• No results found

Do language differences have a significant effect on labour migration within the EU?

N/A
N/A
Protected

Academic year: 2021

Share "Do language differences have a significant effect on labour migration within the EU?"

Copied!
25
0
0

Bezig met laden.... (Bekijk nu de volledige tekst)

Hele tekst

(1)

1

Do language differences have a significant effect on

labour migration within the EU?

Bachelor Thesis

Author: Gijs van Benthem Student number: 10583068

Supervisor: Ed Westerhout

(2)

2 Statement of originality

This document is written by Gijs van Benthem, who declares to take full responsibility for the contents of this document. I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it. The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.

(3)

3 Abstract

Language is proven to have a significant effect on worldwide migration flows. However, it is unclear if this is also the case in common markets, such as the EU. This thesis aims to find out whether linguistic differences have a significant effect on migration flows within the EU. In this thesis, it is hypothesized that linguistic differences do have a negative effect on migration flows. This hypothesis was tested by analyzing migration flows between eight EU countries, a total of 54 observations. All the results indicate that the bigger the difference between two languages, the less migration takes place. However, the insignificance of the results still suggests that linguistic differences don’t influence migration flows within the EU.

(4)

4 Contents

1. Introduction 5

2. Facts and figures 6

3. Literature Review 8

3.1 Advantages of labour mobility 8

3.2 Lack of labour mobility 9

3.3 Linguistic differences 10

4. Relation between language and migration using a simple regression 13

4.1 Model 13

4.2 Data construction 13

4.3 Results 14

5. A model of international migration 15

5.1 Model 15

5.2 Data construction 17

5.3 Results 18

5.4 Leaving out insignificant control variables 20 5.5 Transformation of the dependent variable 𝑚𝑖𝑗 21

6. Conclusion 22

7. Discussion 22

(5)

5 1. Introduction

A common European market aims for unobstructed cross-border movements of goods, services, capital and labour. These freedoms were mostly established with the completion of the single European market in 1993. Freedom of labour movement had already been

achieved substantially longer, since 1968. Although free labour movement has been the case in the European Union for a long time, it has not led to an increase in the mutual exchange of workers (Tassinopoulos, Werner & Kristensen, 1998). The European Union aims to stimulate labour mobility within the EU (Eurofound, 2014). According to this article, there are still labour and skill shortages in some European countries and regions. The European Commission has maintained its focus on geographical mobility and efficient allocation of labour within the EU (Eurofound, 2014). Moreover, this article mentions that increased labour mobility within the EU is a key driver of future growth and an important mechanism for distorted labour markets. The lack of labour movement between members of the European Union is often critiqued with the comment that workers should go to where they are most productive and thus where they can obtain the highest wages. An efficient

distribution of workers would optimize an economy, according to Tassinopoulos, Werner and Kristensen (1998). In their Eurofound publication, Riso, Secher and Andersen (2014) state that Labour migration rates remain lower than expected and this is interpreted as a market failure of the European labour market by many.

The intra-EU labour mobility rate is often compared to that of the United States. When you compare labour mobility rates between these two regions, you find that the European Union has a notably low labour mobility rate, according to Riso, Secher and Andersen (2014). Their paper also states that the main differences between the European Union and the United States are languages, cultures and labour legislation.

The aim of this paper is to look at the effect of language differences on labour migration between European Union members. Recently, Adera and Pytlikova (2015)

investigated how linguistic differences are related to migration flows around the world. They found that the more first official languages differ, the less migration takes place between the two countries. They suggest that language differences itself affect an individual’s migration cost, which influences the probability of migrating to that particular country.

(6)

6 In this paper, the effects of linguistic differences in the European Union on labour migration will be examined. The results of this analysis will contribute to answering the following research question of this article: Do language differences have a significant effect on labour migration within the EU?

In this paper, linguistic differences and labour migration flows between eight European Union members are examined. After which multiple regressions and t-tests are done to determine whether the effect of language differences on labour migration flows are significant.

This paper starts with a section that is dedicated to examining facts and figures to observe namely the level of labour mobility and migration within the EU, compared to other regoins. This will be followed up by a literature review. In this section, previous research on mainly labour migration and linguistic differences are examined. After which simple and multiple regressions are done to determine whether linguistic differences do have a

significant effect on migration flows within the EU. Based on the results that are found in this paper, a conclusion is formed. In the last section, limitations of this article will be mentioned combined with suggestions for further research.

The purpose of this thesis is to give possible insight as to why the labour migration between European Union members is relatively low. And what the effects of language differences are in a common market with free labour movement, such as the European Union.

2. Facts and Figures

Migration within the European Union seems to be low compared to other regions with a common market. In this section, facts and figures will be displayed to analyze and confirm whether labour migration in the European Union is actually relatively low.

First off, a comparison of intra-EU labour mobility with labour mobility in other regions.

(7)

7 Figure 1. Annual cross-border and within-region mobility as a percentage of total population,

2010.

Figure 1 displays the relatively low labour mobility rate within the EU, compared to other regions. The source for this figure is Eurofound (2014) and they used data from OECD (2012) to construct this figure. The labour migration rate within the European Union is often

compared to that of the United States. As can be seen in Figure 1, the U.S. has a labour mobility rate that is eight times as high as the labour mobility rate within the whole European Union.

A detailed analysis of work-related geographical mobility within the European Union was done by Eurofound in 2014. According to this publication, data show that EU27

nationals employed in other EU countries represented 3.1% of the total working population in 2012. In comparison, the amount of third-country nationals being employed in the EU in the same year was 4% (Eurofound, 2014). This article states that although EU citizens can take advantage of the freedom of movement within the EU and third-country nationals are confronted with many obstacles when trying to access the EU labour market, the share of third-country nationals in the EU workforce is substantially larger than the share of EU citizens working outside their origin country.

0,0% 0,5% 1,0% 1,5% 2,0% 2,5% 3,0%

EU15 EU27 US: across 50 states US: across 4 main regions Canada: across 10 provinces Australia: across 8 states

(8)

8 To conclude, data seem to suggest that labour mobility within the EU is relatively low in comparison to other regions. Even when comparing with the share of third-country nationals working in the EU, the share of EU citizens working outside their home country is relatively low. In the next part, a literature review is done to examine whether stimulating the labour migration within the EU is advantageous and what might be causing this low labour mobility rate.

3. Literature review

The theoretical framework will be analyzed and explained in this part of the paper. A few topics will be discussed in this section. Firstly, the advantages of complete labour mobility are analyzed. After which the low migration rate within the EU and possible reasons why this might be the case are discussed. Lastly, different ways to determine linguistic differences are examined.

3.1 Advantages of labour mobility

This paper is based on the premise that labour migration is advantageous, from an

economical point of view. Previous research has shown that this is indeed the case. First off, the European Commission (2014) stated that free movement of labour brings benefits to both workers and employers. Free labour movement bring new job opportunities for individuals. It also helps workers to improve their skills and gain working experience. Moreover, labour mobility helps to address the labour shortages and skill gaps. Also, free movement of labour can help boost competitiveness in the labour market. These are some of the major benefits of promoting and stimulating free labour movement, according to the European Commission (2014).

A lot of research on the efficiency gains of free labour movement has been done. A few articles that analyzed the effect of increasing labour mobility are examined in this section. Firstly, Hamilton and Whalley (1982) did a paper about possible worldwide impacts from the removal of immigration controls. In their paper, they calculated the efficiency gains from liberalized global labour migration by comparing labour migration policies and Gross National Product (GNP) of certain regions. Their results clearly suggest that a free worldwide labour market would have great potential benefits for the worldwide GNP, suggesting that less labour migration restrictions would be more efficient. However, many assumptions are

(9)

9 made in this paper and the authors suggest that their results are not a forecast, but rather suggest policy implications on labour mobility.

Secondly, a more recent and a more sophisticated study on efficiency gains from labour migration was done by Docquier, Machado and Sekkat (2015). They quantified the effect of a complete liberalization of cross-border migration on the world Gross Domestic Product (GDP) and its distribution across regions. They constructed a model to simulate a complete liberalization of labor mobility. They found that, according to their constructed model, 7-18 percent world GDP would be gained by making labour migration unrestricted.

The European Commission suggests that increasing labour mobility has many advantages for the EU and previous research seems to suggest that increased labour mobility and less international labour migration restrictions would increase the GDP of that region. However, many estimates have been made using simplified assumptions, inaccurate data and simulation models, making the predictions less credible. Though, according to existing literature there seems to be a clear relation between increase in labour mobility and an increase in welfare in the EU.

3.2 Lack of labour mobility

Economic theory suggests that in a perfect market, workers would move to search for jobs with higher wages and firms move in search of the lowest wages (Zimmermann, 1995). However, in reality not many move from one European Union country to another for work related reasons. Free movement of labour between European Union members is one of the main objectives of the European Union. Still, labour mobility in the EU is low, especially when it is compared to the labour mobility in the United States and Australia (Van Dalen and Henkens, 2012). Riso, Sercher and Andersen (2014) state that even though the European Union conducts policies to stimulate labour movement within the EU, European and national data suggest that the level of labour migration and the level of mobility remains low and stable.

Literature on this topic suggests that labour migration between European Union

members is comparatively low. An important question Van Dalen and Henkens (2012) raise is what factors are the cause of these low labour mobility rates. They suggest that the lack of international labour migration in the European Union is caused by economic, social and

(10)

10 psychological factors. It is mentioned in this paper that language differences partly explain why so few people migrate between European Union countries. Studies have shown that a lack of proficiency in the native language of the destination country creates many barriers to integrate, which may prevent workers from moving to another country (Riso, Secher & Andersen, 2014). According to Riso, Secher and Andersen (2014), empirical evidence suggests that migration flows between countries with similar languages tend to be larger than between countries with more unrelated languages. In the next sub-section linguistic differences are analyzed and explained in more depth.

3.3 Linguistic differences

Many articles suggest that the more languages differ, the harder it is to acquire the skill to perceive and comprehend that language. Chriswick and Miller (2005), for instance,

developed a measurement system based on the ability of Americans to learn a variety of languages in fixed periods of time. They found that the greater the distance between these languages and English, the lower the scores were on the standardized proficiency tests. A more recent study done by Isphording and Otten (2011) also found that linguistic distance has a significant negative effect on the language acquisition than immigrants with a closer linguistic background. In their article, four different methods of determining linguistic differences were used and they all resulted in the same outcome, being that the linguistic distance between languages is negatively correlated with the ability to master the language of the destination country. Adera and Pytlikova (2015) found a negative relation between linguistic differences and migration flows in a worldwide empirical study. They argued that the harder it is to acquire the skill to perceive and comprehend the language of the

destination country, the bigger the migration cost will be. The higher migration cost will lead to a decrease in the probability of the migrant to migrate to that particular country (Adera and Pylikova, 2015).

So previous research seems to indicate that language differences are negatively correlated with migration flows. However, many different methods of measuring linguistic differences have been used in previous research and each method has its strengths and weaknesses (Isphording and Otten, 2011). In this section, a few methods of measuring language differences are examined. The selection of linguistic difference indexes that are

(11)

11 examined in this section is based on the linguistic proximity measurement systems that were used in the article written by Adera and Pytlikova in 2015.

Firstly, a newly constructed Linguistic Proximity Index (LPI), based on the information from Ethnologue. This measurement system captures the linguistic proximity of two

languages by looking at the linguistic family tree, which represents the historical

relationships between all languages using a tree as metaphor. This measurement results in a number ranging from 0 to 1. Different kinds of relations between the two languages are defined by a set of increasing weights. The most aggregated linguistic tree level yields a weight of 0.1, for example Indo-European versus Uralic (Finnish, Estonian and Hungarian). The closer the languages are related, the heavier the weight factor.

Secondly, a measurement system developed by Max Planck called ‘the Levenshtein disctance’. This form of measuring linguistic differences looks at cognates. Cognates in this measurement system are defined as translation equivalents with high orthographic overlap. These words can either be identical or similar. An example is the Dutch-English translation pair sigaret – cigarette. This is a form-similar cognate. An example of an identical cognate is the word president, which is exactly the same in both languages. In this measurement system, a selection of a number of often used words across languages is made after which the translations are compared. The final step is computing how many steps are needed to move from one word expressed in one language to the same word expressed in another language. This measurement system results in an index number ranging from 0 till 106.39. The higher the index number, the less similar the languages are.

Lastly, a linguistic proximity measure proposed by Dyen et al. in 1992. This method of measuring language differences is similar to the Levenshtein distance approach. First, a selection of commonly used words is made. After which the similarity of this selection of words is compared between languages. With these measurements, an index is constructed that contains continuous metrics of proximity between any pair of languages. The results of this system can range from 0 till 1. The higher the index number, the more similar the languages are.

In this paper, only the linguistic proximity measure proposed by Dyen et al. will be used to estimate the effect of linguistic differences on migration flows within the European

(12)

12 Union. There are a few reasons why this particular linguistic difference index is used in this paper. Firstly, data on this LPI were easy to access. Data on other LPI’s were harder to find and get a hold of. Secondly, this LPI measures linguistic differences based on similarities between words that are often used. Resulting in an index that relates more to real life practical differences, instead of only family tree differences between languages, which are more theoretical. Lastly, the LPI proposed by Dyen et al. in 1992 is one of the most recently constructed linguistic differences index. Languages don’t tend to change substantially over time. However, when looking at differences between languages using a comparison between the most common, regular and real life used words, applying one of the most recent LPI’s seems beneficial for the measurement accuracy of current language differences.

Table 1 displays the LPI proposed by Dyen et al. between the 8 countries that are analyzed in this paper.

Languages Italian French Spanish German Dutch Danish English Greek

Italian 1 0.803 0.788 0.265 0.260 0.263 0.247 0.178 French 0.803 1 0.734 0.244 0.244 0.241 0.236 0.157 Spanish 0.788 0.734 1 0.253 0.258 0.250 0.240 0.167 German 0.265 0.244 0.253 1 0.838 0.707 0.578 0.188 Dutch 0.260 0.244 0.258 0.838 1 0.663 0.608 0.188 Danish 0.263 0.241 0.250 0.707 0.663 1 0.593 0.183 English 0.247 0.236 0.240 0.578 0.608 0.593 1 0.162 Greek 0.178 0.157 0.167 0.188 0.188 0.183 0.162 1

Table 1. Dyen matrix of linguistic distance (higher values mean smaller distance)

This LPI results in a number ranging from 0 to 1 to measure the linguistic difference between two languages. As can be seen in the table above, linguistic differences have a wide range in the 8 countries that are analyzed in this paper with index numbers ranging from 0.157 to 0.838.

According to the literature mentioned above, free labour movement should be efficient and advantageous. However, labour migration within the EU does not seem to occur frequently. Existing literature suggests that linguistic differences could be a possible barrier for labour movement. In this paper, the hypothesis is tested whether language has a

(13)

13 significant effect on labour migration in the European Union. The hypothesis is formulated as follows:

H0: Language differences have no significant effect on the labour migration between European Union members.

H1: Language differences have a significant negative impact on the effect migration between European Union members.

4. Relation between language and migration using a simple regression

In this section the used data and methodology are explained. The effect of language differences on labour migration between European Union members in this section is measured by conducting a simple regression.

4.1 Model

This section is aimed to look at the effect of linguistic differences on labour migration flows between European Union members by using a simple regression. The only variable that is used in this regression is the variable of interest, which is linguistic differences in this case. In this section, the following model is constructed:

𝑚𝑖𝑗= 𝛼1+ 𝛼2(𝐿𝑖𝑗) + Ɛ𝑖𝑗𝑡

where the dependent variable (𝑚𝑖𝑗) denotes the gross migration flows from country i to country j divided by the population of the country of origin i. These gross migration flows are calculated by taking the migration from country i to country j per 1000 inhabitants in country i.

The only independent variable in this model represents the linguistic differences between the origin and destination country using the LPI proposed by Dyen et al. in 1992. This regression should indicate whether language differences have a significant effect on migration flows between European Union members.

4.2 Data construction

For this analysis, data of two variables were needed: Linguistic differences and migration flows. Data on migration flows between European Union members were gathered by using the OECD data base and the article written by Pedersen, Pytlikova and Smith (2008). There is

(14)

14 no time dimension in this regression, because of the fact that language differences don’t alter too much over time. In order to compensate for the absence of a time element in this model, multiple years on migration data were analyzed. The variable 𝑚𝑖𝑗 is calculated by taking the average of all the migration flows between country pairs from 2000 until 2013. In total 54 observation were made on migration flows between a country of origin and

destination.

In this paper, linguistic differences are measured by using a LPI proposed by Dyen et al. (1992). This LPI measures the similarity between samples of words in different languages. These samples of words consist of often used words in a certain language. The Dyen index between country pairs can range from a value of 0 to 1.

4.3 Results

Table 2 shows the results from a regression on the model where only linguistic differences using the method proposed by Dyen et al. (1992) is used as an explanatory variable.

Coefficient t-value p-value

𝐿𝑖𝑗 -0.018 -0.19 0.850

Constant 0.156 3.62 0.001

Table 2. Effect of language differences on migration between European Union members using a

simple regression

These results show that, based on a simple regression on migration and linguistic distance data between eight European Union members, language differences do have a negative effect on migration flows between two countries. However, estimated coefficient of linguistic differences is not significant. Hence the null hypothesis is not rejected.

According to the results shown in Table 2, language differences could even be positive. This could be caused by the fact that not many observations were made in this regression. Also, the residuals in this regression substantially outweigh what the independent variable predicts, which could suggest that more variables are needed in order to construct a model where a more accurate and significant estimation of the effect of linguistic differences is measured. The latter is applied in the next section of this article.

(15)

15 5. A model of international migration

A problem with the simple regression is the substantial amount of residuals. In order to diminish these residuals and to get a more accurate measurement of the effect of linguistic differences, a new model is constructed where more explanatory variable are added. Just like in the simple regression, no time dimension is included and data is gathered by considering the years 2000 until 2013.

5.1 Model

In this part of the paper, more sophisticated approach is applied to measure the effect of language differences on labour migration between European Union members. Adera and Pytlikova did a worldwide study in 2015 where they constructed an econometric model to measure the effect of namely language differences on gross flows of migrants between countries. In this part, their methodology and econometric model will be replicated in order to measure the effect on language differences on labour migration flows in the EU. The following econometric model assumes that immigration rates between countries are driven by differences in wages, employment rates between countries and the cost of migration:

ln(𝑚𝑖𝑗) = Ƴ1+ Ƴ2ln (𝑔𝑑𝑝𝑖) + Ƴ3ln(𝑔𝑑𝑝𝑗) + Ƴ4ln (𝑢𝑗) + Ƴ5ln(𝑢𝑖) + Ƴ6ln(𝑝𝑠𝑒𝑗) + Ƴ7ln(𝑠𝑖𝑗) + Ƴ8(𝐿𝑖𝑗) + Ƴ9ln(𝐷𝑖𝑗) + Ƴ10(𝑁𝑖𝑗) + Ƴ11ln(𝑇𝑖𝑗) + Ƴ12ln(𝐹𝐻𝑗) + Ƴ13ln(𝑝𝑖𝑗) + Ɛ𝑖𝑗𝑡

The dependent variable in this econometric model, 𝑚𝑖𝑗, denotes the gross flow of migrants from country i to country j divided by the population of the country of origin i at time t. These gross migration flows are calculated by taking the migration from country i to country j per 1000 inhabitants in country i.

Adera and Pytlikova (2015) state that based on the methodology of previous studies, they proxy wages by GDP per capita and employment prospects in the sending and receiving countries. The first independent variables (𝑔𝑑𝑝𝑖 and 𝑔𝑑𝑝𝑗) represent the GDP per capita levels of the country of origin and the destination country. The unemployment rates (𝑢𝑖 and 𝑢𝑖) in both the destination country and the country of origin are also included in the model.

(16)

16 In addition, Borjas (1999) suggests that generous social security payment structures (𝑝𝑠𝑒𝑗) may influence migration flows. He argues that certain welfare programs can attract immigrants who otherwise would not have migrated to that country or region. Although little systematic study of the effect of social security payment structures on migration flows has been done, this variable is still concluded in the econometric model in this article. Mostly because Adera, Pytlikova (2015) and Borjas (1999) see potential importance of the effects of these safety nets for immigrants. Similar to Adera and Pytlikova (2015), in the model used in this article 𝑝𝑠𝑒𝑗 is measured by taking public social expenditure as a percentage of national GDP.

The cost of migration is expected to increase with larger physical, cultural and linguistic distance between the origin and destination country. However, the cost of migration might fall if migration networks are present. A migration network results in an easier entrant into the labour market for its members and it is also able to channel them into higher paying occupations (Munshi, 2003). In this econometric model, 𝑠𝑖𝑗 is defined by the total foreign population from country i living in country j per population of the country of origin i.

The following variable 𝐿𝑖𝑗 measures the effect of language differences on migration flows. This variable is aimed to answer the research question and the main hypothesis of this paper, namely whether linguistic differences have a significant effect on migration flows between European Union members. The way linguistic differences are measured in this article is by using the LPI proposed by Dyen et al. in 1992 and Table 1 shows the linguistic distance between all country pairs that are analyzed in this paper.

Even though transportation and communication have been improved and getting cheaper substantially in recent times, the actual distance between the origin and destination country are expected to raise the direct cost of migrating and therefore influence migration flows (Adera & Pytlikova (2015). To account for the effect of distance on migration flows, the variable 𝐷𝑖𝑗 is included in the model. This variable is defined by the distance in kilometers between the capital areas in the origin and destination country. Also included in this model is a dummy that indicates whether the origin and destination country are neighboring

(17)

17 countries, 𝑁𝑖𝑗, where the value of this variable is one if both countries are neighbors, zero otherwise.

The effect of trade between the origin and destination country is also included in this model. The variable 𝑇𝑖𝑗 represents the trade volume and it is defined as the total trade values (both imports and exports) for all pairs of countries. Adera and Pylikova (2015) expect that the business ties and relationships between two countries, represented by the volume of trade in this article, effects the immigration between the two countries.

A measurement of the degree of freedom in political rights and civil liberties in each country is also represented in this model: 𝐹𝐻𝑗. This variable represents a score that ranges from one to seven, with one being the highest degree of freedom and seven the lowest. The source used to for this variable: Pedersen, Pytlikova and Smith (2008).

Finally, the last variable that is included in this econometric model, 𝑝𝑖𝑗, measures relative population size of the country of origin with respect to the destination country. Just like in the model used by Adera and Pytlikova (2015), this variable is included to control for demographic differences between countries.

All the variables in this model, except for the dummy variable and the LPI, are expressed in logarithms. The main reason as to why most variables are transformed into logarithms is because this model is based off of the model used in the paper written by Adera and Pylikova in 2015. The model used in this part of the paper, is an attempt to replicate their study and methodology.

5.2 Data construction

Data regarding migration flows and linguistic differences is the same as in the simple regression. No different sources were used and no different measurement method was applied. The datasets are identical for these two variables.

Data on GDP of all the countries used in this analysis is gathered by using the World Bank database. The average GDP of the period between 2000 and 2013 is calculated and used in this analysis. Also, the World Bank database functioned as a source for the unemployment rate data used in this regression. Just like the GDP variable, the

(18)

18 2000 until 2013. The World Bank database was also used to find data on public social

expenditure of countries and trade volume between countries. Again, the average of the period from 2000 until 2013 is calculated to determine the values of these variables in this regression.

Data regarding the percentage of people from the origin country living in the destination country was collected using the OECD database.

The distance variable was determined by the distance between capital cities of the country of origin and the destination country. These distances were calculated by measuring the shortest distance between two points or alternatively, a straight line between two points.

Lastly, the Freedom House variable was determined by data used in the article of Pedersen, Pytlikova and Smith (2008). They took multiple reports of namely the Annual Freedom in the World Country Scores into account and applied these Freedom House scores in their article.

5.3 Results

The results of the multiple variable regression are shown in Table 3. The goal of adding more variables was to decrease the amount of residuals and to increase the significance of the estimates.

(19)

19

Coefficient t-value p-value

ln (𝑔𝑑𝑝𝑖) -0.75 -6.38 0.000 ln(𝑔𝑑𝑝𝑗) -2.34 -1.27 0.210 ln (𝑢𝑗) -0.92 -1.20 0.237 ln(𝑢𝑖) -0.03 -0.13 0.895 ln(𝑝𝑠𝑒𝑗) -0.37 -0.37 0.717 ln(𝑠𝑖𝑗) 0.35 3.15 0.003 𝐿𝑖𝑗 -0.08 -0.12 0.903 ln (𝐷𝑖𝑗) -0.02 -0.07 0.948 𝑁𝑖𝑗 -0.07 -0.21 0.834 ln(𝑇𝑖𝑗) 0.03 0.21 0.838 ln(𝐹𝐻𝑗) 4.10 1.45 0.154 ln(𝑝𝑖𝑗) -0.15 -1.49 0.145 Constant 5.59 0.19 0.847

Table 3. Effect of language differences on migration between European Union members using a multiple variable regression

First off, the relative amount of residuals decreased substantially. This model predicts noticeably better than the simple regression, which is due to the increased number of

variables. However, the estimate for the effect of linguistic differences got even less significant. Thus the null hypothesis is still not rejected.

Moreover, as shown in Table 3, many included variables are insignificant. A probable causing factor could be the lack of observations. An increase in observations would probably make the confidence intervals of each variable narrower, thus causing better estimates of the variables. Another possible problem with this model could be that too many control variables are added. Due to a low number of observations, a high number of variables can create a lot of noise and therefor inaccurate estimates in a regression. In order to see if having this many control variables in the model might be the cause of this many insignificant estimates a new model is constructed. In the next part of this paper, this new model is specified and a new regression is done.

(20)

20 5.4 Leaving out insignificant control variables

Most of the control variables that were added to the original model ended up being insignificant in the regression. In this part, insignificant control variables are excluded from the model. In this part, a regression will be done on the following model after excluding the insignificant control variables:

ln(𝑚𝑖𝑗) = Ƴ1+ Ƴ2ln (𝑔𝑑𝑝𝑖) + Ƴ3ln(𝑔𝑑𝑝𝑗) + Ƴ7ln(𝑠𝑖𝑗) + Ƴ8(𝐿𝑖𝑗) + Ɛ𝑖𝑗𝑡 Again, 54 observations were made and the data used to estimate the included variables remained the same as in the previous multiple variable regression. The results of the regression on this model are given below:

Coefficient t-value p-value

ln (𝑔𝑑𝑝𝑖) -0.75 -6.40 0.000

ln(𝑔𝑑𝑝𝑗) -1.20 -2.47 0.017

ln(𝑠𝑖𝑗) 0.56 8.09 0.000

𝐿𝑖𝑗 -0.46 -1.10 0.276

Constant 19.94 3.80 0.000

Table 4. Effect of language differences on migration between European Union members including only significant control variables.

When only the significant and close to significant control variables are used in the regression, the results show an increase in the significance of all these variables. The effect of linguistic differences also increased, both the coefficient and the t–value. This seems to suggest that given the fact that the amount of observations was substantially low, the large amount of control variables caused complications in the model. The coefficients were hard to estimate, thus creating substantial confidence intervals and low t-values. Due to excluding the insignificant control variables, the model seems to estimate the coefficients of the remaining variables noticeably better. Even though both the coefficient and t-value of 𝐿𝑖𝑗 increased, they remain insignificant. Which means that the null hypothesis will still not be rejected which suggests that linguistic differences might not have an effect on labour migration within the European Union.

(21)

21 5.5 Transformation of the dependent variable 𝒎𝒊𝒋

In this part, the dependent variable will be transformed in an attempt to make the relationship between the dependent and the independent variables more linear. Thus in order to estimate the effect of linguistic differences on migration flows within the European Union more accurately. The following model is regressed in this section:

ln(𝑚𝑖𝑗) = Ƴ1+ Ƴ2ln (𝑔𝑑𝑝𝑖) + Ƴ3ln(𝑔𝑑𝑝𝑗) + Ƴ7ln(𝑠𝑖𝑗) + Ƴ8(𝐿𝑖𝑗) + Ɛ𝑖𝑗𝑡

The model is similar to the one applied in subpart 5.4. However, there are two differences between this model, and the model used in the previous section. Firstly the dependent variable, ln(𝑚𝑖𝑗), is now transformed:

ln(𝑚𝑖𝑗) = ln ( 𝑚𝑖𝑗 1 − 𝑚𝑖𝑗

)

Secondly, 𝑚𝑖𝑗 is defined differently in this model. The dependent variable is now defined as a ratio: the migration flow from country i to country j, divided by the total migration out of country i.

Again, 54 observations were made and the data used to do this regression is the same as the data used in the previous sections. The results of this regression are shown in Table 5:

Coefficient t-value p-value

ln (𝑔𝑑𝑝𝑖) 0.14 1.25 0.216

ln(𝑔𝑑𝑝𝑗) -1.16 -2.45 0.018

ln(𝑠𝑖𝑗) 0.62 9.17 0.000

𝐿𝑖𝑗 -0.37 -0.92 0.363

Constant 10.33 2.02 0.049

Table 5. Effect of language differences on migration between European Union members, using a transformed version of 𝑚𝑖𝑗.

After transforming the variable that represents migration flow from one country to another, the results still indicate that there is a negative relation between linguistic

differences and migration flows within the EU. However, the results are still not significant. In all the regressions conducted in this thesis, no statistically significant results are found.

(22)

22 Therefor the null hypothesis is not rejected, suggesting that linguistic differences do not have a significant effect on migration flows within in the EU.

6. Conclusion

In this paper, the effect of linguistic differences on migration flows within the EU is

investigated. Specifically, eight European Union members have been analyzed to determine whether linguistic differences have a significant effect on migration flows between EU countries. Data from 2000 until 2013 has been used to calculate the values of the variables that are time sensitive. Moreover, the LPI proposed by Dyen et al. (1992) has been applied to determine the linguistic differences between the eight countries that were analyzed in this paper. The aim of this thesis has been to answer the following research question: Do language differences have a significant effect on labour migration within the EU?

To answer this question, firstly, a literature review has been conducted to examine what literature says on this topic. Existing literature seems to suggest that linguistic

differences do have an effect on migration flows. Secondly, the relation between linguistic differences and migration flows within the EU has been tested empirically through

conducting multiple regressions. The results of the regressions done in this paper show that linguistic differences do not have a significant effect on migration flows within the European Union. However, all the coefficient estimates of the effect of linguistic differences ended up being negative, suggesting that there is a negative relation between linguistic differences and migration flows within the EU.

To conclude, this paper provides insight on the effect of linguistic differences on migration flows within the EU. The results found in this thesis are not significant. However, they seem to suggest that an efficient distribution of labour across all EU countries is

partially hindered by the linguistic differences that exist between European Union members. 7. Discussion

In this thesis, an attempt to replicate the methodology of the paper written by Adera and Pylikova (2015) has been done. In their paper, they found a robust and significant effect of linguistic distances on migration flows worldwide. In this thesis, no such robust and

significant results are found. One of the possible causing factors for these insignificant results could be the small sample size. The number of observations in all the regressions

(23)

23 done in this paper did not exceed 54. Adera and Pylikova (2015), for instance, had access to a database that allowed them to make thousands of observations, which increased the significance of the estimated coefficient. A recommendation for further research can be to include more European Union member in order to increase the number of observations.

Also, only one linguistic proximity measurement system is used in this thesis. Further research could include more ways to measure linguistic differences in order to get a more complete perspective on how linguistic differences affect migration flows.

Lastly, a recommendation for further research is to categorize countries into clusters. In this thesis, only a broad regression is conducted on the eight selected EU countries. Further research could make a distinction between Northern Europe and Southern Europe, for instance. It could be the case that labour migration is on the same level in Northern Europe as it is in the US. However, this is not analyzed in this thesis and could be possible to include in future research.

(24)

24 References

Adera, A., Pytlikova, M., (2015). The Role of Language in Shaping International Migration. The Economic Journal, 125(586), 51-77.

Borjas, G., (1999). Immigration and Welfare Magnets. Current Journals, 17(4), 608-609. Chriswick, B., Miller, P., (2005). Liguistic Distance: A Quantitative Measure of the Distance

between English and Other Languages. Journal of Multilingual and Multicultural Development, 26(1), 1-9.

Docquier, F., Machado, J., Sekkat, K., (2015). Efficiency Gains from Liberalizing Labor Mobility. The Scandinavian Journal of Economics, 117(2), 303-335.

Eurofound (2014). Labour Mobility in the EU: Recent trends and policies. Publications Office of the European Union, Luxembourg, 16-17.

European Commission (2014). Labour Mobility within the EU. Retrieved from: http://europa.eu/rapid/press-release_MEMO-14-541_en.htm

Hamilton, B., Whalley, J., (1982). Efficiency and Distributional Implications of Global Restrictions on Labour Mobility. Journal of Development Economics, 14(1), 61-75.

IMF eLibrary Data (2016). Direction of Trade Statistics. Retrieved from

http://data.imf.org/?sk=9D6028D4-F14A-464C-A2F2-59B2CD424B85&sId=1454703973993

Isphording, I., Otten, S., (2011). Linguistic Barriers in the Destination Language Acquisition of Immigrants. Journal of Economic Behavior and Organization, 105(1), 45-46.

Munshi, K., (2003). Networks in the modern economy: Mexican migrants in the US labor market. Quarterly Journal of Economics, 118(2), 550-554.

OECD, International Migration Database (2016). Inflows of foreign population by nationality. Retrieved from https://stats.oecd.org/Index.aspx?DataSetCode=MIG

OECD, International Migration Database (2016). Stock of foreign population by nationality. Retrieved from https://stats.oecd.org/Index.aspx?DataSetCode=MIG

(25)

25 Pedersen, P., Pytlikova, M., Smith, N., (2008). Selection and Network Effects – Migration

Flows into OECD Countries 1990-2000. European Economic Review, 52(7), 1161-1184.

Tassinopoulos, A., Werner, H., Kristensen, S., (1998). Mobility and migration of labour in the European Union and their specific implications o for young people. European Centre for the Development of Vocational Training, 5-6.

The World Bank, International Labour Organization, Key Indicators of the Labour Market Database (2016). Unemployment, total (% of total labor force). Retrieved from http://data.worldbank.org/indicator/SL.UEM.TOTL.ZS?year_low_desc=false

The World Bank, World Bank National Accounts Data and OECD National Accounts Data Files (2016). GDP per capita. Retrieved from

http://data.worldbank.org/indicator/NY.GDP.PCAP.CD

The World Bank, World Bank National Accounts Data and OECD National Accounts Data Files (2016). General government final consumption expenditure (% of GDP). Retrieved from http://data.worldbank.org/indicator/NE.CON.GOVT.ZS

Van Dalen, H., Henkens, K., (2012). Explaining Low International Labour Mobility: the Role of Networks, Personality and Perceived Labour Market Opportunities. Population, Space and Place, 18(1), 31-35.

Zimmermann, K., F., (1995). Tackling the European Migration Problem. The Journal of Economic Perspectives, 9(2), 52-57.

Referenties

GERELATEERDE DOCUMENTEN

According to The European Action Coalition for the Right to Housing and the City who organized in 2016 a protest when the housing ministers of all EU-countries met in Amsterdam to

Tevens wordt er verwacht dat de deelnemers die hoog scoren op de extraversievragenlijst niet verschillen op de disidentificatievragenlijst wanneer er geen sprake is van een

On the geopolitical level we’ve identified four factors that are important for the EU: Iran’s nuclear program, the relationship between Iran and its regional rival Saudi Arabia,

INDUSTRY-LINKED PROJECT WORK: INTERDISCIPLINARITY WITH DESIGN, ENGINEERING AND MANAGEMENT

Zur Frage der Entstehung Maligner Tumoren (Fischer). Castellanos, E., Dominguez, P., and Gonzalez, C. Centrosome dysfunction in Drosophila neural stem cells causes tumors that are

In the B2C market the secondary and case study data indicate that culture has its effect on e- commerce in customer loyalty (trust), site attractiveness (web-design) and

The direction of Infrastructure shows negative coefficients for all the pooled OLS model specifications and generally positive coefficients for the BE model, with the

Dependent variable is firm performance (Q) measured as the market to book value of assets, net working capital (NWC) measured as the ratio of accounts receivables plus