Content uploaded by Darrel Robinson
Author content
All content in this area was uploaded by Darrel Robinson on Jan 11, 2019
Content may be subject to copyright.
Department of Economics
Working Paper 2018:17
Limited time or secure residence?
A study on the short-term effects of temporary
and permanent residence permits on labour
market participation*
KRISTOFFER JUTVIK and DARREL ROBINSON
Department of Economics Working Paper 2018:17
Uppsala University December 2018
Box 513 ISSN 1653-6975
751 20 Uppsala
Sweden
Limited time or secure residence?
A study on the short-term effects of temporary and permanent residence permits on
labour market participation*
KRISTOFFER JUTVIK and DARREL ROBINSON
Papers in the Working Paper Series are published on internet in PDF formats.
Download from http://www.nek.uu.se or from S-WoPEC http://swopec.hhs.se/uunewp/
Limited time or secure residence?
A study on the short-term effects of temporary and
permanent residence permits on labour market
participation∗
KRISTOFFER JUTVIK†DARREL ROBINSON‡
December 19, 2018
Abstract
In this study we exploit a sudden policy change implemented in Sweden in order
to evaluate the effects of permanent residency on labour market participation. In
short, the policy change implied that Syrians were granted permanent instead of
temporary residency as before the new regulations. Using detailed Swedish reg-
istry data, we examine the effect of the introduction of permanent residency on
three measures of labour market inclusion in the short-term. We analyze the data
through a simple difference-in-means as well as through comparison to groups un-
affected by the policy in a difference-in-differences design and a synthetic control
group approach. Our conclusions are twofold. On the one hand, we conclude that
temporary residents that are subject to a relatively less-inclusive situation earn
more and are unemployed less. However, at the same time, they are less likely to
spend time in education than are those with permanent residency.
Keywords: Labor Market Inclusion, Asylum Policy, Sweden, Residence Permits
JEL Classifications: J15; J24; O15
∗We are grateful to Emma Holmqvist, Susanne Urban, Henrik Andersson, Per Adman, Gunnar Myr-
berg and seminar participants at the Institute for Housing and Urban Research and Uppsala Center for
Labour Studies for helpful comments and suggestions on earlier versions of this text. We are also grateful
to Camilla Scheinert for help with parts of the graphical work.
†Institute for Housing and Urban Research and Department of Government, Uppsala University. mail:
kristoffer.jutvik@ibf.uu.se.
‡Department of Government, Uppsala University, mail: darrel.robinson@statsvet.uu.se
1
1 Introduction
Humanitarian crises throughout the Middle East, South Asia, and Africa have greatly
increased immigrant inflows to Europe in recent years which has subsequently impacted
upon the politics of European recipient nations. Anti-immigration parties have grown
– largely on the basis of ethnocentric and nationalistic rhetoric – and a large debate
has arisen as how best to integrate those individuals that are eligible for asylum in host
countries into the labour market and wider society.
In recent decades, there has been a shift in migration politics across European coun-
tries. This shift has frequently been labeled as a convergence towards increasingly assim-
ilatory politics (Joppke, 2004, 2017) and a retrenchment of multiculturalism (Vertovec
et al., 2010). From a theoretical standpoint it has been noted that the shift has de-
creased emphasis on individual rights for a greater insistence on individual responsibili-
ties (Borevi, 2010). In short, responsibilities-based models argue that individuals need to
make an effort to belong, whereas rights-based models perceive individual rights as the
driver of inclusion. Of particular focus has been residency status. Responsibilities-based
models argue that migrants should be given short-term permits subject to re-evaluation
for which only those that succeed in integrating into the labour market should be given
right to stay. Rights-based models on the other hand argue that migrants should be
given permanent residency, and that this right to stay should not be conditional on
labour market inclusion. This inclusion is in turn intended to improve migrants’ possibil-
ities to integrate into the labour market and wider society. However, the implications of
residency status on labour market inclusion are largely unknown. The research question
this paper addresses is thus: What is the effect of permanent residency status on labour
market inclusion?
Looking at the current state of literature, there are a number of works that study
mandatory integration which base residency or citizenship on language skills, norm adher-
ence, and cultural-historical knowledge shown by the migrant (Strik et al., 2010; Goodman
and Wright, 2015).1Other lines of research focus on the impact of residency status on
non-labour outcomes such as health or psychological well-being (Ryan et al., 2008; Bogic
et al., 2015; Bakker et al., 2014), or have compared labour market outcomes across dif-
ferent nation-states (Koopmans, 2010; Kogan, 2007, 2006; Mansouri et al., 2010; Kesler,
2006; Bevelander and Pendakur, 2014). Although providing valuable insights, compar-
isons between nation-states are problematic as the included cases may differ along other
dimensions, for instance, internal migrant populations, institutional setups, pre-existing
1Mandatory integration policies demand acquisition of citizen-like skills, such as host country lan-
guage, norms, culture, or history, in order to become a full member. See Goodman and Wright (2015)
for a detailed description.
2
immigrant networks, or reception policies, all which may obstruct causal inference. Two
recent contributions study the effect of residency permit explicitly, but have led to mixed
conclusions (Larsen et al., 2018; Blomqvist et al., 2018). Furthermore, these studies in-
clude groups that largely consist of refugee-status migrants which limits the scope of
the treatment – residency permit – as the factors that define refugee-status are largely
time-invariant.2
With this in mind, this paper takes a somewhat different approach by exploiting a
swift change of policy implemented in Sweden concerning asylum seekers from Syria.
Before implementation of the policy change, most Syrian asylum seekers that came to
Sweden were given temporary residency. In 2013 the Swedish Migration Agency (SMA
henceforth) abruptly re-assessed their evaluation of the Syrian conflict which had as a
consequence that all asylum seekers from Syria were to be given permanent rather than
temporary residency.3As the policy change was implemented without prior warning,
it provides a threshold between Syrian refugees that applied for residency under two
different regulations. We exploit this threshold as a quasi-experiment to study the effect
of permanent residency on labour market inclusion in the short term.
This study brings a few additions to the existing literature. First and foremost, it
explicitly focuses on the effects of residency status on labour market inclusion. Until this
point, there is little knowledge of the specific impact of temporary and permanent set-
tlement on these outcomes in existing studies. This is of interest because labour market
outcomes are the most commonly debated in policy and theory in which both models,
the rights-based and responsibilities-based, are claimed to be superior. Second, we make
use of detailed individual data containing the full population under scrutiny. The dataset
allows us to sort between different categories of migrants and identify refugees, as opposed
to labour migrants, family migrants and students. Opposite to comparative approaches,
this study investigates the effect of temporary and permanent residence permits within
one nation thus holding confounding variables such as institutional, cultural, and histor-
ical variation constant. Lastly, while this study only focuses on the short-term inclusion
into the labour market, we do so with estimates of outcomes after two different lengths
of time in Sweden. The importance of early settlement have been emphasized in previous
2These papers likely suffer from difficulties in estimating a treatment effect because the incentive
mechanism does not manifest. If protection status is based upon one of the largely non time-variant
criteria that define refugees according to the Geneva convention, for example persecution due to gender,
sexual orientation, or ethnicity among others, the granting of temporary residency should not generate
the same incentives to integrate into the labour market; upon re-evaluation of one’s case for renewed
residency, the factors that led to refugee status are still present. Our sample on the other hand consist
almost entirely of individuals granted subsidiary protection status due to the Syrian conflict itself, not
the above mentioned criteria (see Table 1).
3Note that the recognition rate for Syrian asylum seekers was 100 % before and after the policy
change.
3
studies and this approach allows us to identify at which point observed outcomes as the
result of residency permit manifest (OECD, 2016a,b).
The conclusions of this study are two-fold. We find that temporary residents have
higher incomes and are unemployed less. However, at the same time, they are less likely to
spend time in education than are those with permanent residency. Given these findings
it is clear that less secure residence status is beneficial to labour market inclusion in
the short term. However, the greater focus on education among those with permanent
residency raises the possibility that long-term inclusion may not follow the same patterns.
Our results therefore suggest that both approaches to migration policy can be supported
empirically, albeit with different metrics for success. In our view then, the issue should
be viewed as largely normative. Rather than debate whether one approach will lead to
greater inclusion than the other, focus should be shifted to discussing the type of inclusion
that the different approaches are likely to provide.
2 A Theoretical Framework on Security of Residence
We propose a theoretical model, inspired by Borevi (2002) and Koopmans (2005), in
which residency status is defined as a uni-dimensional scale ranging between different
levels of security. The position at the left-side of the continuum, insecure residency,
refers to a position in which individuals are largely excluded from full membership and
residency status relies exclusively on attachment to a specific job or studies. If the
required attachment is lost, then the basis for residency is immediately withdrawn. Hence,
this position refers to an unpredictable and insecure status in which the individual has
to perform in a pre-stipulated manner in order to maintain residency. At the right end of
the continuum, secure residency refers to a position in which full membership is granted
into a community of citizens. In this position, residency cannot be withdrawn and hence
refers to an increasingly predictable and secure type of residency.
Residency
status
Guest workers Temporary residents Permanent residents Citizens
Insecure residence Secure residence
Figure 1: Theoretical model on security of residence
Notes: Figure displays the continuum in residency status ranging from insecure residency to
secure residency.
In our theoretical model, we argue that temporary and permanent residency are found
in between the above extremes. This is visualized in Figure 1. Temporary residency is
4
found closer to the left, insecure, end of the scale because an individual’s status will
be re-assessed after a pre-determined period of time. In the event that the grounds
for residency change, and one lacks attachment to the host society such as through
employment, residence status is not typically renewed upon re-assessment. However,
temporary residents do not rely on a specific attachment during their limited time of
residence. On the other hand, permanent residence is found on the right-side of the scale.
Individuals granted permanent residency are not yet fully included in the membership of
citizens but they benefit from secure residency that is not subject to re-assessment after
a given time period, and thereby is not conditional on employment or study.
3 Residency Status and the Labor Market
Human capital theory stipulates that an individual’s earnings are determined as a func-
tion of work experience and skills such as those developed in education (Mincer, 1974;
Chiswick, 1978; Dahlstedt and Bevelander, 2010). If individuals can bear the cost of
education in the short-term, as well as the indirect costs of loss of income and less expe-
rience, long term benefits will lead one to prefer education. This model has been highly
influential in understanding why individuals with higher education have higher earnings
on average. The human capital framework can be applied to explaine migrant earnings
as well with respect to education in a recipient country. Migrants that study in their new
country, whether it be language training (Chiswick, 1991; Lemaˆıtre and Liebig, 2007;
Ferrer et al., 2006; Delander et al., 2005), higher education (Nekby et al., 2002; Ham-
marstedt, 2003; Ald´en and Hammarstedt, 2014), or a specific labour market certificate
(Dahlstedt and Bevelander, 2010) are consistently found to have higher incomes in the
long term than those migrants that do not study upon arrival to their new country as
they develop skills adapted to the local labor market.
For those granted permanent residency renewal of residency status is guaranteed and
the short-term long-term trade-off as outlined in the human capital function remains
unchanged. However, an insecure residency status has the possibility to influence this
model by altering the short-term long-term trade-off. The literature on civic integra-
tion suggests that residence status should depend on performance such as labour market
attachment. Migrants that fail to accommodate the desired requirements are refused re-
newal of residency (Bee and Pachi, 2014; Borevi, 2010; Koopmans, 2010). This incentive,
or restriction, means that individuals that aim to remain in their recipient country must
discount long-term earnings that may come from education in favour of short-term labour
market attachment that will ensure residency renewal. In line with this argumentation,
we propose the following hypotheses:
5
•H1: Permanent residency should lead individuals to work less than temporary res-
idents in the short term.
•H2: Permanent residency should lead individuals to study more than temporary
residents in the short term.
4 The Institutional Setting
After the outbreak of the conflict in Syria the Swedish Migration Agency (SMA) crafted a
number of internal documents containing guidelines and descriptions of the development
in the conflict. These documents, referred to as RCI (Instructions from the General
Counsel),4served to guide case workers in the assessment of the mounting number of
applications from Syria. Without going into the specifics of these documents,5the general
guideline prior to the 2013 change which we exploit was that Syrian asylum seekers were
to be granted temporary residence in Sweden, allowing three years of settlement (RCI
14/2012, 2012). However, permanent residency could be granted if individuals were
considered to be convention refugees.6In 2012 the share of temporary residence permits
was 61% of all granted Syrian applications. This figure had risen somewhat in 2013 in
which the share of temporary residence permits was roughly 73% of all granted Syrian
applications.
On September 3, 2013 the SMA made a new evaluation of the conflict in Syria (RCI
14/2013, 2013) which resulted in the policy change of focus in this study. The General
Counsel stated that the conflict was in a dead lock position, in which both sides believed
close victory was possible. The SMA also noted that the number of actors participating in
the conflict had increased. As a consequence of the increased complexity of the conflict,
the General Counsel made the judgment that the unrest in Syria would go on for an
extensive period of time and stated that all Syrian asylum seekers should be granted
permanent residence. Hence, after the policy change, 100% of applications were granted
permanent residence permits even if not considered convention refugees.
There are a few important details about the policy change that have significant impli-
cations for the choice and implementation of our research design. First, the change was
implemented immediately, providing us with a cut off between those awarded temporary
and permanent residence permits. As shown in Figure 2, there is a clear jump at the
4”R¨attschefens instruktioner” in Swedish.
5For a more detailed description of the the guidelines provided by the SMA, see Andersson and Jutvik
(2018).
6In accordance with the Geneva Convention, a person is a refugee if he or she has a well-founded
fear of persecution due to race, nationality, religious or political beliefs, gender, sexual orientation or
affiliation to a particular social group.
6
0 20 40 60 80 100
Share of permanent residence permits (%)
2012m9
2013m2
2013m7
2013m9
2013m12
2014m5
2014m10
Month
Figure 2: The share of permanent residence permits among all granted residence permits
(among Syrian applications)
Notes: Share of permanent residence permits among all granted residence permits. Data concerns
distribution among Syrian applications only. Note that the recognition rate was 100% before and
after the policy change.
Source: The Swedish Migration Agency (2018).
implementation of the new directives, where the share of permanent residence permits
sharply rises from about 35% to 100%. Second, the change of directives were implemented
by the SMA without prior announcement. Hence, it was not a political decision or the
result of a long parliamentary debate – no awareness was made of an impending change
in the period prior to the reform. This setup made it impossible to react to the change
before the actual implementation. The sudden implementation, in combination with
detailed data from the SMA, allows us to identity a treatment group unaffected by any
potential sorting – those individuals that applied before, but were granted residency after,
the change in policy. Lastly, the policy change implied that all individuals that had been
granted temporary residence permits prior the change could apply for a re-evaluation of
their permits in order to make them permanent. The majority of Syrians with temporary
permits also applied for the re-evaluation immediately. These applications for extensions
were largely processed by the SMA before the end of the year.7
4.1 The Introduction Program
The introduction program is available to all newly arrived migrants during their first
two years of residence in Sweden. The program contains language training (SFI) and
civic orientation, but also the development of an individual plan aiming to fasten labour
7Data from the SMA indicate that 99% of all applications for re-evaluation were handed in to the
SMA before the end of 2013. The SMA had processed 77% of these applications in the end of 2013 and
91% in the end of January 2014.
7
market introduction (Larsson, 2015). Although the program is not mandatory, those who
choose not to participate do not receive any attached economic benefits. Hence, there is a
strong incentive to participate. The economic support provided in the program is slightly
higher than the general social assistance available to all Swedes and is not affected by
the income of other household members. Participants are also allowed to work during
the program (OECD, 2016a).
Given the structure of the program, in which benefits are dependent on participation,
we assume that the individuals under scrutiny here were part of the program during the
time of investigation. The program gives several alternatives for participants in terms
of language training, preparation for work, and educational activities in combination
with activity on the labour market (OECD, 2016a). In that manner, the program, in
combination with the cut off of the policy change, allows us to to track individual behavior
during the extension of the program as well as attachment to the labor market in the
treatment and control group.
5 Methodological Setup, Data and Samples
This study relies on a detailed database of individual-level Swedish register data. The
database, GEOSWEDEN, contains anonymous information on all residents with a reg-
istered address in Sweden between 1990 and 2014. Assessing this data-set we obtain
information on, among many other things, individual demographics, labor market status,
education, country of birth, month of granted residence permit, and reason for approved
application (grund f¨or bos¨attning).
In our methodological setup, we make use of the sharp introduction of SMA’s policy
change which provides a cut-off point determining residency status. Because the decision
from the SMA was so sudden and without prior indications of an impending change, those
that were approved for asylum after, but who applied before, September 3, 2013 were
entirely unaware that they would be guaranteed to receive permanent residency upon
approval. As such, their applications would have been made with the knowledge that
the majority of successful asylum applicants received temporary status. This group of
individuals, those that applied before but for whom a decision was taken after the change
in policy, make up our treatment group.
We do not observe the date of application and date of decision at the individual-
level in GEOSWEDEN. Rather, our data allows us to see only the month a residency
decision was granted. This creates two potential problems in defining the start and end
of our treatment group, however, we circumvent this issue in two ways. First, in defining
the start of our treatment group we rely on the fact that the change was made at the
8
beginning of the month, on September 3, 2013. That means that effectively all individuals
for whom a decision was made in the month of September were subject to the new rules.8
Inferring application outcomes based on monthly-level data is therefore not problematic
and allows us to overcome the issue of treatment “start” in the absence of daily data.
The second issue is that we cannot define the “end” of our treatment period at the
individual-level. That is, we know that all individuals that were awarded residency after
September 3, 2013 were given permanent residence, but if we include individuals into
our treatment group that applied for residency after September 3 our sample will have
self-selected into treatment. In order to solve this problem, we make use of data from
the SMA containing anonymous individual-level application and decision dates (though
we are unable to connect this to individuals in the GEOSWEDEN database through lack
of identifying information). The data from the SMA shows us that 96.38% of all Syrian
asylum decisions made in September 2013 were based on applications that had been
submitted prior to the September 3 threshold. This proportion decreases quite rapidly in
the following months; only 57.6% of applications granted in October were based on pre-
reform submissions, and this drops further to 23.13% in November. As such, we can with
confidence define our treatment window as all of those individuals for whom an asylum
decision was granted in September 2013. After removing those below the age of 18 and
over 65 from our sample, we obtain a treatment group comprised of 629 individuals.
We define our control group as all individuals from Syria that were granted residency in
September of 2012, exactly one year prior to our treatment group. Such a definition allows
us to hold constant the amount of time individuals had been in Sweden across treatment
and control groups when we measure our dependent variables. Further, migration is
highly seasonal so the selection of migrants based on granted asylum month should further
ensure comparability. The most natural control group would have been to select those
individuals that were granted residency in the period immediately prior to the change in
policy directive, for example those that were granted residency in August 2013. However,
the provision which allowed all of those individuals who had been previously been granted
temporary residence to apply for a re-evaluation, prevents the use of such a strategy.
Comparing labour market indicators over different years can naturally lead to prob-
lems if economic conditions differ year-to-year. However, for our two years of interest we
8September 3 was a Tuesday which means that only individuals whose decisions were made on Monday,
September 2 would have been subject to the old rule, all others who were decided in September were
subject to the new rule. There were 927 decisions made in the entire month of September which equates
to 44 per working day on average. However, the length of time for a decision decreased drastically after
the rule change (there were on average 37 decisions per day in July 2013 and 33 per day in August 2013)
which would indicate that more decisions were made per day after the rule change than before, and that
the number of decisions made in September 2013 but before the reform is likely lower than this daily
average.
9
see very little change in economic indicators. GDP growth was 1.2% in 2013 and 2.6%
(World Bank) in 2014 and foreign-born unemployment was exactly equal at 16.4% in
both years (OECD). In general, the labour market conditions for our two groups were
largely equal, if not slightly beneficial to the 2013 cohort of permanent residents. How-
ever, it is nevertheless possible that the labour market for new arrivals specifically differs
year-to-year. We therefore leverage the labour-market outcomes of all other non-Syrian
asylum seekers as a comparison group. These approaches will be further described below.
Table 1: Individual characteristic-differences between
asylum seekers arriving 2012 and 2013
Year 2012 2013 Difference
Population (September): N=274 N=629
Men 0,65 0,62 -0,03
Age 36,18 34,89 -1,29
Young 0,49 0,51 -0,02
Middle age 0,40 0,40 0,00
Married 0,49 0,56 0,07**
With children 0,36 0,49 0,13***
With university education 0,20 0,36 0,16***
Protection status (%):
Subsidiary protection 0,82 0,85 0,03
Convention refugees 0,18 0,15 -0,03
Place of residence (%):
Metropolitan cities 0,23 0,14 -0,09***
Stockholm 0,12 0,06 -0,06***
Outcome variables:
Unemployment days 37,10 44,99 7,88***
Declared income 73,68 8,37 -65,30***
Study grants 0,50 2,69 2,180***
Notes: Table display differences in characteristics among those arriving
in September 2012 compared to those arriving in September 2013. The
variables are measured in the end of the year that the residence permit
was granted.
Source: GEOSWEDEN (2018) and the Swedish Migration Agency (2018).
5.1 Data and Sample
The data for our primary analyses are taken from all asylum seekers in the years 2012-
20149which is the height of Syrian migration to Sweden. In Table 1 we present some
descriptive statistics of Syrian asylum seekers for whom residency was granted in Septem-
ber 2012 and September 2013. As seen in the table, gender and the age structure is
relatively stable. However, moving on to the share of married individuals and individuals
with children reveals an increase by 7 and 13 percentage points respectively. In addition
92014 is the latest wave of data available to us.
10
to that, there is also a substantial increase in individuals with university education.10
The focus of this paper is labour market inclusion which motivates the selection of our
dependent variables. We use three measures, number of days registered as unemployed,
amount of study support received (in 100s of Swedish Kronor), and declared income (in
100s of Swedish Kronor). These are all measured as totals for the calendar year.
A primary assumption that our research design relies upon is the comparability of
our control and treatment cohorts. As can be seen in the Table 1, 2012 arrivals are
largely similar to the 2013 arrivals in terms of gender, age and the proportion that are
considered convention refugees. As mentioned, there are some differences. If those that
make up the 2012 cohort have baseline characteristics that make them more likely to
succeed on the labour market than the 2013 cohort, our results could be biased. Level of
education is precisely such a potential factor. However, we find here that the proportion
of university educated is higher among our 2013 cohort than the 2012; in other words,
our treatment group is more highly educated than the control group. As such, based on
the descriptive differences we observe between cohorts, this sample provides a hard test
of our hypotheses that permanent residents should work less and study more in the short
term than temporary residents.
6 Results
6.1 Baseline results - did the change in residency status have
any effect on inclusion?
In Table 2 we present the estimated effects of residency status in a direct comparison
of permanent and temporary residents that arrived in September 2013 versus September
2012 respectively. All outcomes are measured in December of the year after arrival. For
each of the three dependent variables we estimate the effect of residency status with and
without demographic controls. Perm Res is the coefficient of interest which represents the
estimated difference between those with permanent (coded as 1) and temporary residency
(coded as 0). Because this variable is dichotomous, we can interpret it as a difference in
means of the dependent variable for the two residency status groups (conditional on the
demographic covariates in the models for which they are included).
Models 1 and 2 show the estimated effect of residency status on total number of
registered unemployment days in an individual’s first 16 months in Sweden. Based on
the Model 1 estimate, those with permanent residence were registered as unemployed for
10The data regarding education is somewhat problematic since validation of foreign education might
differ over cases. We control for these differences in our estimations but acknowledge that the differences
between the groups might be smaller.
11
Table 2: Baseline results: Effects of residence status
Unemployed Days Study Grants Declared Income
(1) (2) (3) (4) (5) (6)
Basic W. controls Basic W. controls Basic W. controls
Perm Res 23.52*** 17.94** 6.447** 8.357*** -194.1*** -158.5***
(8.449) (8.439) (2.896) (3.002) (47.56) (44.34)
N 903 903 903 903 903 903
Notes: Table displays the estimated effects of residency status. Perm Res shows the comparison between
the 2012 and 2013 cohort after 16 months with a residence permit in Sweden. Controls include age, marital
and parental status, education, and gender. Robust standard errors in parentheses *** p<0.01, ** p<0.05, *
p<0.1
Source: GEOSWEDEN (2018)
24 more days on average than those that arrived in September 2012. When we include
demographic controls the coefficient changes only marginally indicating that those with
permanent residence permits registered 18 more unemployment days.
The variable Study Grant is measured in 100s of SEK (roughly equivalent to $12
USD) and as such the coefficient on Perm Res shows that permanent residents took out
on average roughly 650 SEK more in their first year of residency. This form of financial
assistance is only available to those that are enrolled in a study program so the fact
that temporary residents received less study grants than permanent residents indicates
that they generally spent less time in education. The 2012 September cohort of Syrians
received on average 644 SEK in study grant compared to 1288 among the 2013 September
cohort yielding a difference of 644 SEK, equivalent to the -6.44 point estimate in the basic
model. Such a figure is low in absolute terms, but represents a doubling of expected study
grants among the permanent resident group. When controls are included the magnitude
of the coefficient increases to 8.4 (840 SEK).
Lastly, the estimated effect of residency status on income is such that individuals with
temporary residency, our control group, declared on average 19,400 SEK more after their
first year than did those with permanent residency. The inclusion of demographic controls
reduces this estimate somewhat, but even so the estimate is an increase in declared income
of 15,900 SEK among temporary residents.
In order to further test the robustness of our baseline results, we extend the time pe-
riod of comparison, increasingly adding one month at the time, until December. Although
this test gives larger samples, it also brings problems with sorting as a larger share of indi-
viduals applied after the policy change was instigated. However, the estimates are rather
stable as the time period is extended. We also perform the same analysis but instead of
comparing Syrians, we compare all other newly arrived migrants granted residence in the
same time period. Estimates with this placebo group yield largely insignificant estimates.
More specifically, the estimates regarding unemployment days as well as declared income
12
are insignificant. There is, however, an effect on the usage of study grants in the placebo
group although the estimates negative, opposite of our main test. Lastly, we conduct
estimates with a matched sample derived from propensity score matching. These tests
gives further support to our analysis regarding unemployment days and declared income
but yields insignificant, but positive, estimates considering the usage of study grants.11
6.2 Difference-in-differences
While the above comparison is intuitive, it is potentially invalidated by year-on-year
trends in the outcome variables. In order to wash such potential trends we utilize a
difference-in-differences approach. Effectively, the method sets out to compare the trends
in labour market inclusion observed cohort-to-cohort among Syrian asylum seekers with
that of an assumed comparable control group. The difference in post-policy outcomes of
these two groups is compared to the difference in pre-policy outcomes thereby providing
an estimated effect of treatment that is not confounded by time.
In our first specification we make use of non-Syrian asylum seekers that arrived in the
same period as our Syrian sample.12 That is, we compare the 2012 Syrian cohort to the
2012 cohort of all other asylum seekers to Sweden, and the 2013 Syrian cohort to the 2013
cohort of asylum seekers. We estimate the year-on-year trend as identified by the Year
term in Table 3 which takes on the values of 1 for the 2013 cohort (outcomes measured
in 2014) and 0 for the 2012 cohort (outcomes measured in 2013). Further, we include a
dummy variable which indicates if the observation belongs to the Syrian sample (1) or if
they belong to the group of other countries (0). Lastly, we interact these two terms to
obtain our difference-in-differences estimate. If the cohort trend in the outcome variable
differs for the Syrian sample in comparison to the group of other arrivals, we assume that
this difference is due to the change in SMA directives, and thus, is the result of permanent
residency. We also include the same pre-treatment controls as above. As can be seen
in Table 3, the estimates produced from the difference-in-differences specification are
substantially larger in magnitude compared to the experimental setup for for all included
measures. Having said that, they mirror the base-line results presented above. Further,
statistical significance at the 95% confidence interval is achieved in all 6 models.
There is a clear increase in the usage of unemployment days in the Syrian cohort after
the introduction of permanent residency. The amount of declared income, on the other
hand, decreases after the introduction of permanent residency. Lastly, the usage of study
11For robustness regarding extended time periods and propensity score matching, see Appendix.
12We have restricted our sample of other newly arrived migrants (non-Syrian) to exclude those migrat-
ing from “uncommon” sending-countries with less than 35 individuals over the time period. This is done
to avoid smaller groups of migrants to get significant weights in the synthetic control group approach.
The sample consist of 19 countries/geographical areas.
13
Table 3: Difference-in-differences specification - Other new arrivals
Unemployed Days Study Grants Declared Income
(1) (2) (3) (4) (5) (6)
Basic W. controls Basic W. controls Basic W. controls
Year -12.45* -14.17** -5.347 -4.949 36.28 24.39
(7.12) (6.94) (3.45) (3.19) (31.03) (30.63)
Syrian -4.49 -11.83 -17.94*** -11.99*** 259.07*** 235.25***
(8.89) (8.72) (4.31) (4.01) (38.73) (38.50)
Year x Syrian 35.97*** 32.91*** 11.79** 10.93** -230.37*** -204.27***
(11.17) (10.93) (5.42) (5.03) (48.67) (48.25)
Constant 138.37*** 81.55*** 24.39*** 71.64*** 131.55*** 59.55
(5.24) (9.80) (2.54) (4.51) (22.83) (43.25)
Observations 2,025 2,025 2,025 2,025 2,025 2,025
R20.01 0.05 0.01 0.16 0.02 0.06
Notes: Table displays the estimated effects of residency status. The coefficient of interest is the interaction term (Year x
Syrian). Controls include age, marital and parental status, education, and gender. Robust standard errors in parentheses
*** p<0.01, ** p<0.05, * p<0.1
Source: GEOSWEDEN (2018)
grants increases in the 2013 cohort, having permanent residency.
6.3 Synthetic control group
A difference-in-differences model provides unbiased causal estimates under the assump-
tion that the treated group would have followed the same trend as the comparison group
in the absence of treatment (Angrist and Pischke, 2008). However, in the event that
this assumption does not hold, resulting estimates are potentially invalid. As such, we
further implement a synthetic control group approach which allows one to empirically
define the comparison group in a generalized difference-in-differences approach (Abadie
et al., 2010).
This approach has two distinct advantages which contribute to the overall aim of the
paper. First, it provides an empirically chosen comparison group that perfectly matches
the trend in labour market outcomes for Syrian asylum seekers in the pre-treatment
period. As such, we have a stronger claim that the common trend assumption is valid.
Second, because we must include a second pre-treatment cohort, we must adjust the
measurement period for our dependent variables such that outcomes are measured in the
same year as the migrant arrived in Sweden rather than one year later. This provides
the opportunity to identify not only if there is a shift in expected outcomes as the result
of the change in SMA directive, but also to determine if such differences manifest at the
14
earliest stages of a migrants time in Sweden, or if they are only visible later.
We use three cohorts as the basis for our synthetic method estimates, those granted
residency in 2012, 2013, and 2014. However, we restrict our sample to only those that
arrived in August or prior in each given year because the policy directive changed in
September 2013. This ensures that both the 2012 and 2013 cohorts can be viewed as
fully untreated, and the 2014 cohort as fully treated. A second potential limitation to
this estimation strategy is related to data availability. 2014 is the last year of data
available which means that we must measure the outcomes in the same year in which
one was granted asylum. As such, results should be viewed as the estimated effect of
residency status in the very short term.
The synthetic control method is appropriate when treatment is assigned at the aggre-
gate level(Abadie et al., 2010, 2015; Fowler, 2013). In our case, the treatment assignment
is the assignment of permanent residency which is applied to the group of Syrian asylum
seekers from September 2013 and onwards. Since the change in process affected Syrian
migrants only, the pool of migrants from other countries can be seen as potential control
units which can act as counterfactual cases. Given a set of pre-intervention outcomes,
in our case the average unemployment measured in 2012 and 2013, the synthetic con-
trol method seeks to find a set of weights, w, that can be applied to the donor pool
of potential control cases such that the outcome measures in the pre-treatment period
between the synthetic control and observed treatment group are equal. In other words,
the method finds an weighted combination of control units such that the common trend
assumption is satisfied in the pre-treatment period (we still must make the assumption
that it is satisfied in the post-treatment period, which is inherently untestable). In effect,
the method finds a control group such that:
¯
Yk
t=
j
X
c=1
∗wc∗¯
Yk
c(1)
where Y is the variable of interest, subscripts t and c denotes the treatment and control
cases, j is the number of control units, and superscript k indexes the pre-treatment
measurements of Y.
The donor pool of potential control cases consists of 19 non-EU units (SMA desig-
nated geographical regions – primarily countries but in some instances groups of smaller
countries from a distinct geographic area) – from which at a minimum 30 individuals
were awarded residence permits on humanitarian grounds.13 The synthetic control units
differ for each of our three outcome variables, but in all but one analysis (unemployment
13These are the same 19 geographical regions that make up the comparison group in the standard
difference-in-differences estimates discussed.
15
days), every possible donor unit is given positive weight. However, in all three cases the
empirical control is largely dominated by two units, with the remaining units receiving
marginal weights. See the Appendix for the actual weightings.
Figure 3: Trends in unemployment days (a), declared income (b), and study grants (c)
for Syrians and the synthetic control unit
Syria
Synthetic Control
Policy change September 2013
0 50 100 150
Unemplyment benefits (average)
2012 2013 2014
Year
treated unit synthetic control unit
(a) Unemployment days
Synthetic Control
Syria
Policy change September 2013
0 50 100 150 200
Declared income (average)
2012 2013 2014
Year
treated unit synthetic control unit
(b) Declared income
Synthetic Control
Syria
Policy change September 2013
0246810
Study Grants (average)
2012 2013 2014
Year
treated unit synthetic control unit
(c) Study Grants
Notes: The figures presents the level of unemployment days (a), declared income (b), and study
grants (c) among Syrians and the synthetic control unit across three points in time. The synthetic
control unit consist of a weighted average of all other individuals granted residence permits on
humanitarian grounds in Sweden.
Source: GEOSWEDEN (2018).
Given that the algorithm was successful in finding a pre-treatment match for Syrian
migrants, we can examine the difference in the trends post-treatment in order to draw
inference of the effect of the reform. Figure 3 (a) shows that the level of unemployment
days is lower in the synthetic control unit after the introduction of permanent residency.
Effectively, Syrian migrants in 2014 on average were registered as unemployed for 16 more
days in 2014 than the weighted average of those individuals that make up the control
unit. The results for declared income are similarly perfectly consistent with our main
estimates; the synthetic control group is expected to have declared 7100 SEK more in
2014 than did Syrian migrants. The short time frame of these estimates must be kept in
mind. These estimates are measured at the end of 2014 for individuals that were granted
16
residency in 2014. Given such a short time period, and given that the average income for
the 2014 cohort of Syrian migrants in 2014 was approximately 54,100 SEK, 7,100 SEK is
a non-trivial estimate. The final outcome variable in our study is the use of study grants,
defined as the amount of study grants in SEK that migrants received in a given year.
Here we find that Syrian migrants received on average more study grants than did their
counterparts in the synthetic control group. On average, a Syrian is expected to have
secured 200 SEK more at the end of their first year in Sweden than was observed in the
synthetic control group.
Taken together, the estimates above provide a rather coherent picture, and one which
is highly consistent with our main investigation. In addition, we perform randomization
inference on the other 19 countries in the data-set. For each dependent variable we per-
form the same analysis, while excluding Syrians. Regarding unemployment days, only
three other groups provide a larger increase than the Syrian sample; in the randomization
inference regarding declared income four other groups provideslarger negative estimates;
lastly, regarding study grants, six other populations show similar increases. Therefore,
although connected with some uncertainties, the results consistency of the synthetic con-
trol tests with our other estimation techniques increases our overall confidence in the
effect of permanent residency. See the Appendix for the randomization inference results.
7 Discussion and Concluding Remarks
A number of scholars have pointed to a general trend of convergence towards increasingly
restrictive migration policies and a multicultural backlash across European nation-states
(Joppke, 2007; Vertovec et al., 2010). Similarly, political parties and scholars have empha-
sized that individuals that need to make an effort to remain in the country of destination
in order to have a better chance to integrate into the main society. A few studies have
evaluated the effect of civic integration measures. However, few of these studies explicitly
focus on residency status and its effect on labour market inclusion.
In this study, we have attempted to fill this knowledge gap by exploiting a policy
change concerning the residency status of Syrian asylum seekers in Sweden. We examine
the policy change with respect to labour market inclusion defined as unemployment days,
study grant receipt, and declared income. A few problematic aspects of the policy change
in combination with the difficulty of estimating labor market outcomes led us to perform
several analyses; a difference-in-means approach as well as a difference-in-differences de-
sign and a synthetic control group approach. Our estimation techniques produce very
similar patterns which give greater confidence in our overall results than can be derived
from a single method. Further, the methods provide us with estimates at different time
17
points - outcomes in the cohort study are measure 16 months after residency and out-
comes in the synthetic control models are measured 4 to 8 months after granted residency.
This helps us to understand not only if, but when differences become apparent in the
short term.
We argue that the shift in directives by the SMA represented a fundamental change
in the security of residence for newly arrived Syrians. We argue that a shift to more
secure residency represents a shift towards a more rights-based approach to migration,
which is in itself derived from the normative standpoint of multicultural migration policy.
Consequently, this shift that the SMA implemented represents movement away from a
responsibilities-based approach to migration which is typically grounded in theories of
civic integration. Migration policy in its entirety is defined by many more features than
this single dimension, and we are careful to avoid any claim to the study of different
migration policy regimes. Rather than a weakness, this continuity in migration policy is
a strength of this study as it allows us to identify an individual dimension such that all
other relevant institutional and cultural factors of the recipient country, Sweden, remain
constant. Such an opportunity allows us to estimate the effect of differing levels of
security of residency within a single institutional context, rather than deriving inference
from the comparison of many contexts, in which all factors that define migration policy
are bundled.
In general we find that temporary residents perform significantly better than per-
manent residents with respect to unemployment and declared income. Across both our
cohort studies and the synthetic control we find that Syrian asylum seekers that were
granted temporary residence registered for fewer unemployment days, and that they
have higher declared incomes. With respect to study grants received, our operationaliza-
tion of time spent in education, we find rather that permanent residents outperformed
temporary residents.
The differences we observe between permanent and temporary residents are not large
in absolute terms. However, it must be kept in mind that these differences are observed
only in the short term. It is nevertheless telling that differences are observable after
such a short period of time. Having that said, it must be stressed that there is a not
insignificant element of uncertainty to these results. A series of placebo tests for both
our synthetic control and cohort studies have given mixed results. Specifically, with
regards to our synthetic control study, randomization inference indicates that several
other nationalities had expected differences from a synthetic control group equal or larger
than that observed for Syrians in absolute terms. In several of these cases the nationality
of interest constitutes a very small number of asylum seekers per year which should lead to
greater year-on-year variance observed. With access to samples of equal size to our Syrian
18
cohorts it is very likely that some of these extreme estimates would be reduced. Placebo
tests for the cohort study were much more conclusive with respect to unemployment days
and study grants, finding convincing null effects among non-Syrian asylum seekers from
the four largest sending countries. We do find an increase in placebo incomes among this
group, but one that is substantially smaller than the observed difference among our true
treatment group. In summary, we find quite robust evidence to suggest that permanent
residency led to a larger number of registered unemployment days, and to decreased
declared income. The robustness of the observed difference in study grants is however
less certain as placebo tests from the cohort study were supportive of the overall findings,
but the cohort matching design and randomization inference from the synthetic control
method resulted in statistically insignificant estimates. Nevertheless, in all cases were the
point estimates consistent with the main results.
Advocates of responsibilities-based migration policy frequently claim that such ap-
proaches provide an incentive, or “push”, for individuals to integrate into society and the
labour market. From this perspective our results can be seen as supportive of this pos-
tulation, at least in the short-term - temporary residents that are subject to a relatively
less-inclusive situation earn more and are unemployed less. However, at the same time,
they are less likely to spend time in education than are those with permanent residency.
Given that those that study in their new country after they receive residency have higher
incomes and fewer unemployment days in the long-term, this is a potential worry for the
success of temporary relative to permanent residents in the long-term.
With a discrepancy in the observed differences due to residency status in unemploy-
ment, income, and study grants, such that we find temporary residency beneficial for
employment, and permanent residency beneficial for education, our results cannot be
said to offer a clear adjudication of which approach is empirically advantageous. Rather,
it would appear that both approaches have their own distinct benefits. While the political
debate about the issue is often framed in such empirical terms - that one approach will
lead to greater or worse outcomes for the target group according to a common outcome
- our results suggest that both approaches can be supported empirically, albeit with dif-
ferent metrics for “success”. In our view then, the issue is should be viewed as largely
normative. Rather than debate whether one approach will lead to greater inclusion than
the other, focus should be shifted to discussing the type of inclusion that the differ-
ent approaches are likely to provide - short-term labour-market benefits versus potential
long-term benefits.
Our findings advocate for further research into how other outcomes are affected by
the shift from temporary to permanent residency - or similar shifts in the rights - and
responsibilities-based migration framework. While we study labour market inclusion in
19
the short-term, migration policy is wide-ranging and its potential outcomes innumerable.
So while our conclusions are drawn in relation to labour market inclusion, it is entirely
plausible, and indeed likely, that other outcomes such as inclusion along social, political,
or migrant well-being dimensions could be more clearly differentiated along empirical
lines.
20
References
Abadie, A., A. Diamond, and J. Hainmueller (2010): “Synthetic control meth-
ods for comparative case studies: Estimating the effect of California’s tobacco control
program,” Journal of the American statistical Association, 105, 493–505.
——— (2015): “Comparative politics and the synthetic control method,” American Jour-
nal of Political Science, 59, 495–510.
Ald´
en, L. and M. Hammarstedt (2014): Utrikes f¨odda p˚a den svenska arbetsmark-
naden: en ¨oversikt och en internationell j¨amf¨orelse, Linn´euniversitet.
Andersson, H. and K. Jutvik (2018): “Do Asylum Seekers Respond to Policy
Changes? Evidence from the Swedish-Syrian Case,” Mimeo, Uppsala University.
Angrist, J. and J. Pischke (2008): “Mostly Harmless Econometrics.(2008),” Prince-
ton, NJ: Princeton University.
Bakker, L., J. Dagevos, and G. Engbersen (2014): “The importance of resources
and security in the socio-economic integration of refugees. A study on the impact of
length of stay in asylum accommodation and residence status on socio-economic inte-
gration for the four largest refugee groups in the Netherlands,” Journal of International
Migration and Integration, 15, 431–448.
Bee, C. and D. Pachi (2014): “Active citizenship in the UK: Assessing institutional
political strategies and mechanisms of civic engagement,” Journal of Civil Society, 10,
100–117.
Bevelander, P. and R. Pendakur (2014): “The labour market integration of refugee
and family reunion immigrants: A comparison of outcomes in Canada and Sweden,”
Journal of Ethnic and Migration Studies, 40, 689–709.
Blomqvist, N., P. Skogman Thoursie, and B. Tyrefors (2018): “Restricting
Residence Permits: Short-Run Evidence from A Swedish Reform,” Working Paper.
Bogic, M., A. Njoku, and S. Priebe (2015): “Long-term mental health of war-
refugees: a systematic literature review,” BMC international health and human rights,
15, 29.
Borevi, K. (2002): “V¨alf¨ardsstaten i det m˚angkulturella samh¨allet,” Ph.D. thesis, Acta
Universitatis Upsaliensis.
21
——— (2010): “2. DIMENSIONS OF CITIZENSHIP: EUROPEAN INTEGRATION
POLICIES FROMA SCANDINAVIAN PERSPECTIVE,” Diversity, inclusion and cit-
izenship in Scandinavia, 19.
Chiswick, B. R. (1978): “The effect of Americanization on the earnings of foreign-born
men,” Journal of political Economy, 86, 897–921.
——— (1991): “Speaking, reading, and earnings among low-skilled immigrants,” Journal
of labor economics, 9, 149–170.
Dahlstedt, I. and P. Bevelander (2010): “General versus vocational education and
employment integration of immigrants in Sweden,” Journal of Immigrant & Refugee
Studies, 8, 158–192.
Delander, L., M. Hammarstedt, J. M˚
ansson, and E. Nyberg (2005): “Inte-
gration of immigrants: The role of language proficiency and experience,” Evaluation
Review, 29, 24–41.
Ferrer, A., D. A. Green, and W. C. Riddell (2006): “The effect of literacy on
immigrant earnings,” Journal of Human Resources, 41, 380–410.
Fowler, A. (2013): “Electoral and Policy Consequences of Voter Turnout: Evidence
from Compulsory Voting in Australia,” Quarterly Journal of Political Science, 8, 159–
182.
Goodman, S. W. and M. Wright (2015): “Does mandatory integration matter?
Effects of civic requirements on immigrant socio-economic and political outcomes,”
Journal of Ethnic and Migration Studies, 41, 1885–1908.
Hammarstedt, M. (2003): “Income from work among immigrants in Sweden,” Review
of Income and Wealth, 49, 185–203.
Joppke, C. (2004): “The retreat of multiculturalism in the liberal state: theory and
policy,” The British journal of sociology, 55, 237–257.
——— (2007): “Beyond national models: Civic integration policies for immigrants in
Western Europe,” West European Politics, 30, 1–22.
——— (2017): “Civic integration in Western Europe: three debates,” West European
Politics, 40, 1153–1176.
Kesler, C. (2006): “Social policy and immigrant joblessness in Britain, Germany and
Sweden,” Social Forces, 85, 743–770.
22
Kogan, I. (2006): “Labor markets and economic incorporation among recent immigrants
in Europe,” Social Forces, 85, 697–721.
——— (2007): Working through barriers, Springer.
Koopmans, R. (2005): Contested citizenship: Immigration and cultural diversity in
Europe, vol. 25, U of Minnesota Press.
——— (2010): “Trade-offs between equality and difference: Immigrant integration, mul-
ticulturalism and the welfare state in cross-national perspective,” Journal of ethnic and
migration studies, 36, 1–26.
Larsen, B., M. Kilstr¨
om, and E. Olme (2018): “Should I Stay or Must I Go:
Temporary Protection and Refugee Outcomes,” Working Paper, Copenhagen Business
School, 1-2018.
Larsson, J. K. (2015): Integrationen och arbetets marknad: hur j¨amst¨alldhet, arbete
och annat” svenskt” g¨ors av arbetsf¨ormedlare och privata akt¨orer, Stockholm: Atlas
Akademi.
Lema
ˆ
ıtre, G. and T. Liebig (2007): Jobs for immigrants: Labour market integration
in Australia, Denmark, Germany and Sweden, vol. 1, Publications de l’OCDE.
Mansouri, F., M. Leach, and A. Nethery (2010): “Temporary protection and the
refugee convention in Australia, Denmark, and Germany,” Refuge: Canada’s periodical
on refugees, 26, 135–147.
Mincer, J. (1974): “Schooling, Experience, and Earnings. Human Behavior & Social
Institutions No. 2.” .
Nekby, L. et al. (2002): How long does it take to integrate?: employment convergence
of immigrants and natives in Sweden, Fief Stockholm.
OECD (2016a): “The labour market integration of immigrants and their children in
Sweden,” OECD Publishing.
——— (2016b): “Making Integration Work: Refugees and Others in Need of Protection,”
OECD Publishing.
RCI 14/2012 (2012): “R¨attsligt st¨allningstagande ang˚aende s¨akerhetssituationen i
Syrien,” Swedish Migration Agency.
RCI 14/2013 (2013): “R¨attsligt st¨allningstagande ang˚aende s¨akerhetssituationen i
Syrien,” Swedish Migration Agency.
23
Ryan, D., B. Dooley, and C. Benson (2008): “Theoretical perspectives on post-
migration adaptation and psychological well-being among refugees: Towards a resource-
based model,” Journal of Refugee Studies, 21, 1–18.
Strik, T., A. B¨
ocker, M. Luiten, and R. van Oers (2010): “The INTEC Project:
synthesis report,” Integration and naturalisation tests: the new way to European citi-
zenship. Nijmegen: Centre for Migration Law, Radboud University Nijmegen. V.
Vertovec, S., S. Wessendorf, et al. (2010): Multiculturalism backlash: European
discourses, policies and practices, Routledge.
24
Appendix
A Difference-in-means approach
Figure 4 below visualizes the control and treatment group of the study. The figure furhter
display the time of measurement, which occurs 16 months after a residency permit is
granted in the cohort study.
Nov 2012-Nov 2013
Nov 2013-Nov 2014
Time of meassurement
Time of meassurement
Dec 2013Dec 2014
Treatment group
Control group
Sep 2012S ep 2013
Oct 2013 Oct 2012
Figure 4: Control and treatment group in our main analysis
Figure displays the treatment and control group of our main analysis. The time of measurement (16 month after granting
of residency permit) in December in 2013 and 2014.
Figure 5 below visualizes the estimate size in the sample used in the original article
and with additional months. The point to the left in the graphs compares Syrians arriving
in September 2012 with those arriving in September 2013, 16 months after receiving a
residence permit (as in the original article). Each step to the right adds one additional
month. The point at the right-side of the graphs compares Syrians granted residence
permits in September-December 2012 with September-December 2013. Although this
gives a larger sample (6500 observations in the comparison to the right), it also brings
problems with sorting as a larger share of individuals knew about the policy change before
applying for asylum.
As suggested by the figure, adding additional months does only marginally affect the
significance and the size of the estimates. In general, the results remains quite stable.
In the lower set of graphs, the Syrian cohorts have been replaced by a placebo group,
constituting all other newly arrived asylum seekers except for Syrians. As can be seen in
the figures, there is no effect regarding unemployment days or declared income. However,
the estimates for study grants are negative and significant, but only if more months than
September is included in the sample. Thus, the general trend regarding study grants for
other newly arrived migrants is slightly negative comparing the cohorts 2012 and 2013.
25
0 10 20 30 40 50 60
Estimate size
Sep
Sep-Oct
Sep-Nov
Sep-Dec
Unemployment days (Syria)
0 4 8 12 16 20 24
Estimate size
Sep
Sep-Oct
Sep-Nov
Sep-Dec
Study grants (Syria)
-400 -300 -200 -100 0
Estimate size
Sep
Sep-Oct
Sep-Nov
Sep-Dec
Declared income (Syria)
-40 -30 -20 -10 0 10
Estimate size
Sep
Sep-Oct
Sep-Nov
Sep-Dec
Unemployment days (placebo)
-12-10-8 -6 -4 -2 0 2 4
Estimate size
Sep
Sep-Oct
Sep-Nov
Sep-Dec
Study grants (placebo)
-50 0 50 100
Estimate size
Sep
Sep-Oct
Sep-Nov
Sep-Dec
Declared income (placebo)
Figure 5: Estimate size in treatment (Syria) and placebo group (other newly arrived
migrants)
Figure displays the estimate size and confidence intervals in the Syrian sample and placebo group regarding
unemployment benefits (top-left), study grants (top-right), and declared income (bottom-left) comparing those arriving in
2012 with those arriving in 2013, adding one month to the sample in incremental step to the right.
Source: GEOSWEDEN (2018).
Moving on to Figure 6, graphs below display the distribution in the dependent vari-
ables used in the study. As can be seen from the figures, the distribution is heavily skewed
in which the majority have a low or, zero value on the dependent variable. This mirrored
in the distribution among other newly arrived migrants.
Table 4 displays the outcome using three matching techniques relying on propensity
scores. From the left, the table shows the number of observations in the treatment and
control group, the size of the estimate for each variable and the T-score. Two conclusions
can be drawn. Firstly, the estimate size of each variable is nearly identical to our main
specification. Secondly, the tests regarding unemployment days and declared income are
significant while the tests regarding study grants yield insignificant results.
Given the highly non-normal distributions in our dependent variables and that our
models are based on only 903 observations we calculate bootstrapped confidence inter-
vals (95% confidence interval) for the bivariate models in our main analysis. These are
reported in Table 5.
26
0
2000
4000
0 100 200 300 400
Unemployment Days
Frequency
Country Other Syria
0
5000
10000
15000
20000
0 5000 10000 15000
Declared Income
Frequency
Country Other Syria
0
5000
10000
15000
20000
0 500 1000 1500
Study Grants
Frequency
Country Other Syria
Figure 6: Histograms of the outcome variables among Syrians (blue) and other newly
arrived migrants (red)
Figure displays the distribution in the three dependent variables included in the study. The blur line represent the
distribution in among Syrians and the red line represent the distribution among other new arrived migrants.
Source: GEOSWEDEN (2018).
Table 4: Matching with propensity scores
Variable Treated (N) Control (N) Estimate T-score
(1) Unemployment days 601 231 25,57 2,157
Declared income 601 231 -145,9 -2,792
Study grants 601 231 2,11 0,81
(2) Unemployment days 601 262 22,11 2,468
Declared income 601 262 -155,24 -3,326
Study grants 601 262 2,58 1,002
(3) Unemployment days 601 262 18,96 2,053
Declared income 601 262 -118,42 -2,781
Study grants 601 262 2,67 1,106
Notes: Table display estimates using (1) ATT estimation with nearest Neighbor Mathing method, (2) Kernel matching
method, and (3) ATT Estimation with Stratification method.
27
Table 5: Bootstrap intervalls
Bootstrap Parametric
Lower Upper Lower Upper
Unemp Days -40.14 -8.17 -40.46 -7.12
Study Grant -11.94 -1.04 -12.74 -0.2
Income 108.87 289.88 119.48 278.37
B Synthetic control group approach
The following tables presents: (1) the weights in the synthetic control group (largest
weights in bold), (2) the predictor balance between the treated and the synthetic control
unit, and (3) the placebo tests.
Table 6 display the composition of the synthetic control groups in each analysis. As
can be seen in table, the two biggest weights in the synthetic control group regarding un-
employment benefits constitute individuals from Morocco and Egypt. Regarding declared
income, the two biggest groups in the control unit consist of individuals from Morocco and
China. Lastly in the control unit for study grants, the two largest weights are received
by individuals from Bangladesh and a sample of countries in Central America.
Table 6: Weights in synthetic control units
Unemployment benefits Declared income Study grants
Country Unit Weight Unit Weight Unit Weight
Russia 0,004 0,053 0,018
Ethiopia 0,008 0,038 0,011
Somalia 0,004 0,056 0,019
Morocco 0,336 0,133 0,008
Uganda 0,02 0,049 0,012
Egypt 0,485 0,054 0,012
Eritrea 0,008 0,056 0,017
Rest of Africa 0,007 0,053 0,02
Lebanon 0,018 0,038 0,006
Turkey 0,053 0,045 0,022
Iraq 0,004 0,046 0,012
Iran 0,003 0,043 0,008
Rest of west Asia 0,004 0,044 0,009
China 0 0,094 0,008
Afghanistan 0,008 0,052 0,005
Bangladesh 0,014 0,031 0,489
Pakistan 0,008 0,041 0,009
Rest of Asia 0,008 0,048 0,009
Central America 0,007 0,025 0,307
Notes: Table displays the weights in the synthetic control units for unemployment benefits,
declared income, and study grants.
Source: GEOSWEDEN (2018)
Table 7 below displays the balance between the treated sample (Syrians) and the
28
synthetic control unit. As can be seen in the table, there is generally a strong balance
between the Syrian sample and the control unit in the dependent variables. That is,
the difference and the trend in the dependent variables between the Syrian sample and
the synthetic control unit are similar in the pre-intervention period (2012 and 2013).
However, it is also important to note that there are some discrepancies in the control
variables included. In sum, the age and the share of married individuals are generally
higher in the Syrian sample. Likewise, the share of men and individuals with children is
also higher in the Syrian sample.
Table 7: Predictor balance between treated and
the synthetic control unit
VARIABLES Treated Synthetic
Difference in unemployment days 35,96 35,94
Unemployment days 86,94 86,85
Age 35,46 32,56
Married 0,57 0,49
Sex 0,62 0,54
Children 0,43 0,39
Difference in declared income -97,97 -97,82
Declared income 130,64 130,48
Age 35,46 32,36
Married 0,57 0,46
Sex 0,62 0,53
Children 0,43 0,33
Difference in study grants 0,84 0,83
Study grants 2,19 2,20
Age 35,46 33,39
Married 0,57 0,37
Sex 0,62 0,67
Children 0,43 0,26
Notes: Table displays the predictor balance between the treated
sample and the synthetic control units for unemployment benefits,
declared income, and study grants.
Source: GEOSWEDEN (2018)
Lastly, Table 8 displays the placebo tests in the synthetic control group approach.
Here we excluded the Syrian sample, and performed the same analysis as in the original
article. As can be seen in the table, three other groups of newly arrived migrants (from
Lebanon, Turkey, and Bangladesh) display a higher increase in the usage of unemployment
days. Regarding declared income four other groups (from Morocco, Lebanon, Asia and
a sample from Central America) have larger decreases compared to Syrians. Lastly,
regarding study grants six other groups (Somalia, Egypt, Asia, China Afghanistan, and
Pakistan) have larger increases compared to Syrians in regards of study grants.
29
Table 8: Synthetic control group: Placebo tests
Country Unemployment benefits Declared income Study grants
Syria 16,89 -71,57 2,08
Russia 3,35 114,90 0,58
Etiophia -4,84 -51,57 -2,72
Somalia -6,23 -24,23 7,71
Morocco -11,86 -128,49 -40,20
Uganda 3,77 -5,05 0,70
Egypt -1,41 -13,46 1,18
Eritrea 12,94 -28,35 2,96
Africa -0,17 -16,71 0,19
Lebanon 19,44 -73,04 -1,49
Turkey 18,19 -53,59 -1,30
Iraq -15,18 52,27 -0,96
Iran 0,97 15,00 -0,45
Asia -0,92 -88,12 2,48
China -49,99 442,10 2,63
Afghanistan -11,12 30,22 10,90
Bangladesh 70,65 209,91 1,85
Pakistan 13,79 -37,85 6,51
Rest of Asia -4,84 79,14 1,06
Central America -1,17 -274,11 -1,04
Notes: Table displays the estimates in the placebo test for unemployment benefits, declared
income, and study grants.
Source: GEOSWEDEN (2018)
30