How Do Migrants
Choose Their Destination Country?
An Analysis of Institutional Determinants
Wido Geis, Silke Uebelmesser and Martin Werding
CESifo GmbH
Poschingerstr. 5
81679 Munich
Germany
Phone:
Fax:
E-mail:
Web:
+49 (0) 89 9224-1410
+49 (0) 89 9224-1409
[email protected]
www.cesifo.de
How do Migrants
Choose Their Destination Country?
An Analysis of Institutional Determinants
Wido Geis∗
Silke Uebelmesser∗∗
Martin Werding∗∗∗
Preliminary version, October 2008
To be presented at the CESifo conference:
“Reform of the Welfare State: A New European Model”
31 October - 1 November 2008, Munich, Germany
* Ifo Institute for Economic Research at the University of Munich
** Center for Economic Studies (CES) at the University of Munich & CESifo
*** Ifo Institute for Economic Research at the University of Munich & CESifo
First Author’s address: Ifo Institute for Economic Research, Dept. of Social Policy and Labor Markets,
Poschingerstraße 5, 81679 Munich, Germany. Telephone: ++49 (0)89 9224-1691, Telefax: ++49 (0)89
9224-1608, E-mail: [email protected]
Abstract
For a long time, migration has been subject to intensive economic research. Nevertheless,
empirical evidence regarding the determinants of migration still appears to be incomplete.
In this paper, we analyze the effects of socio-economic and institutional determinants,
especially labor market institutions, on migrants’ choices of their destination countries.
Our analysis is based on a large data set constructed from micro-data for France, Germany,
the UK and the US. We study decisions to migrate to one of the four countries using
a multinomial choice framework. We find that, besides wages and unemployment rates,
unemployment benefits, employment protection and union coverage influence the location
choice of migrants. In addition, health-care and education systems as well as the tax
burden have a significant impact.
1
Introduction
Moving to another country often is a decisive turning point in the life of the migrants.
They have to build up a new social network and get accustomed to a new institutional
framework. Usually, migration is not the result of a spontaneous decision, but the outcome of a long decision process. Therefore, the institutions of possible destination countries should at least play some role in this process. For instance, if public regulation
impedes labor-market entry for “outsiders”, migrant workers should ceteris paribus prefer
destination countries with more flexible labor markets. Similarly, older persons should
prefer countries that give them access to a better health-care system, and parents should
prefer countries that offer their children a better education. The aim of our paper is to
analyze whether these and other institutions play a role for the migration decision and to
quantify their effects.
How migrants choose their destination country is an interesting research question per
se. In addition, the answer to this question has important implications for migration
policy. On the one hand, it can help to estimate migration potentials for the case of
unrestricted mobility which, in turn, may have a strong influence on the final decision
about immigration policy if a country is contemplating some modifications. On the other
hand, it can have an influence on the assessment of migration regulations already in
place. A prominent example for this is the large inflow of Polish people to the UK
after the EU enlargement in 2004. It is argued that a large part of these people would
have come to Germany, if Germany had also opened its labor market immediately (Baas
and Brücker 2007). However, in the relevant years unemployment in the UK was much
lower than in Germany. Thus, one could also argue that these people would have gone
to the UK anyway because of their better labor-market prospects there. Last but not
1
least, knowledge about the determinants of migration decisions can help policy makers
to design effective programs to attract certain groups of foreigners (such as the ”British
Highly Skilled Migrant Programme”, the H1B visa in the US, or the German “Green
Card” for IT specialists).
Over the last few years, a series of papers have emerged that analyze the determinants
of migrants’ location choices (e.g., Pedersen et al. 2005; Mayda 2007; Docquier et al.
2007). These papers are based on international macro-data panels.1 Beside unemployment
rates and GDP per capita, they find that distance plays an important role for migration
decisions. In addition, a common language and colonial ties obviously have a positive
effect on the choice of a particular destination country. However, the use of aggregate
data carries some problems, as the determinants of migration most likely differ between
population groups (e.g., labor-market access may vary by qualifications and experience,
the quality of the destination country’s education system is more important for young
parents than for childless retires, etc.).2 Therefore, we follow another route and base our
analysis on micro-data.
Unfortunately, no large international micro-data base exists that could be used for
our purposes.3 We therefore construct our own data set, merging micro-data of four of
the most important immigration countries, namely France, Germany, the UK and the
US.4 Because of the limited number of countries covered, we can only analyze migrants’
1
See Lundberg (1993) for an earlier study based on cross-section data.
2
Docquier et al. (2007) differentiate between high-skilled and low-skilled migrants, whereas the other
researchers look at total migration between two countries.
3
The European Labour Force Survey would be such a data base but, in its publicly accessible form,
it contains no information on the origin of migrants.
4
Defoort (2007) states that, together with Canada and Australia, these countries attract 77% of all
migrants to the OECD world.
2
choices between the four countries given that they are willing to migrate at all and end
up living in one of these four destination countries.5 We combine these micro-data with
data regarding a number of institutions that potentially have an impact on the location
decision. Using a Multinomial Choice framework, we then estimate the effects of these
institutions on migrants’ choices of a particular destination among our four countries.
From a technical perspective, Constant and D’Agosto (2008) is the paper that is probably
closest to ours. Based on a data set covering Italian scientists living abroad, they analyze
the determinants of their choice of a destination country. In contrast to our approach,
however, they only use individual characteristics and no general features of the destination
countries as explanatory variables.6
To date, the impact of institutions on migration decisions has hardly been studied
in a systematic fashion.7 Thus, our results offer interesting and important new insights
regarding the determinants of migration decisions. Our more conventional findings are
that wages and migrant networks have a positive effect on the probability to migrate to
a particular country, while the unemployment rate has a negative effect. In addition, we
find that public health expenditures and PISA-scores have a positive impact, while the
income tax wedge negatively affects migration. Moreover, the considered labor market
institutions - employment protection, union coverage and benefit replacement - all have
positive effects on the migration decision.
5
For an analysis of the unconditional migration decision, one would also have to observe the populations
and the institutions in the source countries, and one should also be able to add more destination countries.
6
Furthermore, they use a pure Multinomial Logit Model, whereas we use a combination of a Multinomial and a Conditional Logit Model.
7
Borjas (1999) investigated the role of welfare benefit entitlements for migration within the US which
led to his “welfare magnet” hypothesis. Docquier et al. (2007) find a positive effect of social expenditure
and health expenditure. [To our knowledge] there are, however, no studies of labor-market institutions
as potential determinants of migrants’ location choices.
3
The paper is organized as follows. In the next chapter, we explain how our data set
has been constructed. In chapter 3, we present a number of descriptive results regarding
immigration to the four countries we are studying. Chapter 4 deals with determinants of
migration and, in particular, with institutions that may have an influence on migration
decisions. In chapter 5, we discuss our estimation strategy and in chapter 6, we present
our estimation results. Chapter 7 concludes.
2
The data set
Our data set combines micro-data from large official surveys of the British, French, German and US population. The source of our French data is the Enquête Emploi en Continu
2005, a representative survey of about 0.5% of the French population. The German data
are taken from the Mikrozensus 2005, a representative 1% survey (0.7% in the Scientific
Use File we are using). The British data are from the (British) Labour Force Survey for
the first quarter of 2005, a survey of about 0.2% of the population in the UK. For the
US, we use the American Community Survey 2005, a representative 1% survey of the US
population. In order to analyze the motivation of migrants, flow data would actually be
preferable to stock data. However, existing flow data generally contain much less information and are less precise than stock data. Therefore, we rely on data of the latter type,
implying that we actually do not analyze decisions to migrate to another country, but
decisions to migrate to another country and stay there until the sampling period.
An important preliminary step is to specify the definition of migrants. Immigrants
could be defined as persons holding one or more foreign nationalities. Yet, this appraoch
is problematic as naturalization policies of the four countries differ substantially. For
instance, the German naturalization policy is much more restrictive than the American
one. Hence, looking at individuals with foreign nationalities could lead to biased results.
4
Defining immigrants by their country of birth circumvents this problems. However, as
foreign-born children whose parents are both natives are then classified as immigrants,
this definition can also lead to problems, e.g., if a non-marginal part of the foreign-born
population are children of armed forces positioned abroad. Therefore, we choose the
following approach: we define immigrants as foreign-born people, but re-classify persons
with two native parents as natives.8 The effect of this re-classification on the overall
number of immigrants is small, but their composition changes notably (see Geis et al.
2008 for more details).
In the case of Germany, we have to deal with two specific issues. First, in the German
data the country of birth of immigrants is not recorded. We, therefore, use the nationality,
respectively the nationality before naturalization, as a proxy for the country of birth. The
second issue is related to the “(Spät-)Aussiedler ” legislation. According to this legislation,
persons with German ancestors (who sometimes emigrated centuries ago, mainly to countries in Eastern Europe) can acquire the German nationality immediately upon arrival
in Germany. After the fall of the “Iron Curtain”, a large number of “Spät-Aussiedler ”
came to Germany (Koller 1997). Yet, in spite of their quantitative importance, official
statistics in Germany hardly collect any data on this group. In our data set, we are able
to identify them immigrants,9 but we cannot assign to them a country of birth.
For the source countries, or countries of birth, we choose the following classification:
EU countries, non-EU Europe (including Russia and Turkey), West Asia (from Lebanon
8
For the UK, we re-classify persons who state to be “ethnically British” respectively.
9
Alternative explanations for why Germans with German parents should have “migrated” to Germany
are highly unlikely. For instance, since World War II Germany had hardly any armed forces positioned
abroad. Also, all persons with German nationality who came to Germany before 1949 are automatically
defined as natives.
5
to Iran), East Asia and Oceania, Africa, Latin America, Canada10 and ”unclassified”11 .
A more detailed differentiation is not possible, due to the existing classifications in the
German and French data sources. For the econometric analysis, people who migrate
between our four destination countries also have to be excluded,12 but the descriptive
results reported in the next section include these migrants.
As a further step, we have to standardize a number of other variables we are using.
The only institution for which the standardization is not trivial is education. Here, we
classify educational attainments of our observations using the International Standard
Classification of Education (ISCED) 1997. For the German data, we use the algorithm
proposed by Schrödter et al. (2006) and for the American data the mapping between years
of schooling and ISCED levels given in Institute for Education Sciences (2007). The French
data already contain education levels in the ISCED classification. For the British data,
our re-classification follows the LFS User Guide (2007) with two deviations.13 Also, we do
not use all ISCED levels, but form four categories: no secondary educational attainment
(ISCED 0-1), lower-secondary educational attainment (ISCED 2), upper-secondary and
post-secondary non-tertiary educational attainment (ISCED 3-4) and tertiary educational
attainment (ISCED 5-6). Differentiations between ISCED 3 and 4 and between ISCED 5
and 6 are hardly comparable across countries.
10
In the case of Germany, Canadians are excluded, as we cannot distinguish them from US Americans.
11
By far the largest part of them being German “Aussiedler ”.
12
The reason is that, with respect to migration between the four countries, we can only observe the
potential outcomes of migration to three destination countries. Decisions to stay in the home country or
to migrate there, though vastly different, cannot be told apart.
13
First, we classify people who state to have been in school, but have not acquired any formal degree
as ISCED 1, not ISCED 2. Second, we do not classify people who state to have “other qualifications”
as ISCED 3, but assign them the median ISCED level of people with the same age and the same (last)
occupation. For this, we use the SOC (Standard Occupational Classification) 2000 unit-level classification
which distinguishes between 353 different occupations. An assignment of educational levels is necessary,
as most foreign degrees are recorded as “other qualification” in the British LFS.
6
In the last step, we merge the standardized variables from the four national data sets
to form one large data base, using the weights from the original data sources. As these
weights make the data sets representative for the different countries, our data base should
also be representative. Since the Enquête Emploi does not contain information on persons
who are younger than 15, our descriptive results only refer to people aged 15 and over.
For the econometric analysis, we further drop all individuals who are younger than 25,
as many of these people have not yet reached their final education level. Including these
observations could lead to biased results.
3
Some descriptive results
Before entering the econometric analysis, we present some descriptive statistics from our
data. These statistics do not only serve as background information for our estimation results, they are also interesting in themselves. Applying a consistent definition of migrants,
our data give a very precise picture of the migrant population in the four countries.14
Comparing the shares of immigrants in the population aged 15 and older in the four
countries already leads to a surprising result (cf. table 1). We find the highest share
of immigrants in Germany, with 16.8%, followed by the US with 14.4%, France with
8.5% and the UK with 8.2%. The large share of immigrants in Germany, a country
that is actually well-known for its restrictive immigration policy, has two reasons. The
German “guest worker” agreements with Turkey, Italy, Yugoslavia, Spain and Portugal
caused a large immigration wave between 1955 and 1973, leading to a continuous inflow of migrants due to family re-unification programs. Probably even more important is
the “(Spät-)Aussiedler ” legislation mentioned above. The other shares are in line with
14
For a larger set of descriptive results that are based on the same data base, see Geis et al. (2008).
7
expectations: the US as an “immigration country” has a much larger share of immigrants
than France and the UK. Effects of the recent, more liberal immigration policy in the
UK, especially the opening of the labor market for people from Eastern Europe in 2004,
are not yet visible in the data for 2005.
Table 1 also gives an overview over the most important countries of origin of migrants
to the four countries. In France, these are above all neighboring countries in Europe and
Northern Africa. In Germany, Southern and Eastern European countries are the most important countries of origin; at the same time, one third of all German immigrants cannot
be classified, most of them probably being “(Spät-)Aussiedler ”. In contrast to Germany
and France, the most important source countries of immigrants to the UK are former
colonies outside Europe, in addition to Ireland and Poland. For the US, countries in Central and Caribbean American and the large East Asian countries are the most important
ones. It is remarkable that almost one third of the American immigrant population comes
from Mexico. In none of the European countries, immigration is similarly concentrated
on one country of origin. However, the European countries also differ with respect to
the concentration: 38.8% of the immigrants to France, but only 26.8% and 24.5% of the
immigrants to Germany respectively the UK are from the three respective most important
countries of origin.
There are not only differences regarding the countries of origin of immigrants, but
also regarding their structure in terms of educational attainments. Table 2 shows how the
immigrants aged 25 to 54 are distributed over the educational groups defined above; for
comparison, we add the corresponding distribution of natives. The share of “high-skiled”
immigrants (ISCED 5+6) is highest in the US, followed by the UK, Germany and France.
The picture is similar for “qualified” immigrants, i.e. for those with at least an upper
secondary degree (ISCED 3–6). Obviously, the Anglo-Saxon countries attract people
8
with higher qualifications than the countries in Continental Europe. At the same time,
immigrant populations living in a particular country are far from being homogeneous.
For instance, the share of high-skilled immigrants from Mexico to the US is far below
that of natives; for immigrants from other Latin American countries, this share is also
below that of natives, but the difference is much smaller; for immigrants from non-Latin
American countries, however, the share by far exceeds that of natives. This leads to a
U -shaped pattern of educational attainments of all immigrants to the US. In Europe,
there are similar differences between various immigrant groups, e.g., between Turkish and
other immigrants to Germany, but they are much smaller than in the US.15
A further interesting aspect is the economic integration of immigrants. As a rough
measure, we include unemployment rates (following the ILO definition) differentiated by
educational attainments in table 2. In all European countries, unemployment rates of
immigrants are much higher than those of natives, unemployment rates of immigrants in
the UK still being much lower than those in France and Germany. In the US, however,
unemployment rates of immigrants fall short of the ones of natives, except for the highest
education level (ISCED 5+6). Note that this cannot be explained by different selections
into unemployment and non-participation, as the ratio between the participation rates of
immigrants and natives in the US is not smaller than in Europe. These observations clearly
indicate that all the European countries considered have more difficulties in integrating
immigrants into their labor markets than the US. More importantly, they also show that,
when analyzing the determinants of migration, relying on country-wide averages is less
suitable than using specific information on immigrants.
15
See, again, Geis et al. (2008) for more details.
9
4
Determinants of migration
In the economic migration literature, wages and unemployment rates are generally considered to be the most important determinants of migration (see the seminal papers by
Sjaastad 1962; Todaro 1969; Harris and Todaro 1970). As these two factors vary strongly
across different population groups, detailed data are needed for a well-founded econometric analysis. In our four data sets, unemployment is recorded following the ILO definition.
Building on these data, it is straightforward to calculate specific unemployment rates of
immigrants differentiated by education and gender.
Obtaining consistent data on wages, however, is very difficult in general and still far
from easy with our micro-data, since the wage data provided in our data sets are not
comparable across countries. Nevertheless, we tried hard to generate consistent wage
information from our four national data sources. In a first step, we calculate wages per
hour using information on wage earnings and working hours contained in all datasets.
As our German dataset actually contains income and not wage data, we consider only
persons stating to have no other income than wages.16 In a next step, we calculate for each
country wages of immigrants for the various gender-education groups relative to average
wages. In the last step, we multiply these relative wages with data on GDP per capita
from OECD (2007a). We cannot directly compare our intermediate results regarding
wages per hour as for the European countries we observe net wages, while for the US we
observe gross wages. Note that this means that the dispersion of our wage measure for
the US is probably exaggerated compared to that in the European countries. Still, we
think our measure of wages is superior to the (uniform) GDP per capita which is used
16
For all other measures we consider the complete dataset.
10
in many other studies on the determinants of migration (see, e.g., Pedersen et al. 2005;
Mayda 2007; Docquier et al. 2007).
Another very important determinant of migration are migrant networks (see Munshi
2003 for a comprehensive analysis of Mexican migrant networks in the US). These networks
facilitate migration as they offer new members detailed information about the destination
country and provide a social network once they have arrived. Furthermore, where such
networks exist many people have the opportunity to use preferential family re-unification
programs to immigrate. In our econometric analysis, we use the share of persons from
a certain source country in the population of the destination country as a measure of
the strength of the migrant network. Due to data limitations, we can actually do the
calculations only for immigrant groups representing at least 0.2% of the population in the
destination country. This need not be a problem as smaller groups are probably lacking
the critical mass to deliver the potential benefits of a network. As the effect of the size
of the network on migration decisions may not be linear – in smaller networks, additional
persons are probably more important than in larger ones – we also use the square of this
measure.
In addition, immigration policy and the openness of a country relative to immigrants
may also influence the migration decision. However, immigration policy is difficult to
measure – immigration laws are usually complex and rather case-specific – and there does
not exist a consistent indicator of immigration policy or openness for all four destination
countries in our sample.17 Thus, we cannot observe this determinant directly. Yet, as one
would assume that in the long run a more open country attracts more immigrants, we
17
For the European countries, the British Council and Migration Policy Group has proposed such an
indicator, called MIPEX. However, it does not contain any information regarding the US.
11
use the total share of foreign-born persons in a country as a rough control for openness
to migrants.
Beside the potential factors discussed so far, unemployment benefits should also have
an influence on the migration decision. Expected income in the destination country is
basically given by the employment rate times wages plus the unemployment rate times
these benefits. However, the quantification of unemployment benefits is complicated as
benefit entitlements often depend on the time a person has been unemployed. For our
set-up, the most convincing measure that is available are average replacement rates for
the first five years of unemployment as provided by the OECD (2004).18 The role of
unemployment benefits in a given country may also depend on the unemployment rate.
If unemployment is low, migrants expect to find work, and benefits have next to no
influence on the decision for this country. However, if unemployment is high, migrants
expect to become unemployed with some probability, and the benefits really matter for
their potential income. To control for this effect, we interact the replacement rate with
the unemployment rate.
Other factors which affect expected income in the destination country are income taxes
and social-security contributions. As we are unable to fully capture the different schemes
by which these levies redistribute income from highly productive to less productive individuals we use the total tax wedges (including social-security contributions) for average
high and low income workers without children and for average workers with childrenas
indicated by the OECD (2006b) as a measure for the fiscal burdens that arise.
There are further labor-market institutions that may also have an impact on the
18
Unfortunately, these data do not allow for a differentiation by educational levels. The replacement
rate may be higher for low-skilled than for high-skilled individuals if part of the benefits are a lump sum.
12
location decision of migrants. For people who have to build up a new existence abroad, job
security is probably an important criterion. A good measure for job security is the (overall)
employment protection legislation (EPL) indicator calculated by the OECD (2004). It
ranks the legal requirements for dismissals in various countries on a scale from 0 to 6 where
higher values indicate stricter regulation. Another important labor-market institution is
the power of trade unions. To capture this, we use the share of employment contracts
covered by collective wage agreements (OECD 2004). Employment protection and union
power, though attractive for those covered or represented, may also lead to insider-outsider
problems. Therefore, we additionally interact them with the unemployment rate.
When considering to migrate, people may not only look at their labor-market prospects
but also at other institutions. One important factor may be the health care system in
potential destination countries. We use the public health care expenditure relative to
GDP from the OECD (2007a) as a rough measure for the quality of the health care
system. For young families (and persons who think about having children), the education
system in the destination country may also play a role. We thus include PISA science
scores (OECD 2006a) as a measure for the quality of the education system. At the
same time, people who do not (plan to) have children may not prefer high-quality public
education as this requires higher taxes. Additionally, a generous old-age pension system
could also have a positive impact on the location choice but, since migrants first have to
pay contributions, the effect can also be negative. In any case, we use pension replacement
rates differentiated by wage brackets from the OECD (2007b) to control for this aspect.
Last but not least, the education structure of a destination country can affect the choice of
potential immigrants. Countries with a high share of high qualified people are potentially
more innovative than others and thus more likely to generate higher growth. We therefore
include the share of people with ISCED 5+6 from our micro data as a measure for the
13
education structure. There are certainly many more institutions that may also play a
role for the decision to migrate to a particular country. We believe, however, that the
institutions described here and summarized in table 3 are the most important ones, at
least among those that can be measured in a meaningful way.
5
Estimation strategy
For the estimation, we use a combination of a Conditional and a Multinomial Logit
Model (CMNL).19 The basic idea of the model is that among a range J of options – in
our case, among destination countries, individuals choose the one that offers them the
highest utility, Vij ; here, i denotes the individual and j the option. This utility, in turn,
depends on option-dependent explanatory variables, Xij , and on option-invariant ones,
Zi . Assuming a linear relation and adding an error term, utility levels are represented by
the following equation:
Vij = Xij0 β + Zi0 γj + ij
(1)
The observed variable yij indicates which option an individual has chosen. Thus, for
k ∈ J , yik = 1 and yi¬k = 0 if Vik = maxj (Vij ). Furthermore, it is assumed that the
error terms, ij , are independent and log-Weibull-distributed; the density of this function
is e(−ij −e
−ij
)
. It can be shown that the probability function has the following form (see
Amemiya 1981):
0
0
eXij β+Zi γj
pij = P rob(yij = 1|X, Z) = PJ
0
0
Xil β+Zi γl
l=1 e
(2)
For the estimation, this CMNL has to be transformed into a pure Conditional Logit
Model. Following Cameron and Trivedi (2005), we use the following probability function
19
Although this combination is well-known in the econometric literature, it has no particular name. It
is sometimes called Mixed or Multinomial Logit Model, but these labels also refer to other models.
14
for the estimation:
∗0 ∗
0
∗
eXij β+Zij γ
pij = P rob(yij = 1|X, Z ) = PJ
0
∗0 ∗
Xil β+Zil γ
l=1 e
(3)
where Z ∗ is the Kronecker product of Z and a J × J identity matrix I, Z ∗ = Z ⊗ I,
and γ ∗ = [00 , γ20 , . . . , γJ0 ]; γ1 = 0 is a normalization. The model is estimated by maximum
likelihood. The resulting first-order condition is given by:
N X
M
X
yij (xij − x̄i ) = 0
(4)
i=1 j=1
with x̄i =
Pm
l=1
pil xij . The marginal effects of changes in the option-dependent explana-
tory variables can be calculated as follows (cf. Cameron and Trivedi 2005):
∂pij
= pij (δijk − pik )β
∂xik
(5)
The equation gives the effect of a change in the independent variable for option k on the
probability that option j is chosen; δijk is equal to 1 if j = k and 0 otherwise. Elasticities
are given by:
∂pij xik
= xik (δijk − pik )β
∂xik pij
(6)
It can be shown that the resulting estimates are consistent, asymptotically normal and
asymptotically efficient. A characteristic of the Conditional Logit Model which is often
criticized is the independence of irrelevant alternatives. In our case, this is actually an
advantage, as we can only observe a limited number of countries. Our results would be
of very limited relevance if the possibility to go to Spain had an effect on choices between
Germany and the US.
The low variation in our institutional variables – most of them are country-specific –
clearly presents a challenge. On the one hand, considering all of them in a single regression
is not possible, as this would lead to multi-collinearity. On the other hand, more detailed
15
information is not available, and adding more destination countries to our data set is
all but easy. Therefore, we choose to expand the number of estimations using different
combinations of the various institutions captured by our data. The following individualspecific variables are included in all regressions: level of education, gender, age (and age
squared), (squared) years since migration and region of the country of birth. Furthermore,
all regressions contain information on wages, unemployment rates and the (squared) size
of migrant networks, as these are variables which are conventionally found to have a
strong impact on migrants’ location decisions. In a first step, the institutional variables
are then included one by one in the regressions. As there could also be interactions
between the institutions, we repeat the estimations with all possible pairs and triplets of
institutions (while including four or more institutional variables in a single estimation may
lead to multi-collinearity). If the dispersion of estimated coefficients for an explanatory
variable is not too large, the estimate should not be affected by an omitted-variables
problem. Similar approaches have been proposed in other areas of economics and social
sciences (for instance, Sala-i-Martin 1997 uses a similar approach to explain economic
growth; Hegre and Salaris 2007 do the same to explain civil wars). In addition, we use
the extreme-bound criterion proposed by Leamer (1985) to test the significance of our
estimates.20
20
The lower (upper) extreme bound is given by the minimum (maximum) estimate minus (plus) two
times the corresponding standard deviation. We also experimented with the criterion proposed by Salai-Martin (1997). However, in our case (low standard errors of the estimates but relatively high variation
over specifications) this criterion is inappropriate, as it gives no weight to the variation of the regressors
over specifications.
16
6
Estimation results
The estimation results of the regressions in which we control for wages, unemployment
rates, networks (squared) and one further institutional variable (cf. table 3) are shown
in table 4. They are all significant at the 1% level. Due to space limitations, estimates
for the individual-level characteristics are not reported; except for the country-of-birth
dummies for Canada and for those not classified (mainly German “Spät-Aussiedler ”),
they are also significant at the 1% level. The pseudo-R2 of about 0.63 indicates that
our explanatory variables are indeed important determinants of migrants’ choices of a
destination country. All variables, except for the unemployment rate, the tax wedge and
the pension replacement rate, squared network and the interactions of benefit replacement
and union coverage with unemployment have a positive effect.
Table 5 displays the median results derived from all estimations. The medians of
the estimates have the same signs as the estimates in table 4.21 This indicates that the
estimated effects are stable across differing specifications. We find for wages the expected
positive and for the unemployment rate the expected negative effect. Additionally, as
expected immigrant networks have a decreasing positive effect (the effect is decreasing as
the squared network variable has a negative sign). This indicates that networks really
facilitate the immigration to a country; however, when the network is already large, an
increase in the network has hardly an additional positive effect. Moreover, we find that
open countries, with a high share of foreign born people, are indeed more attractive for
immigrants than countries with a low share. Less clear a priori, employment protection,
union coverage and benefit replacement have positive effects indicating that migrants
21
The average estimates for benefit replacement and union coverage have a negative sign while it is
positive if the median estimates are considered.
17
prefer destination countries where they are protected from labor-market risks. In addition,
it indicates that in the four countries considered the immigrants in our data-sets are not
outsiders on the labor market. If this were the case immigrants would hardly be covered
by the protection measures; moreover, as these measures hamper the access to the labor
market they would be detrimental for immigrants. Nevertheless, the negative interactions
of employment protection and union coverage with the unemployment rate indicate that
if unemployment becomes large an insider-outsider effect may occur.
We find a negative effect of income tax wedges on the migration decision, although
higher taxes are potentially connected with better public services. The negative effect
of pension replacement rates could be explained by the fact that more generous pension
systems usually involve higher contributions and may also be subject to higher political
risks than less ambitious schemes. Although public health expenditures and a good education system are related to taxes which also have to be paid by healthy immigrants
without children, we find that overall they both have a positive effect on the immigration
decision. The negative effect of the share of high-skilled people is puzzling. However, a
potential explanation is that many migrants are high skilled and have to compete against
these people; we will discuss this in more detail below when we analyze the results for
high-skilled and low-skilled migrants separately.
We have repeated the calculations for the sub-group of individuals who have migrated
after 1995; the results are given in table 6. The estimates confirm our previous results;
nevertheless three estimates change their sign. We find now the expected positive effect
of the share of high-skilled natives. Moreover, the estimates for union coverage and the
PISA-scores become negative. In the case of union coverage, this can be explained by
an insider-outsider effect, as discussed above. In the case of PISA-scores this is puzzling;
one explanation could be that better education is generally connected with higher public
18
expenditures which are alone relevant for immigrants without children.
To assess the quantitative importance of our estimates, we calculate a matrix of elasticities for the socio-economic and institutional variables that is presented in table 7.
Among other things, we find that a 1% increase in the unemployment rate in the US
decreases the probability to migrate to the US by 0.14%, while it increases the one to
go to Germany by 0.07%, to the UK by 0.02% and to France by 0.04%.22 A 1% increase
in the unemployment rate in France decreases the probability to go to France by 0.82%
(the large difference between the US and France being due to the fact that a 1% increase
equals a total change by 0.07 percentage points in the US but by 0.19 percentage points
in France). Also, the ex-ante probability to go to the US is higher than the probability to
go to France. The elasticities with respect to wages have the same magnitude as those for
unemployment rates, but with the opposite signs. Most of the elasticities regarding the
institutional variables are even larger than the ones for wages and unemployment rates.
Note, however, that this is partly due to the scaling and the actual range of variation of
the variables.23 In any case, they show that the role of the labor-market institutions and
other institutional characteristics of potential destination countries is not only statistically
but also economically significant for migrants’ location choices.
Determinants of location choices are very likely different for high-skilled and low-skilled
migrants. Therefore, we repeat our estimates running separate regressions for low-skilled
(ISCED 0-2) and qualified (ISCED 3-6) migrants.24 Note that, in contrast to existing
22
Note that, by definition, these latter effects should sum up to 0.14% (the actually resulting 0.13% is
due to rounding), exactly absorbing the change in Prob(US).
23
For instance, the employment protection indicator effectively ranges between 0.067 and 1.000, while
the PISA scores lie between 489 and 516 points.
24
In this case, we exclude interactions with the unemployment rate, as they could lead to multicollinearity problems in this smaller data set.
19
studies based on macro data, we already control for differences between skill levels in the
analysis of the full sample. However, the estimated coefficients only represent average
effects, and skill-related differences are captured in option-invariant variables and in the
error term.
Table 8 summarizes the estimated results for low-skilled and qualified migrants. For
wages, networks, share of foreigners and health expenditures we find for both groups positive effects, as in the full data set; for unemployment and tax wedges we find negative
effects. The other estimates differ between the two groups. The estimated effect of employment protection is negative for low-skilled and positive for qualified immigrants. This
indicates, that for low-skilled immigrants the hardening of the access to the labor market
outweighs the potential positive effect. For high-skilled immigrants, employment protection obviously has less of a negative effect on their access to the labor market. Union
coverage and benefit replacement rate have positive effects for low-skilled immigrants and
negative ones for qualified. This could be explained by the fact, that low-skilled people
generally benefit more from high unemployment benefits and tariff wages than high-skilled
ones. Moreover, unemployment benefits are generally correlated with costs which have
to be paid over-proportionally by more highly skilled people. Pension replacement has
a negative effect for low-skilled people and a positive one for high skilled; this may be
due to the fact that the height of pensions relative to former income is more important for high skilled, whereas social security payments are more important for low-skilled
immigrants.PISA-scores have the expected positive sign for high-skilled immigrants and
a negative sign for low skilled. This negative sign is puzzling. For this constellation not
even state education expenditures are a valid argument, as high skilled generally pay more
taxes than low skilled; moreover, low-skilled immigrants have generally more children than
high-skilled. The share of high-skilled people shows the expected signs. For high-skilled
20
immigrants, who have to compete with high-skilled natives, it is negative; for low-skilled
immigrants, who are probably complements, it is positive.
7
Conclusions
The decision to migrate to a particular destination country is a complex process and
may be affected by many different factors. Economists conventionally expect wages and
unemployment rates to have an impact on this decision. In this paper, we have shown
that the institutional setting in potential destination countries also plays an important
role. Moreover, our results indicate that wages and unemployment rates alone do by far
not suffice to explain location choices of (“non-refugee”) migrants.
In particular, we find that public health expenditures and PISA-scores have positive
effects on the migration decision, indicating that migrants value good education and
health systems. For employment protection, union coverage and unemployment benefits,
the effects are also positive. Thus, protection against labor market risks is obviously
important for immigrants. Our estimate for pension benefits is significantly negative.
This is not counterintuitive, however, as pensions are generally financed by contributions
of workers. If migrants expect that the pension scheme will become less generous over time
– an assumption that is plausible in countries with low fertility rates – these contributions
are sunk costs for them. Additionally, we find a strong, but declining, effect of the size of
immigrant networks.
We are unable to consider all characteristics of destination countries that are potentially important for the migration decision. For instance, we are lacking any measures for
the access of migrants to housing.25 Also, some of the proxies we are using, e.g., for edu-
25
Climate and natural beauty are also very likely to play a role for migration decisions. However, their
21
cation systems, health protection as well as immigration policies, possess some limitations
which result from the lack of consistent data. Another limitation of our analysis arises
from the fact that, for some of the variables we include, there is actually little variation
in the data. Combining micro-data from four major destination countries is clearly an
important step in order to understand better the determinents of the migration decision,
but for some of the institutions we investigate, it is difficult to reconstruct all variation
at the individual level, while others are simply fixed at a national level, i.e., they are the
same for all migrants living in one country.
Still, we are convinced that the approach we have chosen helps to provide new insights
into the question of which institutions play a role for the migration decision. As we have
seen labor-market institutions - beside wages and unemployment rates - have a strong
influence on the migrants’ choice of a destination country.
References
[1] Amemiya, Takeshi (1981), “Qualitative Response Models: A Survey”, Journal of Economic Literature 19(4), 1483–1536.
[2] Baas, Timo and Herbert Brücker (2008), “Macroeconomic consequences of migration diversion – A
CGE simulation for Germany and the UK”, IAB Discussion Paper No. 03/2008, Nürnberg.
[3] Borjas, George J. (1999), “Immigration and Welfare Magnets”, Journal of Labor Economics 17(4),
607–637.
[4] British Council and Migration Policy Group (2007), Migration Integration Policy Index, Brussels.
effects can only be analyzed in a meaningful way at a regional, not at a national level.
22
[5] Cameron, A. Colin and Pravin K. Trivedi (2005), Microeconometrics: Methods and Applications,
Cambridge University Press, New York.
[6] Constant, Amelie F. and Elena D’Agosto (2008), “Where Do the Brainy Italians Go?”, IZA Discussion Paper No. 3325, Bonn.
[7] Defoort, Cécily (2006), “Tendances de long terme des migrations internationales: Analyse à partir
des 6 principaux pays receveurs”, mimeo.
[8] Docquier, Frédéric, Oliver Lohest and Abdesalam Marfouk (2008), “What determines migrants’
destination choice?”, mimeo.
[9] Geis, Wido, Silke Uebelmesser and Martin Werding (2008), “Why go to France or Germany, if you
could as well go to the UK or the US? Selective Features of Immigration to Four Major OECD
Countries ”,CESifo Working Paper No. 2427.
[10] Harris, John R. and Michael P. Todaro (1969), “Migration, Unemployment and Development: A
Two Sector Analysis”, American Economic Review 60(1), 126–142.
[11] Hegre, Håvard and Nicholas Sambanis (2006), “Sensitivity Analysis of Empirical Results on Civil
War Onset”, Journal of Conflict Resolution 50(4), 508–535.
[12] Institute for Education Sciences (2007), Comparative Indicators of Education in the United States
and Other G-8 Countries: 2006, National Center for Education statistics, Washington.
[13] Koller, Barbara (1997); “Aussiedler der großen Zuwanderungswellen – was ist aus ihnen geworden? die Eingliederungssituation von Aussiedlerinnen und Aussiedlern auf dem Arbeitsmarkt in
Deutschland”, Mitteilungen aus der Arbeitsmarkt- und Berufsforschung 30(4), 766–789.
[14] Leamer, Edward E. (1985), “Sensitivity Analyses Would Help”, American Economic Review 75(3),
308–313.
23
[15] Lundborg, Per (1993), “Determinants of Migration in the Nordic Labor Market”, Scandinavian
Journal of Economics 93(3), 363–375.
[16] Mayda, Anna-Maria (2007), “International migration: A panel data analysis of the determinants
of bilateral flows”, Centre for Research and Analysis of Migration, CReAM discussion paper No.
07/07, London.
[17] Munshi, Kavian (2003), “Networks in the Modern Economy: Mexican Migrants in the U.S. Labor
Market”, Quarterly Journal of Economics 118(2), 549–599.
[18] OECD (2004), Employment Outlook: 2004 Edition, Paris.
[19] OECD (2006a), PISA 2006: Science Competencies for Tomorrow’s World, Paris.
[20] OECD (2006b), Taxing Wages 2005–2006, Paris.
[21] OECD (2007a), OECD in Figures: 2007 Edition, Paris.
[22] OECD (2007b), Pensions at a Glance: 2007 Edition, Paris.
[23] Office for National Statistics (2007), LFS User Guide – Volume 9: EuroStat & EuroStat Derived
Variables, Newport.
[24] Office for National Statistics (2000), Standard Occupational Classification 2000, Newport.
[25] Office for National Statistics, Social and Vital Statistics Division and Northern Ireland Statistics
and Research Agency, Central Survey Unit (2005), Quarterly Labour Force Survey, March – May
2005 [computer file], Colchester, Essex [distributor: UK Data Archive, SN: 5211].
[26] Pedersen, Peder J., Mariola Pytlikova and Nina Smith (2005), “Selection or Network Effects? Migration Flows into 27 OECD Countries, 1990-2000”, IZA Discussion Paper No. 1104, Bonn.
24
[27] Sala-i-Martin, Xavier X. (1997), “I Just Ran Two Million Regressions”, American Economic Review
87(2), Papers and Proceedings, 178–183.
[28] Sjaastad, Larry A. (1962), “The Costs and Returns on Human Migration”, Journal of Political
Economy 70(5, Part 2Part 2: Investment in Human Beings.), 80–93.
[29] Schroedter, Julia H., Yvonne Lechert and Paul Lüttinger (2006), “ Die Umsetzung der Bildungsskala
ISCED-1997 für die Volkszählung 1970, die Mikrozensus- Zusatzerhebung 1971 und die Mikrozensen
1976-2004”, ZUMA-Methodenbericht No. 2006/08.
[30] Ruggles, Steven, Matthew Sobek, Trent Alexander, Catherine A. Fitch, Ronald Goeken, Patricia
Kelly Hall, Miriam King, and Chad Ronnander (2004), Integrated Public Use Microdata Series:
Version 3.0 [Machine-readable database], Minnesota Population Center, Minneapolis, MN.
[31] Todaro, Michael P. (1970), “A Model of Labor Migration and Unemployment in Less Developed
Countries”, American Economic Review 59(1), 138–148.
25
26
2.0%
2.4%
not classifiable
Total
Natives
Immigrants
of which from
1. Turkey
2. Russia
3. Poland
4. Italy
5. Serbia and
Montenegro
6. Croatia
7. Greece
8. Bosnia and
Herzegovina
33.7%
2.2%
1.9%
1.9%
260,469
231,670
224,118
4,029,519
12.1%
8.3%
6.4%
3.6%
2.4%
83.2%
16.8%
Share
1,442,949
992,851
762,334
432,790
286,169
Number
71,183,550
59,229,636
11,953,914
Germany
Source: National micro-data sets; authors’ calculations
86,325
102,868
9. Germany
10. UK
4.8%
4.6%
2.5%
202,259
195,768
105,322
6. Tunisia
7. Turkey
8. Poland
91.5%
8.5%
13.7%
12.7%
12.4%
7.4%
6.3%
Share
Number
Total
50,033,805
Natives
45,805,640
Immigrants
4,228,165
of which from
1. Algeria
579,313
2. Portugal
534,994
3. Morocco
525,982
4. Italy
313,420
5. Spain
267,343
France
10. USA
9. Nigeria
6. South Africa
7. Poland
8. Kenya
Total
Natives
Immigrants
of which from
1. India
2. Ireland
3. Pakistan
4. Bangladesh
5. Jamaica
88,639
96,388
113,698
106,089
104,969
422,204
278,612
264,049
179,661
127,322
Number
47,891,659
43,948,682
3,930,175
UK
2.3%
2.5%
2.9%
2.7%
2.7%
10.7%
7.1%
6.7%
4.6%
3.2%
91.8%
8.2%
Share
Table 1: Important immigrant groups (15 years and older)
9. Dominican
Republic
10. UK
6. El Salvador
7. Korea
8. Cuba
Total
Natives
Immigrants
of which from
1. Mexico
2. Philippines
3. India
4. China
5. Vietnam
646,985
672,573
945,666
923,535
872,350
10,136,329
1,521,699
1,337,894
1,115,409
1,038,901
Number
227,783,897
195,049,054
32,734,843
USA
2.0%
2.1%
2.9%
2.8%
2.7%
31.0%
4.6%
4.1%
3.4%
3.2%
85.6%
14.4%
Share
Table 2: Educational achievements of immigrants (25-54)
Immigrants
ISCED 0-1
Number
Share
Participation rate
Unemployment rate
Wage*
ISCED 2
Number
Share
Participation rate
Unemployment rate
Wage*
ISCED 3+4
Number
Share
Participation rate
Unemployment rate
Wage*
ISCED 5+6
Number
Share
Participation rate
Unemployment rate
Wage*
France
Germany
UK
USA
699,323
28.56%
67.98%
19.15%
$12.91
718,828
11.70%
60.36%
26.86%
$13.51
509,257
21.13%
49.95%
9.25%
$12.47
3,884,751
18.27%
73.25%
7.99%
$11.39
512,363
20.92%
76.05%
21.55%
$13.22
1,596,041
25.97%
75.08%
20.65%
$13.42
305,096
12.66%
78.99%
7.65%
$15.98
2,659,406
12.51%
74.26%
7.80%
$12.84
701,190
28.63%
81.39%
17.19%
$14.23
2,547,618
41.46%
84.46%
15.56%
$14.71
880,387
36.53%
84.08%
5.65%
$19.77
7,583,786
35.67%
78.40%
6.26%
$16.38
535,926
21.89%
80.90%
15.81%
$19.56
1,282,602
20.87%
81.55%
12.69%
$20.02
715,139
29.68%
87.75%
5.43%
$26.16
7,132,580
33.55%
81.23%
4.24%
$30.08
Natives
France
Germany
UK
ISCED 0-1
Number
1,613,090
368,143 2,590,481
Share
7.13%
1.24%
11.96%
Participation rate
74.75%
68.14%
61.94%
Unemployment rate
13.18%
29.47%
6.98%
Wage*
$12.87
$9.61
$14.17
ISCED 2
Number
4,478,207
3,003,786 3,905,006
Share
19.78%
10.14%
18.03%
Participation rate
84.92%
79.40%
82.30%
12.10%
18.42%
4.27%
Unemployment rate
$14.44
$13.31
$15.99
Wage*
ISCED 3+4
Number
10,167,941 17,763,323 8,428,241
Share
44.92%
59.96%
38.91%
Participation rate
90.00%
88.62%
88.44%
Unemployment rate
6.90%
9.87%
2.78%
$15.32
$15.30
$18.60
Wage*
ISCED 5+6
Number
6,375,285
8,490,608 6,736,941
Share
28.17%
28.66%
31.10%
Participation rate
91.67%
90.61%
93.08%
Unemployment rate
5.45%
3.92%
1.81%
Wage*
$20.86
$20.87
$25.87
* Hourly wages are derived as described in chapter 4
Source: National micro-data sets; authors’ calculations
USA
27
1,667,184
1.63%
51.41%
13.33%
$14.24
7,655,447
7.47%
67.58%
14.63%
$14.13
53,448,746
52.18%
81.64%
6.39%
$18.75
39,661,288
38.72%
87.84%
3.01%
$30.68
28
PISA-scores in sciences
Share of persons with ISCED 5+6 on the
population in the immigration country
PISA-scores
Share of high skilled
Pension replacement
Tax wedge
Benefit replacement
Union coverage
Employment protection legislation indicator;
Range 0 (not restrictive) – 2 (extremely restrictive)
Share of workers who are covered by collective wage agreements
Benefit replacement rate in the first five years
of unemployment
Income tax wedge (including employer and
employee social security contributions)
Net pension replacement rate
Definition
Immigrant specific unemployment rates
(ILO-definition)
Immigrant specific wages in US$ (PPP) as
derived in chapter 4
Share of persons with the same country of
birth on the population in the immigration
country (0 if share < 0.2%)
Share of foreign born persons on the population in the immigration country
Share of public health expenditures on GDP
Employment protection
Health expenditures
Share of foreigners
Network
Wage
Name
Unemployment
OECD Employment
Outlook: 2004
OECD Employment
Outlook: 2004
OECD
Taxing
Wages 2005-2006
OECD Pensions at a
Glance: 2007
OECD PISA 2006
Own calculations
OECD in Figures:
2007
OECD Employment
Outlook: 2004
489
27.63
30.6
11.7
13.8
14
0.067
6.90
10.30
0.00
Own calculations
Own calculations
11.39
Min. Obs.
4.24
Own calculations
Source
Own calculations
Table 3: Socio-economic and and institutional variables
516
37.93
78.4
52.5
39.4
93
1.000
8.85
16.75
5.66
30.08
Max. Obs.
26.86
child-income group
specific
income group specific
country specific
country specific
country specific
country specific
country specific
country specific
country specific
Type of Variation
gender-education
group specific
gender-education
group specific
individual specific
29
-0.0445***
(0.00015)
0.0269***
(0.00016)
2.286***
(0.0013)
-0.456***
(0.00035)
-0.0474***
(0.00015)
0.0296***
(0.00016)
2.298***
(0.0013)
-0.460***
(0.00035)
0.179***
(0.0012)
0.00858***
(0.000037)
-0.0557***
(0.00015)
0.0200***
(0.00017)
2.256***
(0.0013)
-0.448***
(0.00035)
2.032***
(0.0097)
-0.0154***
(0.00050)
-0.0468***
(0.00037)
0.0203***
(0.00017)
2.254***
(0.0013)
-0.447***
(0.00035)
0.0211***
(0.00012)
-0.0000496***
(0.0000059)
-0.0520***
(0.00040)
0.0202***
(0.00017)
2.256***
(0.0013)
-0.448***
(0.00035)
Log likelihood
-23549781 -23538723
-23522967
-23519230
-23525137
Pseudo R2
0.6292
0.6294
0.6296
0.6297
0.6296
Observations
319431
319431
319431
319431
319431
Standard errors are in parenthesis. Source: Authors’ calculations / estimations.
Share of high skilled
PISA-scores
Pension replacement
Tax wedge
Brr*u
Benefit replacement
Uc*u
Union coverage
Epl*u
Employment protection
Health expenditures
Share of foreigners
Network2
Network
Wage
Unemployment
-23524163
0.6296
319431
0.0641***
(0.00036)
0.0000776***
(0.000017)
-0.0569***
(0.00052)
0.0205***
(0.00017)
2.257***
(0.0013)
-0.448***
(0.00035)
Table 4: Estimation results (complete data set)
-23514460
0.6297
319431
-0.0292***
(0.00011)
-0.0369***
(0.00015)
0.0383***
(0.00017)
2.305***
(0.0013)
-0.462***
(0.00035)
-23541865
0.6293
319431
-0.0144***
(0.00011)
-0.0406***
(0.00015)
0.0195***
(0.00017)
2.280***
(0.0013)
-0.454***
(0.00035)
-23544724
0.6293
319431
0.0224***
(0.00022)
-0.0487***
(0.00015)
0.0237***
(0.00017)
2.280***
(0.0013)
-0.454***
(0.00035)
-0.113***
(0.00059)
-23531623
0.6295
319431
-0.0542***
(0.00015)
0.0198***
(0.00017)
2.261***
(0.0013)
-0.449***
(0.00035)
30
Average
Minimum
Maximum
Unemployment
-0.0520 -0.1094
-0.6310
-0.0200
Wage
0.0209
0.0222
0.0000
0.0402
Network
2.2620
2.2647
2.2450
2.3140
Network2
-0.4510 -0.4508
-0.4650
-0.4430
Share of foreigners
0.2990
0.2569
-1.3050
0.6660
Health expenditures
0.0114
0.0522
-0.5060
1.0270
Employment protection 16.6100 29.2114
2.0230
135.1000
Epl*u
-0.0370
-0.6595
-2.0710
-0.0150
Union coverage
0.0220
-0.3094
-3.9830
0.4510
Uc*u
-0.0001
0.0009
-0.0240
0.0247
Benefit replacement
0.0479
-0.2976
-7.6760
9.3380
Brr*u
0.0004
0.0202
0.0000
0.0737
Tax wedge
-0.0460 -0.0427
-0.0480
-0.0280
Pension replacement
-0.0110 -0.0107
-0.0170
-0.0040
PISA-scores
0.0076
0.0272
-0.5960
0.4780
Share of high skilled
-0.0240
-0.0522
-7.5210
2.1520
Bold numbers are significant by the extreme bound criterion
Source: Authors’ calculations / estimations.
Median
Cross
variance
0.0235
0.0001
0.0002
0.0000
0.0779
0.0527
1052.4663
0.6633
0.5529
0.0002
5.1093
0.0007
0.0001
0.0000
0.0271
1.9072
Within standard error
0.0005
0.0002
0.0013
0.0004
0.0015
0.0005
0.0878
0.0021
0.0014
0.0000
0.0040
0.0001
0.0001
0.0001
0.0005
0.0023
Total standard error
0.1533
0.0088
0.0124
0.0038
0.2791
0.2296
32.4419
0.8144
0.7435
0.0130
2.2604
0.0271
0.0073
0.0029
0.1647
1.3810
Number of
regressions
130
130
130
130
37
37
37
37
37
37
37
37
37
37
37
37
Table 5: Aggregated estimation results (complete data set)
Lower extreme
bound
-0.6344
0.0026
2.2424
-0.4657
-1.3214
-0.5090
2.0030
-2.0820
-4.0030
-0.0241
-7.7420
0.0000
-0.0482
-0.0173
-0.6012
-7.5590
Upper extreme
bound
-0.0188
0.0405
2.3166
-0.4423
0.6694
1.0356
136.1800
-0.0140
0.4538
0.0248
9.3860
0.0741
-0.0278
-0.0038
0.4804
2.1698
31
Average
Minimum
Maximum
Unemployment
-0.0520
-0.1122
-0.6660
0.0080
Wage
0.0176
0.0178
-0.0060
0.0395
Network
2.8650
2.8674
2.8480
2.9110
Network2
-0.5710 -0.5702
-0.5840
-0.5610
Share of foreigners
0.6050
0.5835
0.2280
0.9760
Health expenditures
0.0183
0.1159
-0.9800
2.1490
Employment protection 17.5300 37.9240
-21.0900
146.6000
Epl*u
-0.0490 -0.8651
-2.7820
-0.0180
Union coverage
-0.1020
-0.5433
-5.8790
0.5390
Uc*u
-0.0002
0.0032
-0.0240
0.0328
Benefit replacement
0.1490
-0.1896
-16.0900
13.9900
Brr*u
0.0002
0.0198
-0.0010
0.0729
Tax wedge
-0.0650 -0.0599
-0.0690
-0.0390
Pension replacement
-0.0170 -0.0177
-0.0280
-0.0070
PISA-scores
-0.0440
-0.0107
-1.3180
0.7050
B old numbers are significant by the extreme bound criterion
Source: Authors’ calculations / estimations.
Median
Cross
variance
0.0266
0.0001
0.0002
0.0000
0.0216
0.1974
1822.6216
1.2282
1.1238
0.0003
15.0076
0.0008
0.0001
0.0000
0.1030
Within standard error
0.0006
0.0002
0.0017
0.0004
0.0022
0.0008
0.1179
0.0026
0.0020
0.0000
0.0056
0.0001
0.0002
0.0002
0.0007
Total standard error
0.1630
0.0108
0.0125
0.0044
0.1470
0.4443
42.6923
1.1082
1.0601
0.0163
3.8740
0.0277
0.0102
0.0056
0.3210
Number of
regressions
130
130
130
130
37
37
37
37
37
37
37
37
37
37
37
Table 6: Aggregated estimation results; people immigrated after 1995
Lower extreme
bound
-0.6692
-0.0064
2.8446
-0.5849
0.2040
-0.9842
-21.4300
-2.7946
-5.9070
-0.0242
-16.1860
-0.0010
-0.0693
-0.0283
-1.3256
Upper extreme
bound
0.0095
0.0399
2.9144
-0.5601
0.9802
2.1616
147.1600
-0.0168
0.5420
0.0330
14.0580
0.0734
-0.0387
-0.0067
0.7094
Table 7: Median elasticities (complete data set)
1% increase in
unemployment
rate in
1% increase in
wage per hour in
1% increase in
Network in
1% increase in
share of foreign
born in
1% increase in
state health
expenditures
in % of GDP in
1% increase in
employment
protection
indicator in
US
Germany
UK
France
Change in
Prob(US)
-0.138
0.543
0.205
0.571
Change in
Prob(GE)
0.071
-0.715
0.065
0.187
Change in
Prob(UK)
0.024
0.062
-0.310
0.064
Change in
Prob(FR)
0.043
0.110
0.040
-0.822
Average
value
6.92%
17.59%
6.64%
18.50%
US
Germany
UK
France
Change in
Prob(US)
0.134
-0.198
-0.260
-0.200
Change in
Prob(GE)
-0.071
0.251
-0.078
-0.059
Change in
Prob(UK)
-0.027
-0.022
0.378
-0.022
Change in
Prob(FR)
-0.036
-0.032
-0.040
0.281
Average
value
$19.18
$15.34
$20.21
$15.52
US
Germany
UK
France
Change in
Prob(US)
0.016
-0.026
-0.074
-0.029
Change in
Prob(GE)
-0.004
0.053
-0.011
-0.043
Change in
Prob(UK)
-0.009
-0.006
0.091
-0.014
Change in
Prob(FR)
-0.003
-0.022
-0.006
0.087
Average
value
0.90%
0.55%
0.06%
0.11%
US
Germany
UK
France
Change in
Prob(US)
1.658
-3.007
-1.849
-2.095
Change in
Prob(GE)
-0.863
3.869
-0.575
-0.651
Change in
Prob(UK)
-0.310
-0.336
2.747
-0.234
Change in
Prob(FR)
-0.486
-0.526
-0.324
2.980
Average
value
15.46%
16.75%
10.30%
11.67%
US
Germany
UK
France
Change in
Prob(US)
2.827
-5.644
-4.959
-6.076
Change in
Prob(GE)
-1.471
7.262
-1.541
-1.888
Change in
Prob(UK)
-0.528
-0.629
7.368
-0.678
Change in
Prob(FR)
-0.829
-0.988
-0.868
8.642
Average
value
6.90%
8.23%
7.23%
8.86%
US
Germany
UK
France
Change in
Prob(US)
0.398
-7.824
-2.329
-9.980
Change in
Prob(GE)
-0.207
10.072
-0.724
-3.104
Change in
Prob(UK)
-0.075
-0.880
3.461
-1.122
Change in
Prob(FR)
-0.116
-1.369
-0.408
14.206
Average
value
0.066
0.784
0.233
1.000
32
Table 7 (continued)
1% increase in
union coverage in
1% increase in
benefit replacement
rate in
1% increase in
tax wedge in
1% increase in
pension replacement
rate in
1% increase in
PISA-score in
1% increase in
share of high skilled
persons (ISCED 5+6)
in
US
Germany
UK
France
Change in
Prob(US)
0.111
-0.900
-0.437
-1.231
Change in
Prob(GE)
-0.058
1.159
-0.136
-0.382
Change in
Prob(UK)
-0.021
-0.101
0.649
-0.138
Change in
Prob(FR)
-0.032
-0.158
-0.077
1.751
Average
value
14%
68%
33%
93%
US
Germany
UK
France
Change in
Prob(US)
0.237
-0.841
-0.469
-1.134
Change in
Prob(GE)
-0.124
1.082
-0.146
-0.353
Change in
Prob(UK)
-0.044
-0.094
0.697
-0.127
Change in
Prob(FR)
-0.070
-0.147
-0.082
1.614
Average
value
13.8%
29.2%
16.3%
39.4%
US
Germany
UK
France
Change in
Prob(US)
-0.387
1.262
0.866
1.300
Change in
Prob(GE)
0.197
-1.640
0.272
0.406
Change in
Prob(UK)
0.071
0.145
-1.293
0.147
Change in
Prob(FR)
0.120
0.233
0.154
-1.854
Average
value
22.17%
45.73%
31.23%
46.77%
US
Germany
UK
France
Change in
Prob(US)
-0.249
0.400
0.316
0.445
Change in
Prob(GE)
0.128
-0.515
0.099
0.139
Change in
Prob(UK)
0.046
0.045
-0.474
0.050
Change in
Prob(FR)
0.075
0.070
0.060
-0.633
Average
value
58.6%
57.3%
45.7%
63.9%
US
Germany
UK
France
Change in
Prob(US)
1.349
-2.382
-2.377
-2.285
Change in
Prob(GE)
-0.702
3.064
-0.739
-0.710
Change in
Prob(UK)
-0.252
-0.266
3.532
-0.255
Change in
Prob(FR)
-0.395
-0.417
-0.416
3.250
Average
value
489
516
515
495
US
Germany
UK
France
Change in
Prob(US)
-0.329
0.400
0.445
0.401
Change in
Prob(GE)
0.170
-0.516
0.138
0.124
Change in
Prob(UK)
0.062
0.045
-0.661
0.045
Change in
Prob(FR)
0.097
0.070
0.078
-0.571
Average
value
37.93%
27.63%
30.70%
27.68%
Source: Authors’ calculations / estimations.
33
34
-0.0170
0.0465
1.9260
-0.4280
0.7350
0.0057
15.2700
-0.0470
-0.0310
-0.0400
0.0129
0.0793
-0.0920
Median
-0.0085
0.0428
2.9825
-0.5710
-0.0330
0.0094
-2.1290
0.0056
0.1330
-0.0520
-0.0350
-0.0560
0.1950
Unemployment
Wage
Network
Network2
Share of foreigners
Health expenditures
Employment protection
Union coverage
Benefit replacement
Tax wedge
Pension replacement
PISA-scores
Share of high skilled
Low skilled
Unemployment
Wage
Network
Network2
Share of foreigners
Health expenditures
Employment protection
Union coverage
Benefit replacement
Tax wedge
Pension replacement
PISA-scores
Share of high skilled
-0.0106
0.0154
2.9867
-0.5722
0.0119
0.0079
-6.6864
0.0237
0.2431
-0.0513
-0.0362
-0.0554
0.1543
Average
-0.0196
0.0472
1.9279
-0.4290
0.7145
0.1259
22.8702
-0.4071
-0.5239
-0.0398
0.0349
0.1013
0.1238
Average
-0.0260
-0.1240
2.9780
-0.5820
-0.3270
-0.1260
-98.3600
-0.4790
-1.3680
-0.0580
-0.0530
-0.0990
-0.2470
Minimum
-0.4050
-0.0270
1.9180
-0.4340
0.5250
-0.7260
-3.4900
-4.9250
-18.7100
-0.0420
-0.1830
-1.4810
-8.3800
Minimum
0.0001
0.1160
3.0190
-0.5690
1.6340
0.1870
4.0350
1.2400
0.9550
-0.0390
-0.0270
0.0184
1.3140
Maximum
0.3560
0.1200
1.9440
-0.4260
0.7620
2.4840
87.7200
0.5710
11.7800
-0.0350
0.2320
0.8530
5.0600
Maximum
B old numbers are significant by the extreme bound criterion
Source: Authors’ calculations / estimations.
Median
Qualified
Cross
variance
0.0001
0.0041
0.0001
0.0000
0.0826
0.0055
289.6025
0.0678
0.1582
0.0000
0.0000
0.0004
0.0745
Cross
variance
0.0206
0.0026
0.0000
0.0000
0.0026
0.2404
679.3646
0.8299
15.5632
0.0000
0.0131
0.1253
3.8498
Within standard error
0.0003
0.0014
0.0025
0.0006
0.0030
0.0011
0.1321
0.0023
0.0069
0.0002
0.0005
0.0009
0.0046
Within standard error
0.0011
0.0006
0.0016
0.0004
0.0036
0.0014
0.1502
0.0031
0.0088
0.0002
0.0011
0.0012
0.0061
Total standard error
0.0083
0.0637
0.0090
0.0032
0.2874
0.0740
17.0182
0.2605
0.3978
0.0051
0.0066
0.0203
0.2730
Total standard error
0.1437
0.0514
0.0050
0.0016
0.0511
0.4903
26.0651
0.9110
3.9450
0.0016
0.1146
0.3539
1.9621
Number of
regressions
130
130
130
130
37
37
37
37
37
37
37
37
37
Number of
regressions
130
130
130
130
37
37
37
37
37
37
37
37
37
Table 8: Aggregated estimation results (by skill levels)
Lower extreme
bound
-0.0266
-0.1280
2.9730
-0.5833
-0.3348
-0.1438
-100.1000
-0.4874
-1.4560
-0.0585
-0.0540
-0.1054
-0.2830
Lower extreme
bound
-0.4094
-0.0279
1.9148
-0.43486
0.519
-0.7346
-3.548
-4.967
-18.886
-0.04232
-0.1848
-1.4956
-8.45
Upper extreme
bound
0.0007
0.1184
3.0240
-0.5677
1.6640
0.1904
4.1030
1.2620
1.0890
-0.0386
-0.0260
0.0288
1.3780
Upper extreme
bound
0.3606
0.12128
1.9472
-0.42516
0.782
2.508
88.58
0.5766
11.882
-0.0347
0.2352
0.8604
5.108
Table A1: Median elasticities; people immigrated after 1995
1% increase in
unemployment rate in
US
Germany
UK
France
Change in
Prob(US)
-0.150
0.520
0.196
0.547
1% increase in
wage per hour in
US
Germany
UK
France
Change in
Prob(US)
0.117
-0.156
-0.206
-0.158
Change in
Prob(GE)
-0.062
0.204
-0.069
-0.052
Change in
Prob(UK)
-0.020
-0.016
0.314
-0.016
Change in
Prob(FR)
-0.035
-0.031
-0.040
0.226
1% increase in
share of foreign born in
US
Germany
UK
France
Change in
Prob(US)
3.561
-5.761
-3.543
-4.014
Change in
Prob(GE)
-1.861
7.603
-1.240
-1.405
Change in
Prob(UK)
-0.546
-0.592
5.551
-0.412
Change in
Prob(FR)
-1.156
-1.253
-0.770
5.831
US
Germany
UK
France
Change in
Prob(US)
4.822
-8.573
-7.532
-9.229
Change in
Prob(GE)
-2.509
11.332
-2.628
-3.220
Change in
Prob(UK)
-0.742
-0.885
11.807
-0.953
Change in
Prob(FR)
-1.571
-1.874
-1.646
13.402
US
Germany
UK
France
Change in
Prob(US)
0.446
-7.808
-2.324
-9.960
Change in
Prob(GE)
-0.232
10.318
-0.813
-3.483
Change in
Prob(UK)
-0.069
-0.807
3.644
-1.029
Change in
Prob(FR)
-0.145
-1.703
-0.507
14.472
1% increase in
union coverage in
US
Germany
UK
France
Change in
Prob(US)
-0.544
3.947
1.915
5.398
Change in
Prob(GE)
0.283
-5.216
0.667
1.879
Change in
Prob(UK)
0.083
0.405
-3.001
0.553
Change in
Prob(FR)
0.178
0.864
0.419
-7.830
1% increase in
benefit replacement
rate in
US
Germany
UK
France
Change in
Prob(US)
0.784
-2.474
-1.381
-3.338
Change in
Prob(GE)
-0.408
3.270
-0.482
-1.164
Change in
Prob(UK)
-0.121
-0.255
2.165
-0.344
Change in
Prob(FR)
-0.256
-0.541
-0.302
4.847
1% increase in
tax wedge in
US
Germany
UK
France
Change in
Prob(US)
-0.589
1.670
1.148
1.721
Change in
Prob(GE)
0.300
-2.236
0.407
0.607
Change in
Prob(UK)
0.087
0.177
-1.810
0.179
Change in
Prob(FR)
0.202
0.389
0.255
-2.508
US
Germany
UK
France
Change in
Prob(US)
-0.402
0.572
0.452
0.636
Change in
Prob(GE)
0.207
-0.756
0.160
0.223
Change in
Prob(UK)
0.061
0.059
-0.719
0.066
Change in
Prob(FR)
0.134
0.124
0.107
-0.926
1% increase in
Pisa score in
US
Germany
UK
France
Change in
Prob(US)
-8.233
12.984
12.959
12.456
Change in
Prob(GE)
4.299
-17.135
4.528
4.352
Change in
Prob(UK)
1.251
1.320
-20.313
1.266
Change in
Prob(FR)
2.683
2.831
2.826
-18.074
1% increase in
share of high skilled persons (ISCED 5+6) in
US
Germany
UK
France
Change in
Prob(US)
4.289
-4.670
-5.189
-4.678
Change in
Prob(GE)
-2.240
6.163
-1.813
-1.635
Change in
Prob(UK)
-0.652
-0.475
8.133
-0.476
Change in
Prob(FR)
-1.398
-1.018
-1.131
6.788
1% increase in
Network in
US
Germany
UK
France
Change in
Prob(US)
0.016
-0.033
-0.079
-0.037
Change in
Prob(GE)
-0.003
0.073
-0.012
-0.063
Change in
Prob(UK)
-0.009
-0.006
0.099
-0.015
Change in
Prob(FR)
-0.004
-0.034
-0.007
0.114
1% increase in
state health expenditures
in % of GDP in
1% increase in
employment protection
indicator in
1% increase in
pension replacement
rate in
Source: Authors’ calculations / estimations.
35
Change in
Prob(GE)
0.076
-0.707
0.070
0.202
Change in
Prob(UK)
0.022
0.055
-0.314
0.056
Change in
Prob(FR)
0.052
0.133
0.048
-0.805
Table A2: Median elasticities; qualified immigrants
1% increase in
unemployment rate in
US
Germany
UK
France
Change in
Prob(US)
-0.026
0.136
0.053
0.164
1% increase in
wage per hour in
US
Germany
UK
France
Change in
Prob(US)
0.293
-0.455
-0.623
-0.458
Change in
Prob(GE)
-0.159
0.553
-0.164
-0.120
Change in
Prob(UK)
-0.067
-0.049
0.854
-0.049
Change in
Prob(FR)
-0.067
-0.049
-0.067
0.627
1% increase in
share of foreign born in
US
Germany
UK
France
Change in
Prob(US)
3.655
-8.052
-4.951
-5.610
Change in
Prob(GE)
-2.039
9.802
-1.359
-1.539
Change in
Prob(UK)
-0.800
-0.867
6.853
-0.604
Change in
Prob(FR)
-0.815
-0.883
-0.543
7.753
US
Germany
UK
France
Change in
Prob(US)
1.208
-2.930
-2.574
-3.154
Change in
Prob(GE)
-0.709
3.567
-0.743
-0.910
Change in
Prob(UK)
-0.265
-0.315
3.563
-0.340
Change in
Prob(FR)
-0.270
-0.321
-0.282
4.359
US
Germany
UK
France
Change in
Prob(US)
0.328
-7.832
-2.331
-9.990
Change in
Prob(GE)
-0.183
9.537
-0.639
-2.740
Change in
Prob(UK)
-0.072
-0.846
3.226
-1.079
Change in
Prob(FR)
-0.073
-0.860
-0.256
13.810
1% increase in
union coverage in
US
Germany
UK
France
Change in
Prob(US)
-0.212
2.090
1.014
2.858
Change in
Prob(GE)
0.118
-2.545
0.278
0.784
Change in
Prob(UK)
0.046
0.226
-1.404
0.309
Change in
Prob(FR)
0.047
0.230
0.111
-3.951
1% increase in
benefit replacement
rate in
US
Germany
UK
France
Change in
Prob(US)
-0.138
0.594
0.332
0.802
Change in
Prob(GE)
0.077
-0.724
0.091
0.220
Change in
Prob(UK)
0.030
0.064
-0.459
0.087
Change in
Prob(FR)
0.031
0.065
0.036
-1.107
1% increase in
tax wedge in
US
Germany
UK
France
Change in
Prob(US)
-0.305
1.221
0.846
1.261
Change in
Prob(GE)
0.169
-1.496
0.236
0.351
Change in
Prob(UK)
0.066
0.135
-1.178
0.138
Change in
Prob(FR)
0.071
0.141
0.096
-1.750
US
Germany
UK
France
Change in
Prob(US)
0.228
-0.488
-0.343
-0.533
Change in
Prob(GE)
-0.128
0.594
-0.096
-0.147
Change in
Prob(UK)
-0.050
-0.053
0.478
-0.058
Change in
Prob(FR)
-0.050
-0.053
-0.038
0.738
1% increase in
Pisa scores in
US
Germany
UK
France
Change in
Prob(US)
12.484
-26.774
-26.722
-25.684
Change in
Prob(GE)
-6.954
32.610
-7.324
-7.039
Change in
Prob(UK)
-2.742
-2.893
36.983
-2.775
Change in
Prob(FR)
-2.789
-2.943
-2.937
35.499
1% increase in
Network in
US
Germany
UK
France
Change in
Prob(US)
0.011
-0.027
-0.087
-0.029
Change in
Prob(GE)
-0.003
0.040
-0.011
-0.023
Change in
Prob(UK)
-0.006
-0.004
0.103
-0.012
Change in
Prob(FR)
-0.002
-0.008
-0.005
0.064
1% increase in
share of high skilled
graduates (ISCED 5+6) in
US
Germany
UK
France
Change in
Prob(US)
-1.127
1.669
1.854
1.672
Change in
Prob(GE)
0.629
-2.031
0.509
0.459
Change in
Prob(UK)
0.247
0.180
-2.566
0.180
Change in
Prob(FR)
0.251
0.183
0.203
-2.310
1% increase in
state health expenditures in
relative to GDP
1% increase in
employment protection
indicator in
1% increase in
pension replacement
rate in
Change in
Prob(GE)
0.018
-0.166
0.018
0.055
Change in
Prob(UK)
0.006
0.015
-0.073
0.017
Change in
Prob(FR)
0.006
0.015
0.006
-0.226
Source: Authors’ calculations / estimations.
36
Table A3: Median elasticities; low skilled immigrants
1% increase in
unemployment rate in
US
Germany
UK
France
Change in
Prob(US)
-0.038
0.116
0.042
0.099
1% increase in
wage per hour in
US
Germany
UK
France
Change in
Prob(US)
0.222
-0.292
-0.340
-0.296
Change in
Prob(GE)
-0.106
0.416
-0.139
-0.111
Change in
Prob(UK)
-0.033
-0.035
0.583
-0.035
Change in
Prob(FR)
-0.083
-0.089
-0.104
0.443
1% increase in
share of foreign born in
US
Germany
UK
France
Change in
Prob(US)
-0.225
0.300
0.184
0.209
Change in
Prob(GE)
0.107
-0.428
0.071
0.081
Change in
Prob(UK)
0.033
0.036
-0.312
0.025
Change in
Prob(FR)
0.085
0.092
0.057
-0.315
US
Germany
UK
France
Change in
Prob(US)
2.856
-4.180
-3.673
-4.500
Change in
Prob(GE)
-1.354
5.971
-1.419
-1.738
Change in
Prob(UK)
-0.421
-0.502
6.223
-0.541
Change in
Prob(FR)
-1.081
-1.288
-1.132
6.779
US
Germany
UK
France
Change in
Prob(US)
-0.062
0.893
0.266
1.140
Change in
Prob(GE)
0.029
-1.276
0.103
0.440
Change in
Prob(UK)
0.009
0.107
-0.451
0.137
Change in
Prob(FR)
0.023
0.276
0.082
-1.716
1% increase in
union coverage in
US
Germany
UK
France
Change in
Prob(US)
0.034
-0.205
-0.099
-0.280
Change in
Prob(GE)
-0.016
0.293
-0.038
-0.108
Change in
Prob(UK)
-0.005
-0.025
0.168
-0.034
Change in
Prob(FR)
-0.013
-0.063
-0.031
0.422
1% increase in
benefit replacement
rate in
US
Germany
UK
France
Change in
Prob(US)
0.798
-2.074
-1.158
-2.798
Change in
Prob(GE)
-0.378
2.962
-0.447
-1.079
Change in
Prob(UK)
-0.117
-0.248
1.961
-0.335
Change in
Prob(FR)
-0.302
-0.640
-0.357
4.213
1% increase in
tax wedge in
US
Germany
UK
France
Change in
Prob(US)
-0.532
1.255
0.844
1.285
Change in
Prob(GE)
0.242
-1.822
0.331
0.495
Change in
Prob(UK)
0.074
0.152
-1.442
0.154
Change in
Prob(FR)
0.216
0.415
0.267
-1.934
US
Germany
UK
France
Change in
Prob(US)
-1.005
1.066
1.030
1.225
Change in
Prob(GE)
0.472
-1.523
0.373
0.473
Change in
Prob(UK)
0.149
0.128
-1.720
0.147
Change in
Prob(FR)
0.384
0.329
0.316
-1.845
1% increase in
Pisa scores in
US
Germany
UK
France
Change in
Prob(US)
-11.980
15.531
15.500
14.898
Change in
Prob(GE)
5.677
-22.181
5.979
5.747
Change in
Prob(UK)
1.763
1.861
-26.260
1.785
Change in
Prob(FR)
4.540
4.790
4.781
-22.430
1% increase in
Network in
US
Germany
UK
France
Change in
Prob(US)
0.019
-0.021
-0.036
-0.031
Change in
Prob(GE)
-0.005
0.084
-0.010
-0.092
Change in
Prob(UK)
-0.011
-0.008
0.054
-0.018
Change in
Prob(FR)
-0.004
-0.055
-0.008
0.140
1% increase in
share of high skilled
graduates (ISCED 5+6) in
US
Germany
UK
France
Change in
Prob(US)
3.229
-2.890
-3.211
-2.895
Change in
Prob(GE)
-1.531
4.127
-1.239
-1.117
Change in
Prob(UK)
-0.475
-0.346
5.441
-0.347
Change in
Prob(FR)
-1.223
-0.891
-0.990
4.359
1% increase in
state health expenditures in
relative to GDP
1% increase in
employment protection
indicator in
1% increase in
pension replacement
rate in
Change in
Prob(GE)
0.018
-0.166
0.016
0.042
Change in
Prob(UK)
0.005
0.014
-0.070
0.012
Change in
Prob(FR)
0.014
0.036
0.013
-0.153
Source: Authors’ calculations / estimations.
37