germans. Note that although Germany
in the 1960s also experienced a major inflow of Italians, our sample of Italians is too small
to consider them a separate group. Instead, we remove the migrants from the other parts
of Europe, America, Africa and Asia from our data, and the resulting sample consists of
7,147 individuals.
Table 1 reports some descriptive statistics. Migrants are on average either younger or
have the same age as natives (30 to 36 years vs. 36 years). Russians and migrants from
Central and Eastern Europe/former Yugoslavia have been staying on average longer in
Germany than migrants from Turkey (12 to 13 years vs. 10 years). Migrants tend to live
predominantly in West Germany (84%-93% vs. 63%), are more often married (47%-54%
vs. 38%) and have more children than natives (35%-51% vs. 31% have children, 0.51-0.89
vs. 0.48 on average). In addition, the share of unemployment benefit recipients among the
group of migrants is lower than among the group of native Germans (73%-77% vs. 79%).
The educational background of migrants varies a lot. Compared to 25% of native Germans,
who have a high school degree, 32% of migrants from Central and Eastern Europe/former
Yugoslavia also have one, whereas this is true of only 17% of Turkish and 18% of Rus-
sian migrants. Similarly, the previous labor market history in the group of migrants is
very diverse. Individuals from Central and Eastern Europe/former Yugoslavia and Rus-
sia experience fewer months of unemployment (measured relative to the years since the
18th birthday) than natives (0.65-0.71 vs. 0.83 months), whereas migrants with a Turkish
background are more often unemployed than natives (0.89 months). At the same time,
however, all migrant groups had fewer previous employment months than natives (again
measured relative to the years since the 18th birthday, 6.12-7.89 vs. 8.45 months). It is sur-
prising that Russians and the Central and Eastern Europeans/former Yugoslavians have
simultaneously shorter unemployment and employment spells, even when age is controlled
for. One reason could be that this group of migrants spends more time in education and
military service.
Individuals are also asked questions regarding their “locus of control”, which is a gen-
eralized expectancy about internal versus external control of reinforcement (Rotter, 1966).
Whereas individuals whose external locus of control personality trait dominates believe
that everything that happens is beyond their control, people with an internal locus of con-
trol are confident that outcomes are contingent on their decisions and behavior.4Natives
4Locus of control is measured by a set of statements to which individuals could reply on a scale of “1”
(I do not agree at all) to “7” (I agree fully), e.g., “How my life takes course is entirely dependent on me”
or “Success is gained through hard work”. We sum up the positive answers and build a single dummy
variable if the answers exceed a certain threshold.
8
do report an average of 5.04 with respect to their locus of control, whereas it is lower for
Russians (4.87) and people from Central Eastern Europe/former Yugoslavia (4.95) and
Turkey (4.81).
The key variable for our analysis, the entry rate into treatment η, is measured by the
answer to the question how likely it is that ALMP participation occurs conditional on re-
maining unemployed in the next three months. This explicitly merges all ALMP measures
(the main ALMP for short-term unemployed workers are training, job search assistance,
and subsidized work). The answers range from 0 (“very unlikely”) to 10 (“very likely”).
For the analysis we construct a binary measure by grouping 0−4 into the category “η-low”
and 5 −10 into “η-high”. The search effort sis operationalized as the number of search
channels used where the maximum number is 10.5This is in line with e.g. Van den Berg
and Van der Klaauw (2006) and references therein who also use this outcome as an indica-
tor of search effort. On average, 57% of native Germans find it highly likely to participate
in a program of ALMP (see Table 2). The probability is higher for migrants. Note that the
Turkish most often report that it is likely they will participate in an ALMP (69%), whereas
Russians (63%) and unemployed migrants from Central and Eastern Europe/former Yu-
goslavia (61%) are more similar to natives in that respect. Major differences in the average
values of the variables describing the search strategies only exist between the native Ger-
mans and the migrants with a Turkish background. The average reservation wage and the
average number of search channels of native Germans are e6.89/hour and 5.12 channels,
whereas Turkish migrants have on average a higher reservation wage (e7.35) and a smaller
number of search channels (4.75), despite the fact that they are, for example, less educated
than the average native German. Russians and Central and Eastern Europeans/former
Yugoslavians have an average reservation wage of e6.72 Euro and e7.02 respectively, and
use on average 4.92 and 5.23 search channels.
4 Empirical Analysis
As a first step we conduct a simple regression analysis of the logarithm of the reserva-
tion wage (log φ) and the number of search channels (s) on the expected participation
probability η. We include a large set of additional explanatory variables, for example,
5Individuals were asked the following questions, where multiple entries were allowed: “What have you
done in order to find an apprenticeship or employment? Have you searched... 1: through job advertisements
in the newspaper, 2: by personally advertising as a job seeker, 3: through a job information system,
4: through contact with acquaintances, relatives, other private contacts, 5: through an agent from the
employment agency, 6: through internet research, 7: through a private agent with agency voucher, 8:
through a private agent without agency voucher, 9: through blind application at companies 10: other, 11:
nothing of its kind.” We take the sum of all answers as the number of search channels.
9
years since migration as a percentage of age, individual past labor market history, bene-
fit level, education, regional indicators, marital status, number of children, age, means of
communications, and month of entry in unemployment. For second-generation immigrants,
years since migration is set to zero, as for natives. Additionally we control for personality
traits such as locus of control, openness, conscientiousness, extraversion and neuroticism
which have proven to be important in recent labor market research (see e.g. Borghans,
Duckworth, Heckman, and Ter Weel, 2008). We pool all individuals independently of their
origin. However, we introduce dummy variables for the different migration groups as well
as interaction terms that combine the migration background and ηin order to accomplish
our goal to investigate whether there are differences in search strategies and differences in
the ex-ante effects.
The regression results on the determinants of the reservation wage in Table 3 confirm
the descriptive results that Turkish migrants have a higher reservation wage than all other
groups. The dummy on Turkish origin is positive and highly significant (10%). Furthermore,
the results suggest that native Germans reduce their reservation wage in case they belong to
the “η–high” category by 2.8%. The coefficients on the other interaction terms are slightly
larger, however, not significant.
With respect to search channels we find some differences between the different groups
of origins. Russians seem to have a somewhat lower search effort level than the other
groups (the coefficient on Russians is negative, but only significant at the 10% level.) Most
importantly for our analysis of the ex-ante effects, the OLS estimates suggest that native
Germans, Central and Eastern Europeans/former Yugoslavians and Russians significantly
increase their search effort when they expect to participate in an ALMP. Only individuals
with a Turkish background do not change their search effort level in response to their
expected extent of participation in ALMP. This is in contrast to all other groups considered
here. We conclude that the Turkish migrants are quite different in this respect from natives
and the two other migration groups (Central and Eastern Europeans/former Yugoslavians
and Russians). Notice that years since migration does not have a significant effect. One
explanation for this is that this variable contrasts second-generation immigrants and natives
on the one side to first-generation migrants on the other side. We return to this in the
concluding section of the paper.
As a next step, we proceed by using propensity score matching (introduced by Rosen-
baum and Rubin, 1983; see Caliendo and Kopeinig, 2008, Imbens and Wooldridge, 2008,
and Blundell and Costa Dias, 2009, for recent overviews) in order to estimate the average
“treatment–on–the–treated” effect (ATET), where of course in our case the “treatment” is
the entry rate into ALMP participation. As we need a reasonable sample size for matching,
we merge Central and Eastern Europeans/former Yugoslavians and Russians, and call this
group henceforth CEER. Despite matching being able to handle heterogeneity of treatment
10
effects, we omit Turkish migrants, as they seem to react so very differently.6Note that the
treatment consists in being in the “η–high” group as compared to being in the “η–low”
group. Thus, we regard a high subjective probability of participating in an ALMP condi-
tional on staying unemployed as the treatment. After estimating the propensity score for
the probability of being in the “η–high” category (we use the same rich set of conditioning
variables as in the OLS regressions, see Table 4 for the score estimates and Figure 1 for
the score distribution), we perform Kernel-matching7in order to obtain ATET estimates.
The estimation procedure is conducted separately for migrants and natives. In addition,
we also want to investigate whether there are differences in the search behavior between
natives and migrants, conditional on them being in a certain “η” category. In this case
we interpret being a migrant as a treatment vs. being a native German. One can think of
this approach as a thought experiment which answers the question: What would happen
to the reservation wage and the search intensity of a native German with the average char-
acteristics of a migrant and who would become a migrant. Here, we omit the ”years since
migration” as it would be heavily related to the outcome measure.
Let us turn to the estimation results for the matching analyses. For both native Germans
and CEER, we find that the ex ante effects on the reservation wage are negative and
on the search effort are positive, respectively (see Table 5). Having the perspective of
otherwise going into a program of ALMP, native Germans and CEER migrants are willing
to accept worse jobs by lowering their reservation wage. Natives with a high ηlower their
reservation wage by 3.0% and CEER migrants by 3.9%, but the coefficient is not significant
at conventional levels (p-value: 0.12). Both migrants and natives with a high ηalso search
significantly harder, and the effect is larger for migrants (0.46 or 9% more channels) than
natives (0.22 or 4.3% more channels). This shows that both migrants and natives with high
ηactively try to prevent participation.
We now address the question whether there are differences between migrants and native
Germans within the “η–low” or “η–high” group, respectively. The estimates of ATET are
also presented in Table 5, in which we do not find significant differences for the “η–low”
groups. This means that migrants from CEER who are part of the “ η-low” group have
the same average reservation wage and average search effort as similar native Germans.
Furthermore, we are not able to find significant differences in the average reservation wages
for the “η–high” groups. However, we do find that CEER migrants increase the search effort
level even more than similar native Germans in view of a likely participation in ALMP.
6Indeed, when including Turkish migrants the results seem to be partly driven by their inclusion. At
the same time the group of Turkish migrants is too small in order to use matching for them alone. Due to
insufficient overlap we would need to discard 7% of all “treated”.
7For the kernel matching procedure, we use an Epanechnikov kernel with a bandwidth of 0.06 and
impose the common support condition based on the “MinMax” criterion.
11
Thus, CEER migrants try to prevent participation in ALMP even harder than native
Germans by searching more intensively.
Both sets of matching results are in full agreement with the OLS results. With this in
mind, we may return to the OLS estimates for Turkish migrants. Those results suggest
that Turkish migrants try to prevent participation by reducing the reservation wage by a
similar amount to native Germans. However, note that the initial level of the reservation
wage of Turkish migrants is higher than the one of similar natives. At the same time,
Turkish migrants do not show signs of adjustment in their search effort. From this we
conclude that migrants with a Turkish background struggle less than native Germans and
CEER migrants to prevent participation in ALMP.8
In addition, we conduct a separate matching analysis for Russians and for Central
and Eastern Europeans/former Yugoslavians. As the OLS results already indicate, we find
some differences between these two groups with regard to the search effort level. However,
further research is required to investigate any differences between these groups because the
current sample precludes an in-depth investigation; whereas with additional waves of the
panel, we will be able to take advantage of multiple observations per individual and exploit
the information in realized outcomes.
The analysis naturally raises the question where the differences between native Germans
and migrants might originate from, and why there are differences between migrants of
different origins.
One potential explanation could be that individuals of different origins interpret ques-
tions in the interview in different ways. In one sense, we can rule out this possibility,
because in many cases native speakers were used when interviewing migrants. In partic-
ular, Turkish speakers were used to interview Turkish respondents, and Russian speakers
were used for respondents who were fluent in Russian but not in German. What we cannot
rule out is that respondents from certain regions of origin have difficulties with the concept
of a reservation wage as such because they expect to be able to bargain over the wage
and other job characteristics. Another potential explanation is that the groups differ with
respect to the type of jobs they expect and that the differences in reservation wage reflect
compensating wage differentials. However, these explanations only explain different levels
of the reservation wage. As a first-order approximation, it should not be able to explain
differences in ex-ante effects.
Yet another potential explanation for the differences in the ex ante effects is that in-
dividuals of different origins have different degrees of awareness of the program (see De
Graaf-Zijl, Van den Berg and Heyma, 2011). However, in our data only very few migrants
(and native Germans) reported not to know about ALMP programs. Therefore, we can
reject this possibility as well.
8The matching estimates for Turkish, although based on very few observations, support these results.
12
If differences in the information sets are excluded, then, according to the theoretical
model (see Section 2), the differences in the ex ante effects either derive from differences in
the ex post treatment effects due to changes in the labor market position or from differences
in direct costs of treatment. At the current state of research, we can only speculate which
of these aspects might play a role here. The evidence on ex–post treatment effects is
insufficient in order to state with some certainty that they are (partly) responsible for the
heterogeneity in the ex ante effects; but naturally they remain a potential candidate for
the explanation.
One could, however, also suspect that the direct costs of treatment differ between
the different nationalities. For example, the higher search intensity of CEER migrants to
prevent participation could originate from a stronger dislike of governmental intervention
compared to native Germans and Turkish individuals or from the special involvement of
case workers, which makes search cheaper.
In contrast to natives and, even more so, in contrast to CEER migrants, Turkish in-
dividuals do not increase their search intensity if they face participation in ALMP. They
either benefit more from participation in ALMP or their direct costs of participation are
lower. One supporting aspect why the direct costs of participation could be responsible is
related to the higher reservation wages of Turkish individuals compared to other nation-
alities. The higher reservation wage could reflect that Turkish have a lower nonpecuniary
disutility of unemployment (probably due to neighborhood effects of living close to other
unemployed individuals). This lower nonpecuniary disutility of unemployment could also
transfer to lower costs of participation in ALMP.
5 Conclusions
Using a recently developed method to determine ex ante effects of participation in ALMP,
this paper uncovers the heterogeneity of these effects according to migrant status in Ger-
many. We find that the search behavior of Turkish migrants is not affected by the probabil-
ity of future participation in ALMP. There is a moderate threat effect for native Germans,
while individuals from Central and Eastern Europe, Russia and former Yugoslavia increase
their job search behavior most in order to prevent participation. We speculate that next
to differences in the ex post treatment effect, the differences in ex ante effects may be
driven by differences in the direct cost of treatment, perhaps derived from a dislike of
governmental intervention (for the Central and Eastern Europeans, Russians and former
Yugoslavians) and lower disutility of staying unemployed and participating in a program
(Turkish migrants).
Note that prevention of participation is not per se in the interest of society. In order to
return to work fast, individuals might accept jobs in which their productivity is not fully
13
exploited. Therefore, it is interesting to investigate further the sources of the heterogeneity
and to examine realized post-unemployment outcomes. The results of our study suggest
that an important group for this consists of migrants from Central and Eastern Europe,
Russia, and former Yugoslavia, as this group shows the strongest dislike for ALMP.
Another interesting topic for further research would be to explore differences between
first- and second generation migrants. Recent work by Constant et al. (2010a,b) provides
evidence that first- and second-generation migrants differ in terms of attitudes towards risk,
language skills, and reservation wages. From this one may expect differences in the response
to a high perceived probability of ALMP participation as well. Presumably, our explanatory
variable capturing years since migration is not able to capture any such generation-specific
differences. Moreover, stratifying by generation as well as by region of origin would result
in subsamples that are too small to allow for meaningful inference. One would therefore
like to have access to a larger sample of migrants from a specific region of origin. Moreover,
following Constant et al. (2010a), one may extend the theoretical job-search framework by
including behavioral-economic concepts.
14
References
Abbring, J.H. and G.J. van den Berg (2003), “The non-parametric identification of treat-
ment effects in duration models”, Econometrica, 71(5), 1491–1517.
Abbring, J.H. and G.J. van den Berg (2005), “Social experiments and instrumental vari-
ables with duration outcomes”, Working paper 2005:11, IFAU, Uppsala.
Aldashev, A., S.L. Thomsen, and T. Walter (2010), “Short-term training programs for
immigrants: Do effects differ from natives and why?”, Discussion paper 10–021, ZEW,
Mannheim.
Algan, Y., C. Dustmann, A. Glitz, and A. Manning (2010), “The economic situation of
first- and second-generation immigrants in France, Germany, and the United Kingdom”,
Economic Journal, 120(542), F4–F30.
Bernhard, S., K. Hohmeyer, E. Jozwiak, S. Koch, T. Kruppe, G. Stephan and J. Wolff
(2008), “Aktive Arbeitsmarktpolitik in Deutschland und ihre Wirkungen”, Research
report 2/2008, IAB, Nuremberg.
Black, D.A., J.A. Smith, M.C. Berger, and B.J. Noel (2003), “Is the threat of reemployment
services more effective than the services themselves? Evidence from random assignment
in the UI system”, American Economic Review, 93(4), 1313–1327.
Blundell, R. and M. Costa Dias (2009), “Alternative approaches to evaluation in empirical
microeconomics”, Journal of Human Resources, 44(3), 565–640.
Borghans, L., A.L. Duckworth, J.J. Heckman, and B. ter Weel (2008), “The Economics
and Psychology of Personality Traits”, Journal of Human Resources, 43(4), 972–1059.
Caliendo, M., A. Falk, L. Kaiser, H. Schneider, A. Uhlendorff, G.J. van den Berg, and K.F.
Zimmermann (2011), “The IZA Evaluation Data Set: Towards evidence-based labour
policy-making”, International Journal of Manpower, forthcoming.
Caliendo, M., and S. Kopeinig (2008), “Some practical guidance for the implementation
of propensity score matching”, Journal of Economic Surveys, 22(1), 31–72.
Caliendo, M., and S. K¨unn (2011), “Start-Up Subsidies for the Unemployed: Long-Term
Evidence and Effect Heterogeneity”, Journal of Public Economics, forthcoming.
Constant, A.F., A. Krause, U. Rinne and K.F. Zimmermann (2010a), “Economic Prefer-
ences and Attitudes of the Unemployed: Are Natives and Second Generation Migrants
Alike?”, IZA Discussion Paper 5380, IZA, Bonn.
Constant, A.F., A. Krause, U. Rinne and K.F. Zimmermann (2010b), “Reservation Wages
of First and Second Generation Migrants”, IZA Discussion Paper 5396, Bonn.
15
De Graaf-Zijl, M., G.J. van den Berg, G., A. Heyma (2011), “Stepping stones for the
unemployed: the effect of temporary jobs on the duration until (regular) work”, Journal
of Population Economics, 24, 107–139.
De Giorgi, G. (2005), “Long term effects of a mandatory multistage program: The New
Deal for Young People in the UK”, Working paper 05/08, UCL, London.
Eichhorst, W. and K.F. Zimmermann (2007), “And Then There Were Four...How Many
(and Which) Measures of Active Labor Market Policy Do We Still Need?”, Applied
Economics Quarterly, 53(3), 243–272.
Huber, M., M. Lechner, C. Wunsch, and T. Walter (2009), “Do German welfare-to-work
programmes reduce welfare and increase work?”, Discussion Paper 4090, IZA, Bonn.
Imbens, G.W. and J.M. Wooldridge (2009), “Recent developments in the econometrics of
program evaluation”, Journal of Economic Literature, 47(1), 5–86.
Lalive, R., J.C. van Ours, and J. Zweim¨uller (2005), “The effect of benefit sanctions on
the duration of unemployment”, Journal of the European Economic Association, 3(6),
1386–1417.
OECD (2010), International Migration Outlook 2010, OECD, Paris.
Rosenbaum, P.R. and D.B. Rubin (1983): “The central role of the propensity score in
observational studies for causal effects”, Biometrika, 70(1), 41–55.
Rosholm, M. and M. Svarer (2008), “The threat effect of active labour market pro-
grammes”, Scandinavian Journal of Economics, 110(2), 385–401
Rotter, J.B. (1966), Generalized expectancies for internal versus external control of rein-
forcement, American Psychological Association, Washington DC.
Van den Berg, G.J. and B. van der Klaauw (2006), “Counseling and monitoring of unem-
ployed workers: Theory and evidence from a controlled social experiment”, International
Economic Review, 47(3), 895–936.
Van den Berg, G.J., A. Bergemann, and M. Caliendo (2009), “The effect of active labor
market programs on not-yet treated unemployed individuals”, Journal of the European
Economic Association, 7(2-3), 606–616.
Van den Berg, G.J., A. Bozio and M. Costa Dias (2008), “Policy discontinuity and duration
outcomes”, Working paper, VU University Amsterdam.
16
Table 1: Selected Sample Descriptives: Baseline Characteristics for Natives
and Migrants
Variables Native Russian Central & Eastern Turkish
European/
Yugoslavian
N 6181 325 432 209
West Germany 0.63 0.93 0.84 0.93
Female 0.50 0.51 0.58 0.46
German citizenship 1.00 0.93 0.74 0.50
Years since migration 0.00 12.99 12.36 9.96
Age 35.95 32.05 36.09 30.33
Age (17-24 years) 0.22 0.36 0.17 0.29
Age (25-34 years) 0.25 0.29 0.33 0.45
Age (35-44 years) 0.28 0.20 0.26 0.20
Age (45-55 years) 0.26 0.15 0.24 0.06
Married (or cohabiting) 0.38 0.54 0.48 0.47
Number of children 0.48 0.73 0.51 0.89
Children
No children 0.69 0.56 0.65 0.49
One child 0.18 0.24 0.21 0.23
Two (or more) children 0.13 0.20 0.13 0.28
Locus of Control (1=external, 7=internal) 5.04 4.87 4.95 4.81
Unemployment benefit recipient (1=yes) 0.79 0.73 0.76 0.77
Level of UB (missings=0) 510.84 424.39 503.87 478.29
Level of UB (log(ben+1),mis=0) 4.73 4.34 4.61 4.67
School leaving degree
None, special needs, other 0.02 0.03 0.03 0.12
Lower secondary school 0.29 0.34 0.31 0.43
Middle secondary school 0.44 0.45 0.33 0.28
Specialized upper secondary school 0.25 0.18 0.32 0.17
Professional Qualification
None 0.08 0.18 0.14 0.32
Internal or external professional training 0.73 0.65 0.65 0.54
Technical college or university degree 0.19 0.17 0.21 0.14
Months in unemployment (div. by age-18) 0.83 0.65 0.71 0.89
Months in employment (div. by age-18) 8.45 6.14 7.89 7.81
Employment status before unemployment
Employed 0.65 0.59 0.69 0.66
Subsidized employment 0.07 0.08 0.07 0.05
School, apprentice, military, etc. 0.15 0.19 0.10 0.14
Maternity leave 0.05 0.06 0.05 0.06
Other 0.08 0.08 0.09 0.09
Source: IZA Evaluation Data Set, own calculations.
Note: All numbers are shares unless stated otherwise.
17
Table 2: Perceived Treatment Entry and Search Intensity of Natives and
Migrants
Variables Native Russian Central & Eastern Turkish
European/
Yugoslavian
N 6181 325 432 209
Subjective (overall) probability of treatment
participation (0=very low, 10=very high)
4.80 5.24 5.13 5.72
(3.58) (3.45) (3.54) (3.39)
Participation probability (η≥5) 0.57 0.63 0.61 0.69
Reservation wage (in euros) 6.89 6.72 7.02 7.35
(2.32) (2.04) (2.25) (2.09)
Number of search channels 5.12 4.92 5.23 4.75
(1.67) (1.61) (1.67) (1.65)
Source: IZA Evaluation Data Set, own calculations.
Note: All numbers are shares unless stated otherwise; standard deviations in parentheses.
18
Table 3: OLS Estimation Results - Reservation Wage and Number of Search Channels
log φ s
Migration background (Ref.: Natives)
Russians 0.004 -.310∗
Central & Eastern European, former YU. -.002 -.021
Turkish 0.1∗∗∗ -.158
Migration background ×participation probability
Natives ×ηH-.028∗∗∗ 0.231∗∗∗
Russians ×ηH-.039 0.505∗∗∗
Central & Eastern European, former YU. ×ηH-.039 0.377∗∗
Turkish ×ηH-.045 -.018
West Germany 0.161∗∗∗ 0.145∗∗∗
Years since migration (divided by age) -.0002 0.135
Female -.123∗∗∗ 0.019
Married (or cohabiting) -.007 0.139∗∗∗
Children (Ref.: No children)
One child 0.025∗∗∗ -.040
Two (or more) children 0.055∗∗∗ -.234∗∗∗
Unemployment benefit recipient (1=yes) -.037∗∗ 0.033
Level of unemployment benefits (log(ben+1),mis=0) 0.01∗∗∗ 0.028∗∗
Age (Ref.: 17-24 years)
Age (25-34 years) 0.094∗∗∗ -.151∗∗
Age (35-44 years) 0.149∗∗∗ -.028
Age (45-55 years) 0.153∗∗∗ -.074
School leaving degree
None, special needs, other (Ref.)
Lower secondary school 0.064∗∗∗ 0.019
Middle secondary school 0.076∗∗∗ 0.161
Specialized upper secondary school 0.154∗∗∗ -.035
Vocational training
None (Ref.)
Int. or ext. prof. training, others 0.067∗∗∗ 0.18∗∗
Technical college or university degree 0.214∗∗∗ 0.33∗∗∗
Months in unemployment (div. by age-18) -.013∗∗∗ -.007
Months in employment (div. by age-18) 0.001∗0.006∗
Personality traits
Locus of Control (1 = Internal) 0.027∗∗∗ -.097∗∗
Openness (standardized) 0.017∗∗∗ 0.065∗∗∗
Conscientiousness (standardized) -.002 0.102∗∗∗
Extraversion (standardized) 0.009∗∗ 0.05∗∗
Neuroticism (standardized) -.012∗∗∗ -.064∗∗∗
Father has A-level qualifications?
Not known (ref.)
Yes 0.04∗∗ 0.136
No 0.017 0.166∗
Father employed at age 15?
Not known or already dead (ref.)
Yes -.008 -.090
No -.0009 0.021
Employment status before Unemployment (Ref.: Employed)
Subsidized employment -.020 0.019
School, apprentice, military, etc. -.046∗∗∗ -.072
Maternity leave -.028∗-.066
Other -.006 -.130∗
Available means of communication:
Landline telephone -.020∗-.235∗∗∗
Personal mobile phone 0.022∗0.077
Computer 0.003 -.040
Printer -.008 0.27∗∗∗
Internet 0.033∗∗ 0.155
Email 0.029∗∗ 0.232∗∗∗
Observations. 7,147 7,147
Pseudo-R20.303 0.062
Note: Additional control variables used in the estimation: Months of entry into unemployment (June 2007 - April 2008),
time between entry and interview (in weeks) and living situation. Full estimation results are available on request by the
authors.
∗ ∗ ∗/∗ ∗/∗indicate significance at the 1%/5%/10%-level.
19
Table 4: Propensity Score Estimation: General Participation Expectation in ALMP
CEER migr.: Natives: High η: CEER Low η: CEER
High vs. Low ηHigh vs. Low ηvs. Natives vs. Natives
(1) (2) (3) (4)
West Germany -.186 0.297∗∗∗ 1.277∗∗∗ 1.728∗∗∗
Female 0.209 0.153∗∗∗ -.053 -.025
Married (or cohabiting) -.336∗-.095 0.799∗∗∗ 1.053∗∗∗
Children (Ref.: No children)
One child 0.087 -.002 0.258∗0.088
Two (or more) children -.026 0.12 0.078 0.032
Years since migration (divided by age) -.030
Unemployment benefit recipient (1=yes) 0.036 0.29∗∗ -.337 -.142
Level of UB (log(ben+1),mis=0) 0.012 0.008 0.021 0.032
Age (Ref.: 17-24 years)
Age (25-34 years) 0.084 -.167∗-.320∗∗ -.475∗∗
Age (35-44 years) 0.209 -.198∗∗ -.911∗∗∗ -1.158∗∗∗
Age (45-55 years) -.2123 -.373∗∗∗ -1.066∗∗∗ -1.236∗∗∗
School leaving degree
None, special needs, other (Ref.)
Lower secondary school -.346 -.101 -.147 -.305
Middle secondary school -.488 -.251 -.050 -.171
Specialized upper secondary school -.359 -.442∗∗ -.016 -.688
Vocational training
None (Ref.)
Int. or ext. prof. training, others -.346 0.023 -.624∗∗∗ -.204
Technical college or university degree -.488 -.347∗∗∗ -.423∗-.157
Months in unemployment (div. by age-18) -.200∗∗∗ -.081∗∗∗ -.162∗∗∗ -.028
Months in employment (div. by age-18) -.0002 -.007 -.045∗∗∗ -.066∗∗∗
Personality traits
Locus of Control (1=Internal) -.212 0.004 -.201∗-.075
Openness (standardized) -.102 0.053∗-.250∗∗∗ -.094
Conscientiousness (standardized) -.076 0.019 0.001 0.093
Extraversion (standardized) 0.105 -.015 0.009 -.093
Neuroticism (standardized) 0.124 -.030 0.185∗∗∗ 0.038
Father has A-level qualifications?
Not known (ref.)
Yes 0.252 -.150 0.997∗∗∗ 0.476
No 0.057 -.033 0.199 -.017
Father employed at age 15?
Not known or already dead (ref.)
Yes -.101 0.071 -.286 0.016
No -.163 0.226 -.126 0.488
Employment status before UE (Ref.: Employed)
Subsidized employment 0.019 -.048 0.131 0.146
School, apprentice, military, etc. 0.05 0.193∗∗ -.555∗∗∗ -.484∗
Maternity leave -.017 0.085 -.570∗∗ -.442
Other -.123 -.110 -.192 -.148
Rent 0.332∗-.070 0.592∗∗∗ 0.193
Subletting -.037 -.011 0.325 0.335
Other 0.01 -.681∗0.465 -.326
Available communication (non-exclusive)
Landline telephone -.391 -.146 0.206 0.391
Personal mobile phone -.030 0.007 -.698∗∗∗ -.625∗∗∗
Computer 0.397 0.08 0.073 0.004
Printer 0.062 -.098 -.277 -.428∗
Internet -.138 0.104 0.246 0.311
Email -.045 -.068 -.498∗∗ -.427
Observations. 757 6181 3980 2954
Pseudo-R20.066 0.032 0.142 0.14
Note: Estimations are done using a logit model. CEER stands for Central and Eastern Europe, Russia and former
Yugoslavia. Additional control variables used: month of entry into unemployment and time between entry and interview.
∗ ∗ ∗/∗ ∗/∗indicate significance at the 1%/5%/10% level.
20
Table 5: Matching Results: Reservation Wage and Number of Search Channels
Comparison Outcome ATET s.e. t-value “Treated” “Untreated” Off support Bias after Median bias
variable matching after matching
CEER migr.: High vs. Low ηlog φ-0.039 0.023 -1.649 466 291 9 2.296 2.098
s0.458 0.133 3.42 466 291 9 2.296 2.098
Natives: High vs. Low ηlog φ-0.030 0.008 -3.758 3516 2665 0 0.627 0.482
s0.220 0.043 5.068 3516 2665 0 0.627 0.482
High η: CEER vs. Natives log φ-0.009 0.018 -0.512 466 3513 1 1.908 1.643
s0.152 0.089 1.695 466 3513 1 1.908 1.643
Low η: CEER vs. Natives logφ0.009 0.021 0.398 291 2663 5 2.041 1.387
s-0.144 0.112 -1.286 291 2663 5 2.041 1.387
Notes: CEER stands for Central and Eastern Europe, Russia and former Yugoslavia. We apply kernel (Epanechnikov)
matching with common support; for the bandwidth we follow Silverman’s rule-of-thumb and use 0.06. Standard errors
are based on 1000 bootstrap replications. Extensive sensitivity analyses are available on request by the authors. Results
are not sensitive to the kernel or bandwidth choice. Estimations are done using the PSMATCH2 package by Leuven and
Sianesi.
Matching quality: we report the mean (median) standardized bias after matching. In addition, we show the number of
individuals in each group (“treated” and “untreated”) and the number of individuals lost due to missing common support
(off support).
21
Figure 1: Propensity Score Distributions for the Different Comparisons
CEER migrants η= 5 −10 vs. CEER migrants η= 0 −4. Natives η= 5 −10 vs. Natives η= 0 −4
CEER migrants η= 5 −10 vs. Natives η= 5 −10. CEER migrants η= 0 −4 vs. Natives η= 0 −4
Note: Propensity score estimation results are in Table 4. Individuals with high participation expectations (η= 5−10)
are depicted in the upper half, individuals with low participation expectations (η= 0 −4) in the lower half. Migrants
from Central and Eastern Europe, Russia and former Yugoslavia (CEER) are depicted in the upper half, natives in
the lower half of the second row.
22
