Naturalization and Social Capital Investment

Naturalization and Social Capital Investment

Evidence from 32 European Countries

John Eriksson Oscar Svensson

Spring 2019

Abstract

In times of increased migration, knowledge about how to best integrate migrants is crucial. In this paper, we investigate whether citizenship can facilitate integra- tion by increasing investment into social capital. The question is of interest as social cohesion and social capital investment are key determinants of economic growth. Us- ing data from the European Social Survey, we investigate the relationship between citizenship and several social capital investment measures with a Linear Probability model, a Two-stage Least Squares and a Bivariate Probit model. For exogenous variation in citizenship, we create an instrument based on variation in whether the country of origin allows their migrants to acquire dual citizenship. The results are mixed. In the naive linear model we find a positive relationship, but the results are not robust to instrumental variable models that deal with selection into citizenship.

Policy makers should not put too much faith in potential positive effects of natu- ralization on social integration.

Supervisor: Ariel Phil

Master’s thesis in Economics, 30 HECs.

Graduate School, School of Business, Economics and Law, University of Gothenburg.

Acknowledgements

We extend our gratitude to our supervisor Ariel Phil, for professional input and

feedback. Her high regard for quality and her motivational attitude has been a valuable

factor during the entire writing process.

Contents

1 Introduction 1

1.1 Review of the Literature on Citizenship and Social Capital . . . . 3

2 Theoretical Framework 6 2.1 A Model of Expected Stay and Social Capital Investment . . . . 7

3 Data 10 3.1 European Social Survey . . . . 10

3.2 Social Capital Variables . . . . 11

3.3 Naturalization . . . . 14

3.4 Control Variables . . . . 15

3.5 Descriptive Statistics . . . . 17

4 Method 20 4.1 Linear Probability Model . . . . 20

4.2 Two-Stage Least Squares & Bivariate Probit . . . . 21

5 Results 26 5.1 Linear Probability Model - Results . . . . 27

5.2 Two-Stage Least Squares - Results . . . . 29

5.3 Bivariate Probit - Results . . . . 32

5.4 Sensitivity Analysis . . . . 32

6 Discussion 33

7 References 37

A Appendix 41

1 Introduction

Currently, more than 230 million people live as international immigrants (UNDESA 2013), and the number of migrants moving from the ‘global south’ to the ‘global north’ is ex- pected to increase (De Haas 2007). As immigration increases, so does the importance of investigating how to integrate migrants in their host societies. One of the key con- cepts in doing so is social integration, through the formation of social capital (Chan et al.

2006). In this paper, we will use the concept of social capital similar to Putnam (2000), who stresses the importance of civic society, voluntary organizations and social networks.

A potential way of promoting social capital investment is to increase the naturalization rates of immigrants, as it could increase the returns to social capital investment by the migrant, and lead to increased social integration. This could be an effective policy lever to increase integration and social capital investment, which in turn enhances economic ac- tivity through lower monitoring and transaction costs (Brehm & Rahn 1997, Beugelsdijk

& Smulders 2003, Nannestad et al. 2008). While the effects of naturalization on integra- tion and social capital are of great economic importance, the research area has arguably received too little attention (OECD 2011, Hainmueller et al. 2017). Most of the previous economics literature has examined the effects of naturalization on labor outcomes, such as wages and employment, and have left out the aspect of integration. It is therefore impor- tant to investigate and understand the relationship between citizenship and integration, in order to provide better recommendations for policy makers (Zimmermann et al. 2009).

We contribute to this literature by examining the effects of migrants’ citizenship ac- quisition on social capital investment using data from 32 European countries. One key issue of the literature is to handle the exogeneity issue of citizenship acquisition. Because citizenship is a decision, where the migrant needs to apply and thereafter be approved, the citizenship decision is usually contingent on many factors that might correlate with the social capital outcomes of interest. These factors could be education levels, income, ambition or unobserved willingness to integrate. Citizenship acquisition signals attach- ment and dedication to the host country as well as an understanding of host country institutions, both economic and cultural (Bevelander & DeVoretz 2008). It could also make the migrant value host country social capital investment differently and citizenship should therefore be associated with both economic and social integration.

Because of this non-random selection, little is known about the causal effects of natu-

ralization (Baub¨ ock 2006). Previous literature has largely focused on descriptive statistics

and correlations between social capital and citizenship. Examples include, for example, the 2011 report released by the OECD on the effects of naturalization, containing case studies for many countries including Belgium, Canada, the UK, Australia and the Nether- lands (OECD 2011). Although important, these type of studies does not answer the rele- vant policy question of what the real effects of citizenship are. Some studies such as Just

& Anderson (2012) and Hainmueller et al. (2017) have gone further and, similarly to us, tried to deal with the exogeneity issue using instruments. We argue that the contribu- tion in our study compared to their studies lies in larger sample sizes and more plausible exogenous variation in citizenship.

Small sample sizes have been a reoccurring problem in the current literature on the topic (Baub¨ ock 2006, Bevelander & DeVoretz 2008). We deal with this issue by pooling eight waves of the European Social Survey (ESS) and obtain a sample of roughly 35,000 migrants, from almost every country around the world.

We address the endogeneity issue of by using new variation with an instrument vari- able approach, where the instrument is based on whether countries of origin permit dual citizenship or not. One of the main costs of acquiring a new citizenship for a migrant is the cost of renouncing her previous citizenship (Bevelander & DeVoretz 2008). By uti- lizing changes and variation in the dual citizenship policies of the migrants’ countries of origin, we create an instrumental variable which affects the likelihood of naturalization, but arguably does not affect social capital investment in the host-country through any other unobserved channels. Using this instrument, we mitigate the endogeneity problem of self-selection into citizenship.

We investigate the effect of citizenship using a Linear Probability Model (LPM) model, as well as a linear Two-stage Least Squares (2SLS) model and a non-linear Bivariate Probit (BP) model. The 2SLS and the BP are both IV approaches, arguably dealing with the aforementioned endogeneity issue. The 2SLS is the standard linear IV model while the Bivariate Probit is an extension of the Probit model and therefore allows for non-linearity.

The results are mixed. Although a general positive relationship between naturalization and social capital can be seen, the results are not robust to changes in model specification.

In the linear model, citizenship is associated with an approximate 3.5 percentage point

increase in the probability of the migrant stating that she participates at least as much as

the average migrant in social activities. Further, it is associated with a 3 percentage point

increase in the probability of reading newspapers on a daily basis. It is also associated

with a 2 percentage point increase in the probability of participating in non-political

organizations.

These results are not robust to models which handle the endogeneity issue, and the effects could potentially be attributed to self-selection. Nevertheless, in the BP model, acquiring citizenship is associated with 4.5 percentage point increase in the likelihood of participation in a non-political organization. This effect is robust to country of origin, host country and survey year fixed-effects as well as socioeconomic controls. This suggests that there might be a causal effect of naturalization on one of the outcomes of interest.

In sum, the mixed results indicate that the effect of naturalization is subtle and that most of the effect may be attributed to self-selection.

The rest of the paper will be structured as follows: In the next section, we review previous literature on the topic. Section two will lay out the theoretical framework we use to formulate the hypotheses of interest. Section three presents the data used for the analysis. In section four the methodological approach as well as the econometric specifications are explained and motivated. Section five is a presentation and explanation of the results, and section six discusses the implications of the results and concludes.

1.1 Review of the Literature on Citizenship and Social Capital

There exists a current debate regarding whether naturalization should be considered the end station of the integration process, or the beginning of it (Hainmueller et al. 2017).

Naturalization can be seen simply as a sign of a successful integration process and should therefore not be seen as a tool through which integration can be improved. It can however, to the contrary, also be thought of an event which in itself can enhance integration both for labour and social capital related outcomes (Steinhardt 2012, Hainmueller et al. 2017).

Acquiring citizenship could affect migrant social capital outcomes through many dif- ferent channels. Perhaps the most obvious benefits given by citizenship are lowered formal and informal barriers such as access to certain type of jobs, or lower perceived discrimina- tion. Analyzing the labour market effects of naturalization, Bratsberg et al. (2002) argue that the mitigation of such barriers is the cause of the naturalization effect, and find that the migrant wage growth in the US increases after citizenship acquisition. Similarly, citi- zenship acqusition seems to be associated with higher employment rates among migrants in France (Foug` ere & Safi 2009) and increased job chances for immigrants and refugees in the Netherlands (Bevelander & Veenman 2006).

Also investigating the effect of naturalization on labor outcomes, Steinhardt (2012)

uses German longitudinal register data from the Institute for Employment Research (IAB). He circumvents the issues of self-selection into citizenship as the panel data struc- ture allows him to control for the time-invariant part of the unobserved heterogeneity by following and estimating the outcomes for the same individuals before and after natu- ralization. He finds wage growth to be higher for naturalized males, but finds no effect for females. Steinhardt concludes that a large share of the wage premium is due to self- selection into naturalization, but also that naturalization seems to have an effect even after controlling for this. Regarding the female part of the sample he argues that the specification might suffer from an omitted variable bias as he is not able to control for the number of children and marital status, which are key determinants for female labor participation. Given the survey nature of our data set compared to the administrative structure of the data Steinhardt uses, we are able to control for these factors.

More in line with the specific topic of this paper, other studies have shifted the at-

tention from wages and employment to social capital, civic engagement and political

integration, where the political science and sociology literature has made a significant

contribution. Just & Anderson (2012) investigate the effect of citizenship acquisition on

political participation, using an index of different political participation measures as their

outcome variable. The measures include, for example, whether the respondent has par-

ticipated in a political organization or whether they have donated money to charitable

causes. Using the same data set as the one used in this paper, although using fewer

waves, obtaining a sample size of approximately 2600 migrants, they find that citizenship

acquisition and length of stay in the host country has a positive effect on political partici-

pation among migrants, and that regardless of citizenship, migrant political participation

rates remain low. The authors use an instrumental variable strategy where they instru-

ment citizenship with the geographical distance between the country of origin and the

host country. They argue that the distance from the home country should not have an

impact on political participation rates other than through the effect it has on citizenship

acquisition probability. We are not fully convinced by their reasoning regarding this, as

the geographical distance between two countries might correlate with other factors such

as cultural and political proximity, which could affect the political outcome variables in

other ways than through its effect on the naturalization probability. We argue that the

instrument we use in this paper is more reliable, partly because we can exploit time vari-

ation in when the dual citizenship policies were introduced, and because we are able to

control for country of origin fixed effects, which is not possible with their instrument as

geographical distance does not vary over time.

As a part of the OECD (2011) report, Kesler & Demireva (2011) investigate the relationship between naturalization and a number of social capital outcomes. They use the European Social Survey and focus on trust-related outcomes, including trust in the political system or the police, and measures of life satisfaction in general and satisfaction with the host country. As with many of the previous studies on this topic, they do not claim causality as they are unable to disentangle the effects they estimate from possible confounding factors, and also admit they cannot say for certain whether their analysis suffers from bias induced by reverse causality.

Investigating a similar question to our own, Hainmueller et al. (2017) analyze the ef- fect of naturalization on social capital outcomes by exploiting how the Swiss migration system functions in some municipalities, where naturalization requests were decided on by a referendum. Their sample consist of roughly 2000 migrants. In order to circum- vent the endogeneity issue of citizenship they employ an IV strategy where approval or disapproval constitutes the instrument variation, as well as a fuzzy regression disconti- nuity (RD) design around the cutoff where applicants where either just approved or just rejected. The outcomes covering social integration they use are: Plans to stay in Switzer- land, perceived discrimination, active membership in a voluntary club and whether the migrant reads Swiss newspapers. They also construct an integration scale combining the four measures. Due to sample size limitations, they are forced to use quite a wide range (35-65% ‘yes-votes’) as the definition of narrowly lost or narrowly won. They find that naturalization significantly increases the migrant’s score on the integration scale in their RD- as well as in their IV-estimates. In addition to small sample sizes, a shortcoming of their estimations is that they are comparing individuals who received citizenship by win- ning a referendum to those who lost by getting rejected. A migrant being publicly rejected may affect the migrant’s decisions regarding social capital investment as well. Neverthe- less, Hainmueller et al. (2017) conclude that naturalization improved long-term social integration in migrants and that naturalization is a ‘catalyst’, rather than a ‘crown’ for integration, suggesting more restrictive citizenship acquisition policies may hurt migrant integration. The effect seems to be stronger for groups of migrants which are marginalized in Switzerland.

Our paper shares similarities with Hainmueller et al. (2017), but differs in that we

are able to estimate an effect with a larger sample size and with a different identification

strategy. As we are using data covering most of Europe, we are also able to analyze

whether the effects found in Switzerland extend to a broader European setting, and a more generally relevant margin. We are also able to test if the effect of naturalization is different without the non-citizens having been rejected citizenship, which is likely to overestimate the effect of citizenship.

It is important to note that not all social capital is equally desirable as a policy goal. Putnam (2000) makes the distinction between ‘bridging’ social capital and ‘bonding’

social capital. Bridging social capital is developed over diverse groups of people while bonding social capital is created within homogeneous groups. Because the creation of bridging social capital requires diverse groups of people to socialize and cooperate, it creates trust which transcends the group the migrant belongs to. Increased trust decreases overall transaction and monitoring costs and helps solve collective action problems which has many positive effects, including increased economic growth (Brehm & Rahn 1997, Beugelsdijk & Smulders 2003). Nannestad et al. (2008) elaborate on these different types of social capital and argue that bonding social capital might still be beneficial if it creates positive spillover effects into bridging social capital.

Overall, we recognize several recurring themes in previous literature on the topic of the effects of naturalization of migrants. Many studies have acknowledged and tried to deal with the endogeneity of citizenship acquisition, but only a few with convincing methodological approaches. Several papers are also of a more qualitative rather than quantitative nature. Focus has mostly been on labour outcomes, for example impacts on labour force participation or wage effects, while much less time has been spent on exploring how naturalization affects social capital acquisition among migrants. Our contribution thereby lies in our discussion of the effect of naturalization on social capital outcomes with new instrumental variation and a larger sample size than previous studies. In the next section of this paper, we elaborate on the economic model and theoretical framework we use as a baseline for subsequent econometric analysis.

2 Theoretical Framework

We draw inspiration from the return migration modelling framework developed by Dust-

mann (Dustmann 2000, Dustmann & G¨ orlach 2016), and add naturalization as an ex-

planatory factor. In our model, naturalization has an effect on the value of staying in the

host country, and will subsequently positively affect the expected duration of stay in the

host country. This will, in theory, also positively affect the optimal level of social capital

investment of the migrant.

2.1 A Model of Expected Stay and Social Capital Investment

The migrant’s expected stay is determined by the value of staying in the host country:

V N = Ω(π N , c, Y N , X) (1)

For our purposes, the expected stay V _N (Ω) is a function of preferences to consume in the host country (π _N ), wages (Y _N ), consumption (c) and other factors (X), which include for example skills and age. The function is positive in preferences to consume, consumption and income. The Dustmann & G¨ orlach (2016) model includes both country of origin and host countries. We focus only on the latter, but also allow preferences to consume and wages to vary by naturalization status. Naturalization is described by N ∈ {0, 1}, with N equal to 1 if the migrant is naturalized and 0 otherwise. Further, we argue that the preference parameter for consumption in the host country, as well as income, is higher when naturalized:

π 1 > π 0 (2)

and:

Y 1 > Y 0 (3)

Where π ₁ and Y ₁ are the preference parameter and income for a naturalized migrant in the host country, compared to π 0 and Y 0 for a non-naturalized migrant.

A way in which naturalization should affect preferences to consume in the destina- tion country, as seen in eq. (2), is noted by Hainmueller et al. (2017). They state that naturalization strengthens the long term attachment to the host country and reduces un- certainty regarding return migration, since naturalization enables the migrant to sponsor their family members in obtaining their own citizenship. Further, in countries basing their citizenship regime on a Jus Sanguinis system, receiving citizenship automatically ensures the citizenship of the migrant’s children. ¹ This will further strengthen the long- term attachment to the host country. The authors also note that the signalling effects of citizenship acquisition might decrease discrimination, which would increase preferences

1 As opposed to a Jus Soli system, where citizenship is granted by place of birth, a Jus Sanguinis system

is based on citizenship acquisition by descent. This would mean that a child receives the citizenship of

its parents, regardless of where it is born, and not necessarily citizenship in the country of birth.

to consume in the host country. This results in a an increase in acceptance by the native population, and a greater sense of belonging.

Naturalization removes formal and informal obstacles for the migrant, reduces the costs and increases the gains of assimilation, which we predict will lead to a higher preference for consumption in the destination country.

The income increase from naturalization, seen in eq. (3), can be explained by the fact that naturalization provides access to certain types of jobs and greater opportunities in the labor market. From the employer side, this could for example be through a reduction in administrative costs associated with hiring a non-citizen. One example of this is the German principle of a ‘priority-test’ which forces firms to ensure that no German national or EU citizen is available for the job before they hire a non-citizen migrant for the same position (Steinhardt 2012). Citizenship also mitigates uncertainty for the firm regarding the duration of stay of the migrant, as citizenship ensures that the migrant has full legal rights to stay indefinitely in the country. Other formal benefits of naturalization include the right to certain jobs in the public- and security sector, ease of travel due to a new passport (which is an especially important perk of citizenship for white collar workers) and easier obtainment of benefit schemes.

Naturalization further shows that the migrant is committed and dedicated to the country long-term. This could work as a positive signalling effect for employers and an indication of that the migrant is more willing to invest into country-specific human and social capital (Bevelander & Sp˚ ang 2015). This should also positively affect the wages earned by the naturalized migrant.

For the above described reasons, holding all other factors constant, we see in eq.

(1) that V _N (Ω) increases following naturalization. In theory, naturalization therefore positively affects the expected stay in the host country.

However, one should note that there might be other mechanisms and factors that work in the opposite direction. For instance, it is imaginable that a migrant who naturalizes feels she has no more obligation or incentive to stay and integrate in the host country, since she has already received citizenship and all the formal benefits associated with it.

In this scenario, naturalization would impede rather than increase social capital invest- ment. Even so, we expect the positive effects to outweigh the potentially negative effects.

Naturalization should increase the expected duration of stay because the value of staying in the host country is higher.

Based on Dustmann (2000), we describe how the optimal investment in social capital

is a function of expected stay as follows:

SC ^∗ = ω ₁ V _N (Ω) + ω ₂ X (4)

With:

ω ₁ > 0 (5)

Where the optimal investment into social capital, SC ^∗ , depends on the expected stay in the host country V _N (Ω) and other factors X. ω ₁ is the causal effect of increased expected stay on the optimal social capital investment. It is however important to note that the social capital investment level could also be a determinant of the expected duration of stay in the host country, part of X in eq. (1). In a simple linear model, we could therefore obtain a positive association between naturalization and social capital, driven by self-selection.

However, in the instrumental variable framework we are able to exogenously affect naturalization status, leaving us with a, in theory, causal prediction of naturalization positively affecting social capital investment, through a longer expected stay in the host country.

As argued by Dustmann (2000), migrants’ expected stay in the host country affects the marginal benefit of investing in human capital, which should have a similar effect on social capital. This is intuitive, since an increased duration over which country-specific social capital can be utilized will strictly increase the gain from every unit of social capital acquired, provided that the individual discount rate does not change when the migrant is granted citizenship. Investing in social capital by, for example, participating in voluntary organizations, expanding your social network or joining a trade union makes little sense if you do not see yourself staying for a longer time period in which these investments can generate large enough returns.

In sum, we expect naturalization to increase the time the migrant expects to stay in the host country. This makes her more committed to a long-term future in the country, which increases the incentives to invest in social capital. Using this framework, we arrive at the following hypotheses:

Hypothesis 1: Naturalization will increase investment in social capital, through a longer

expected duration of stay in the host country.

Hypothesis 2: Estimates from the LPM will be larger in magnitude compared to the IV results, since there should be endogenous positive selection that is uncontrolled for.

3 Data

3.1 European Social Survey

For the empirical analysis we use the ESS, which is a biennial survey conducted in 32 European countries (European Social Survey 2016). The cross-sectional survey measures behaviors and attitudes of representative individuals above the age of 15, to date con- ducted eight times between 2002 and 2016.

For all subsequent empirical analysis and statistics, we weight the data using ESS post-stratification weights. ²

For some specifications in the sensitivity analysis we multiply the post-stratification weights with a population weight in order to control for size of the host country. The population weight differs by country but is identical for individuals from the same country.

This weight is applied in order to account for the fact that most countries have similar sample sizes in the data but are obviously not similar in population size. If not adjusted for, this would give a disproportionately large weight to smaller countries, and our results would not be representative of Europe as a whole. On the other hand, migration policy is largely a national matter and weighting by population size will give high importance to large countries and low importance to small countries. For this reason, if the effect is small in, for example, Germany, a large country in our data set, the total effect might be small or non-existent although there is an effect in smaller countries. Using only the post-stratification weights will therefore give us the average effect of naturalization on social capital outcomes in the group of countries in our sample. Although we cannot argue for a ‘European’ effect with this weight, it is arguably the one which is of higher policy interest.

Pooling the data across all countries and across the eight survey waves available, we are left with a total of roughly 375,000 individuals, among them 35,000 migrants. This

2 Post-stratification weights take into account the fact that certain groups of individuals might be more

or less likely to take the survey, and over- or under-samples these groups to construct a sample more

representative of the population of interest.

allows us to circumvent one of the issues in previous studies of migration, namely small sample sizes. The survey-design allows us to control for a wide range of variables which would not be available in most administrative data sets. Using administrative data would however not leave us with the issue of hypothetical bias, which might be a concern in our data.

We define an immigrant as a respondent who states that they are not born in the country they are taking the survey in. If the respondent states that they are not born in the survey country, they are asked to provide information on their country of origin and how long they have stayed in the survey country. ³

For our purposes, it is vital that our migrant subsample of the ESS is representative of the migrant population in Europe, and we conduct tests to verify that this is the case. Similar to Just & Anderson (2012), we analyze whether the subsample we use is comparable to the European migrant population as a whole by verifying that the share of migrants in our sample and in Europe as a whole are sufficiently similar. We find that on a country level, the foreign-born share of the population in our sample does not deviate by more than a few percentage points for almost any of the countries (Eurostat 2019). The exception is Cyprus, which deviates by between 12 and 17 percentage points. ⁴ However, Cyprus is one of the smallest countries in our sample both in terms of sample size and population size and the likelihood of this deviation driving any results is therefore small. ⁵

3.2 Social Capital Variables

Since we are specifically interested in social capital investment, the measures of greatest interest are variables which require an active choice by the migrant, as opposed to for example trust levels which are arguably less of a decision and more a latent characteristic.

Because many other papers have focused on social trust when discussing social capital, we show the correlations between our outcome variables and social trust in Table 1. We show that all of the outcome variables are positively and significantly correlated with social trust.

One of our outcome variables is the variable measuring participation in social activities.

3 All questions used for the variables can be found in the codebook available for download at:

https://www.europeansocialsurvey.org/downloadwizard/. Further, a more detailed overview of how the variables we have changed are coded is found in Table A1

4 See Table A2 in the Appendix

5 The total number of migrants living in Cyprus in our data is less than 350 (less than 1% of our total

migrant sample).

Engagement in social activities is a measure of civic engagement and participation in the local community. We argue that this is a relevant measure of bridging social capital, since social activities commonly extend beyond one’s close circle of friends or relatives.

We also include a measure of migrant social networks. This variable is derived from a question on time spent with friends and family and measures a form of bonding social capital which may not build social capital across heterogeneous groups. As previously discussed, bonding social capital could however be positive if it leads to spill-overs into bridging social capital.

In order to obtain results which are easier to interpret and analyze, we modify the variables on social activities and time spent with friends and family by splitting the discrete categories into binary variables, where the cutoff is the median of the original variables.

Trade union membership among migrants is another variable which have been ex- tensively used in previous research as an indicator of social and political participation.

Although some authors have argued that membership in a trade union leads to too little interaction between groups of people and therefore that little social capital is created (Baub¨ ock 2006), being a member of a trade union has also been argued to increase in- clusion and social belonging in society, as well as expand an individual’s social network (Johnson & Jarley 2005). We create a dummy variable which takes the value one if the respondent is or has been a member of a trade union, and zero otherwise.

Another commonly used variable measuring social capital is participation in volun-

tary associations (Knack & Keefer 1997, Beugelsdijk & Smulders 2003). Participation in

voluntary associations show an engagement in the civic community and is a platform for

developing bridging social capital. Since it enables people to meet on a regular basis, it

creates trust across groups of people (Putnam 2000, Varshney 2003, Svendsen & Svendsen

2004, OECD 2012). In our data this variable is measured by the following question: “In

the past 12 months, how often did you get involved in work for voluntary or charitable

organisations?”. Theiss-Morse & Hibbing (2005) criticize the measurement by arguing

that voluntary associations do not necessarily involve diverse groups of individuals and

that participation not necessarily builds social capital by increasing trust between differ-

ent groups of people. In line with Nannestad et al. (2008), we argue that if participation

in voluntary organizations produces negative bonding social capital, the correlation be-

tween participation in voluntary organizations and social trust should be negative. ⁶ If on the other hand, voluntary participation is associated with positive bonding the same correlation should be positive. This is indeed what we find, see Table 1.

Table 1: Variable Correlations

Social Trust Social Activities Friends & Family Trade Union Voluntary Participation Non-Political Participation Newspaper

Social Trust 1

Social Activities 0.089

1

Friends & Family 0.070

0.23

1

Trade Union 0.039

0.000 -0.065

1

Voluntary Participation 0.099

0.167

0.079

-0.0132 1

Non-Political Participation 0.090

0.114

0.072

0.083

0.313

1

Newspaper 0.084

0.087

0.043

0.115

0.126

0.102

1

∗p < 0.05,^∗∗p < 0.01,^∗∗∗p < 0.001

A positive correlation could, however, indicate a reverse relationship. More trusting people could be more likely to participate in voluntary associations and more trusting people could have larger social networks. With regard to this, Brehm & Rahn (1997) argue that participation in voluntary associations causally creates trust to a larger degree than trusting people are inclined to participate in voluntary associations.

Since the majority of migrants do not participate in a voluntary organization there is very little variation in different levels of participation. For this reason, we create a dummy variable indicating whether or not the migrant has participated in any form during the last 12 months.

We further include a variable indicating if the migrant has worked in a non-political organization or association. The question starts with stating that “there are many differ- ent ways to improve things in [country] or help things from going wrong” and then asks if the respondents has worked in a non-political organization or association. The ques- tion therefore measures a sort of non-political civic engagement and investment into the host-country. Similar to voluntary participation, there is little variation in participation rates for those who have participated at all and we therefore create a binary variable that takes the value of one for any participation and zero otherwise.

Finally, we include a variable measuring newspaper reading. Newspaper exposure has previously been used as a measure of social capital, as it measures the degree to which the individual is an aware and participating member of the community (Brehm & Rahn 1997). Although newspaper reading might be increasingly irrelevant due to digitization, the survey question does not distinguish between physical and online newspapers and should therefore cover both types. As some previous authors have noted, it is possible

6 Social trust is measured by asking the respondent: “Generally speaking, on a scale from 1-10, would

you say that most people can be trusted, or that you can’t be too careful in dealing with people?”

that migrants only read ‘ethnic’ newspaper which would not indicate an investment in bridging social capital but rather a bonding type of social capital (OECD 2011). However, Jacobs & Tillie (2004) show that watching and reading ‘ethnic’ television and newspapers is associated with increased civic engagement. The measure might therefore be of interest even if some of the migrants are only reading ‘ethnic’ newspapers. Further, as seen in Table 1, daily newspaper reading is positively correlated with social trust. For daily newspaper reading we create a binary variable which takes the value of one for migrants who read newspapers daily, regardless of how much, and zero for non-daily readers. We do this because the variation in reading or non-reading is arguably more important than the variation in time spent reading.

3.3 Naturalization

Our independent variable is naturalization status. The variable describes whether or not the respondent is a citizen of the country in which the migrant is interviewed. We focus solely on first generation immigrants as they are the ones easily identifiable in the data and because the integration process of second and third generation immigrants is very different and should arguably be analyzed separately (Van Mol & De Valk 2016, Suarez-Orozco 2000).

As previously discussed, there is a risk of misreporting naturalization rates (Mazzolari 2006), since migrants may lie about their citizenship status for various reasons. Specif- ically, it may be the case that migrants who are in the country illegally state that they are naturalized because they do not want their lack of citizenship on record. Van Hook

& Bachmeier (2013) investigate this type of bias using administrative US data and the American Community Survey. They argue that misreporting only occurs for a very spe- cific subset of migrants, for example migrants who very recently arrived. The authors conclude that misreporting is not likely to affect results in a meaningful way. Further, the problem of illegal migrants is likely not very problematic in our setting as the ESS sam- pling is based on individuals, households or addresses data which should not be available for migrants who are not registered in the country.

The question asked in the American Community Service is similar to ours in that it

asks the respondent to provide information on when they ‘first arrived’ in the country. In

our sample of migrants, less than 12% have stayed in the country for less than five years,

suggesting that the group of migrants comparable to the group which was the cause of

misreporting in the US study is relatively small in our data. ⁷

3.4 Control Variables

There are several variables which might affect both naturalization and social capital in- vestment which could bias the result if left uncontrolled for. The control variables we choose to include and their potential effects on the dependent and independent variables are discussed below.

Host countries vary both in terms of how strict their naturalization requirements are and therefore, how likely a migrant is to naturalize, and also in the degree and development of social institutions affecting the variables we use as social capital measures. The extent to which people in different countries for example participate in voluntary organizations, trade unions or social activities is likely to differ by country and we therefore control for host country fixed effects.

Similarly, we control for country of origin fixed effects. Hainmueller & Hangartner (2013) find that the migrants’ country of origin is the most crucial determinant of citizen- ship in Switzerland, where naturalization is decided by a public referendum. Their finding suggest that country of birth is an important factor to control for if the public employee formally deciding on the naturalization decision discriminate (consciously or not) on who gets citizenship based on where they are born. Even if it is not the case that country of birth is a deciding factor in whether to approve a naturalization request, it could be the case that migrants from different countries have a different cultural heritage which makes them more or less likely to naturalize. The cultural heritage of a migrant is also likely to have an impact on participation in social groups or activities.

The difficulty of obtaining citizenship varies over years. The number of migrants and therefore the number of applicants vary over time and formal laws and regulations might also change over time. In most countries, naturalization rates are similar across years but the total number of applicants varies, which increases or decreases the difficulty for any one migrant to obtain citizenship. Because migrants often arrive in waves based on global events, it is important to control for these time-varying effects. It is also likely that investment into social capital varies across time, as noted by previous authors (Putnam 2000). For example, it is possible that participation in voluntary associations

7 We analyzed this conservatively. For example, if migrants stated they had stayed in the host country

for between 5-10 years, we regarded them as having stayed for five years only.

has changed substantially during the years of our survey because of the invention and use of social media. We will run regressions using survey year fixed effects as well as regressions with linear time trend to investigate the robustness of the results to either of the chosen specifications.

Another important control variable is the length of time the migrant has been staying in the country. Length of stay in country is measured by the question on what year the migrant arrived. Because of the way these questions are framed, it is possible that a migrant has lived in the country for several different time periods with time spent outside the country in between. Since this information is not available in the data, we will assume that the migrant has stayed in the country continuously from the year stated until the year of the survey. Temporary return trips should not have a significant impact on our results and are therefore not relevant for the purposes of our study.

Both gender and age are likely to affect naturalization decisions and social capital investments. Yang (1994) argues that men are more likely to naturalize than women because they are more likely to get jobs where the naturalization process pays off. Beve- lander & DeVoretz (2008) argue that age is of importance because older migrants have smaller incentives to naturalize as they have a shorter pay-off time. Both gender and age are also likely to affect the amount invested in social capital.

People with higher education levels are more trusting (Hooghe et al. 2009) and might also be different with regard to other forms of social capital. Education is also very likely correlated with naturalization decisions, as the education level determines both access and returns from naturalizing (Bevelander & DeVoretz 2008), and we therefore control for the migrant’s education level. Similarly, we control for whether one of the official languages of the host country is spoken in the migrant’s home. Since language skills are commonly a prerequisite for naturalization and also an indicator of integration, this needs to be included in the analysis.

There are two variables related to income in the survey. The most straightforward one

simply asks the migrant to estimate the yearly income of the household. Because of the

face-to-face interview style of the survey and the fact that income is a sensitive subject,

there might be substantial bias in this measure. The other income measure instead asks

about the respondents subjective feeling about their income. Although there are obvious

problems with this measure, such as the limited amount of information on actual income,

it exists for almost all individuals and is less likely to suffer from hypothetical bias. This

is therefore the measure we choose to include in the analysis. Previous papers using the

ESS have argued that when controlling for education the variable serves as a sufficiently reliable measure of material well-being (Hooghe et al. 2009).

Two variables whose importance are likely more dominant for women are marital sta- tus and number of children. As previously discussed, some literature using administrative data lack these variables which might bias the estimations for the female part of the sam- ple. Marital status is a key determinant of labor participation, and could similarly be a determinant of social capital investment. It is also likely correlated with naturalization decisions. The number of children in the household is a factor which likely affects both social capital investment and citizenship acquisition. Because there are a lot of missing values for the variable on marital status we will use a binary question on whether the re- spondent lives together with a partner, instead of formal marital status. For the number of children we will use a question on the number of people living in the household. We recognize that the number of people living in the household does not necessarily proxy the number of children. It could for example, show other relatives living in the same house- hold. Nevertheless, household size could be an important predictor of both naturalization status and social capital investment.

For the instrumental variable approaches we include the same set of controls as in the LPM. The country- and survey-year fixed effects are needed because the conditional mean of the error term likely depends on the policy change if they are excluded. There are further reasons for why the socioeconomic controls are important in the IV regressions.

Factors that vary over time and countries are not controlled for by the fixed effects. If, for example, there was a wave of young migrants from a particular country in a particular period from a country which allow dual citizenship this might render the instrument endogenous if we do not control for age.

3.5 Descriptive Statistics

From the descriptive statistics in Table 2 it is obvious that most migrants in the sample are already quite well integrated. For non-citizens, most people have been staying in the host country for more than 10 years and almost 33% have stayed in the host country for more than 20 years. For citizens, almost 70% have stayed in the host country for more than 20 years.

The same pattern is clear with regards to language usage. For non-citizens, 60% of

the migrants speak one of the official languages at home. As expected, this number is

higher for citizens where 74% of the migrants speak one of the official languages of the host country.

Further, migrants with citizenship are on average older, and more likely to be women.

The education levels are similar between the two groups. Migrants with or without citizenship do not seem to significantly differ in how they feel about their income, and they are about as likely to live with a partner. Both groups also have roughly the same household sizes on average.

Lastly, it is evident from Table 2 that citizens overall have higher social capital levels

compared to non-citizens. For all measures except one, the average social capital invest-

ment variable is higher for citizens than non-citizens. In the next section, we will proceed

to test whether this fact holds after employing our econometric approaches.

Table 2: Descriptive Statistics of Migrants in the ESS

Non-citizen Citizen

Social capital investment Obs Mean Obs Mean

Social Activities 12 321 0.55 20 312 0.58

Friends/Family/Colleagues 12 686 0.60 20 633 0.60

Trade Union 12 475 0.30 20 506 0.43

Voluntary Participation 2 977 0.30 4 991 0.33

Non-Political Participation 12 611 0.09 20 075 0.11

Newspaper 7 701 0.66 12 657 0.69

Socio-economic controls Obs Mean Obs Mean

Native Language (Official language = 1) 12 722 0.60 20 837 0.74 Age of Respondent (In years) 12 640 41.69 20 720 49.08 Gender (Male = 1, Female = 0) 12 714 0.48 20 820 0.43 Partner (Lives with partner = 1) 12 637 0.63 20 674 0.61

Household Size (# of people) 12 691 2.86 20 793 2.77

Length of stay Obs % of sample Obs % of sample

Less than one year 442 3.5% 72 0.4%

1-5 years 2 805 22.4% 586 2.8%

6-10 years 2 554 20.4% 1 373 6.6%

11-20 years 2 592 20.7% 4 546 22.0%

More than 20 years 4 105 32.9% 14 115 68.2%

Education Obs % of sample Obs % of sample

Less than lower secondary education 1 638 13.0% 2 237 10.8%

Lower secondary education 2 264 18.0% 3 381 16.3%

Upper secondary education 3 781 30.0% 6 848 33.0%

Post-secondary non-tertiary 650 5.2% 838 4.0%

Tertiary education 4 182 33.2% 7 359 35.5%

Other education 83 0.7% 70 0.3%

Income Obs % of sample Obs % of sample

Very difficult to cope on present income 1 150 9.2% 2 128 10.4%

Difficult to cope on present income 2 951 23.6% 4 526 22.2%

Coping on present income 5 185 41.5% 8 735 42.8%

Living comfortably on present income 3 195 25.6% 5 006 24.6%

Observations 12 722 20 837

Total observations 33 559

4 Method

4.1 Linear Probability Model

We start the analysis by running a number of LPM specifications in the form of eq. (6) described below:

Y _i = α + β ₁ N _i + X _i δ ⁰ + λ _c + ψ _o + γ _t + ε _i (6) Where Y _i is the outcome of interest for individual i. N _i is naturalization status, taking the value one for migrants who have acquired citizenship, and zero otherwise. X _i is a vector of the socio-economic control variables. λ _c are time-invariant host country fixed effects and ψ _o are time-invariant country of birth fixed effects, allowing us to control for factors varying over countries but not over time. γ _t are survey year fixed effects, controlling for factors that do not vary by countries but rather over time. ε i is the error term.

The LPM framework assumes a linear relationship, which puts certain restrictions on the model and the interpretations. This includes potential unreasonable predictions of outcome variables (predicted probabilities below 0 or above 1), imposing linearity on non-linear effects, and non-normality of standard errors. For this reason, we also run the same corresponding specification in a Probit framework as a sensitivity check.

An LPM model without covariates and a binary independent variable will have the same coefficient as the average marginal effect from a Probit model. Therefore, we expect that even with covariates, the difference between the Probit and LPM will be relatively small.

Factors that we are not able to control for in this specification are unobserved char- acteristics such as ambition or a willingness to integrate, which is likely to affect both the probability of naturalization for a migrant as well as the social capital outcomes of interest. For example, it is not unthinkable that migrants who make efforts to naturalize are more positively inclined towards social integration as well. This would lead to an omitted variable bias and an overestimation of the effect of naturalization in the LPM.

In addition, other authors, such as Hainmueller et al. (2017), have noted one additional

problem with citizenship acquisition, arguing that there is double selection into citizen-

ship. The first selection is the one discussed in the previous paragraph, and the second

is the selection where governmental agencies formally decide on which migrants retrieve

citizenship or not. However, in most European countries, i.e. the countries in our sample,

there are formal and specific requirements which need to be fulfilled in order to obtain citizenship. Many of these factors, such as length of stay in the country, are observed in our data and can therefore be controlled for. Further, we do not find any indication that government workers have the capacity to systematically deny citizenship to migrants who fulfill all the formal requirements, except for the case of Switzerland. The first type of selection is therefore arguably more severe for the purposes of this analysis.

4.2 Two-Stage Least Squares & Bivariate Probit

To deal with this selection, we propose an IV approach, in both a 2SLS framework, as well as a BP framework. The linearity problem discussed previously extends to the 2SLS case.

However, in contrast to the LPM case where robustness of coefficients is easily testable with a Probit model, the 2SLS case is a bit more complex. This is because the first stage of the IV regression depends on a linear prediction. Allowing for non-linearity in the second stage makes little sense after imposing linearity in the first stage. The BP model is an estimation strategy where two simultaneous Probit equations are run, similar to the first and second stage of a 2SLS. The correlation between unobserved factors in both Probit equations is identified and used to remove the selection bias in the naturalization variable. Further, the BP framework does not force the second stage to be estimated based on a linear prediction in the first stage and can therefore allow for non-linear effects. It has been argued to be a more suitable choice of specification for instrumental variable specifications where both the outcome and potentially endogenous regressor variables are binary (Rhine et al. 2006, Morris 2007, Angrist & Pischke 2008). On the other hand, it assumes a joint normal distribution of the error term with correlation ρ, which the 2SLS does not.

As for the instrument, we will use variation in whether the countries of origin permit dual citizenship. The variation comes from differences between countries as well as over time if the countries of origin have changed their legislation during the time period of the surveys. ⁸ Permitting dual citizenship reduces the costs associated with acquiring a new citizenship since the migrant is not forced to renounce her previous citizenship (Bevelander

& DeVoretz 2008, Liebig & Haaren 2011), and should not have an impact on the outcomes of interest other than through its effect on naturalization, when controlling for observable factors. Migrants from countries allowing dual citizenship should therefore, all else equal,

8 Countries which have changed their dual citizenship policy can be found in Table A4 in the Appendix.

have higher naturalization rates. Because countries which allow dual citizenship might be systematically different from countries where dual citizenship is banned, we utilize both changes in whether the country of origin permits dual citizenship, as well as control for country of birth fixed effects. Between countries that have always allowed and countries that have never allowed dual citizenship, we assume potential factors that differ to not vary over time. In our sample, there are over 30 countries of origin which have changed their dual citizenship policy during the time frame available in the ESS, and many more who are included as having always allowed or never allowed dual citizenship.

In the first stage, eq. (7), the instrument is based on whether the migrant’s country of origin permits dual citizenship. Using this variation, we create an instrument which is equal to one if the migrant is being surveyed in a year when her country of origin allows dual citizenship, or is surveyed at least one year later than when the change in country of origin dual citizenship policy took place, if the country has recently changed its legislation to allow dual citizenship. ⁹ Further, the instrument is only equal to one for migrants who have moved to a country which also allows dual citizenship. If the host country does not allow dual citizenship, there is no change in the migrants’ incentives to acquire citizenship in the host-country as the migrant cannot acquire the new citizenship without renouncing the old one.

Bevelander & DeVoretz (2008) note that there does not seem to be any clear re- lationship between formal citizenship legislation and aggregate naturalization patterns, and argue that other political, cultural, social and economic factors matter more than legal structure. Despite this, permitting dual citizenship has, surprisingly, seemed to be negatively associated with naturalization in the past (Yang 1994, DeVoretz & Pivnenko 2005). ¹⁰ These authors are not using any time-variation in citizenship policies, mean- ing the variation is based solely on cross-country differences. More recent papers which, similar to us, also utilize changes in dual citizenship laws show that permitting dual citi- zenship has been associated with increases in naturalization rates, as expected (Mazzolari 2009, 2017). The first stage of the 2SLS Instrumental Variable specification is shown in eq. (7).

N _i = α + β ₁ I _i + X _i δ ⁰ + λ _c + ψ _o + γ _t +  _i (7) Where the instrument can be described by the following indicator function:

9 We allow one year for the legislation to take effect. For the few countries that have changed their legislation to not allowing dual citizenship, the instrument is 0 as soon as the change took place.

10 The authors however find a positive correlation when they limit their sample to non-OECD countries.

I _i (OA _o,t , HA _h,t ) = 1[OA o,t = 1 & HA _h,t = 1] (8) Where OA _o,t is Origin country dual citizenship policy and HA _h,t is Host country dual citizenship policy, both taking the value of 1 if the country permits dual citizenship at time t.

The second stage is shown in eq. (9):

Y _i = α + β ₁ N ˆ _i + X _i δ ⁰ + λ _c + ψ _o + γ _t + υ _i (9) Where ˆ N _i is the variation in naturalization status cleaned from the endogenous part of the error term, as predicted by the first stage of the IV regression. υ _i is the error term

The BP follows in a similar fashion in eq. (10), using the same instrument I _i for naturalization as in the 2SLS:

N _i = 1[α + β ₁ I _i + X _i δ ⁰ + λ _c + ψ _o + γ _t +  _i > 0] (10) and eq. (11):

Y _i = 1[α + β ₁ N _i + X _i δ ⁰ + λ _c + ψ _o + γ _t + υ _i > 0] (11) Where N _i and Y _i are the binary outcomes and evaluate to one if the latent function within the brackets exceeds the threshold zero, and zero otherwise. In contrast to the 2SLS this specification is suitable for the binary structure and will assume a more appropriate functional form between zero and one. The correlation between the error terms  _i and υ _i will be used to conduct Wald tests for exogeneity, similar to the Hausman test statistic in the 2SLS case.

In the instrument, the criteria of host country allowing dual citizenship should in-

crease the explanatory relevance of the instrument and allow us to estimate a clearer

relationship, and we therefore choose to impose it, but it should be noted that it could

potentially introduce some endogeneity. If there are unobserved characteristics in some

migrant groups which determines whether they move to a country which allows dual

citizenship given that their home country also allows dual citizenship, this might be prob-

lematic. One example of this could be that migrants with higher unobserved willingness

to integrate decide to move to countries which allow dual citizenship with the intention

of naturalizing without having to renounce their previous citizenship. This unobserved

characteristic, which affects both naturalization and social capital investment, would ren- der the instrument endogenous. For this to be a relevant factor in the migrant’s decision, both the host country and the country of origin must allow dual citizenship at the time of migration, or the migrant must expect her home country to allow dual citizenship in the future. A migrant already living in the host country while the host country changes citizenship policy has already decided on the country of migration and will therefore not select based on whether the host country allows dual citizenship.

Ideally, we would identify all potential migrants for whom this selection could have been an issue, drop them from the analysis and evaluate the change in the results. How- ever, we are unable to collect information on the exact dates for changes in dual citizenship legislation for many of the host countries in our analysis. Instead, to shed some light on the likelihood of this being a problem, we analyze the observable characteristics of mi- grants who come from countries that allow dual citizenship and whether they differ based on if they migrate to countries that also permit dual citizenship. If there is a difference between these groups in countries of origin that allow dual citizenship, but not in countries of origin that does not allow dual citizenship, it might be an indication of that the groups also differ in unobservable characteristics. This in turn could indicate that bias-inducing selection has taken place. However, if there is no difference or if the difference can be observed both in countries of origin that allow dual citizenship and countries of origin that do not allow dual citizenship, it could just as well be the case that some groups of migrants just tend to migrate to certain countries. This would not be problematic for the analysis since we control for country fixed effects. We investigate this issue in Figure (A3) in the Appendix and find that migrants from countries of origin which allow dual citizenship who live in host countries that allow dual citizenship differ from migrants in host countries that do not allow dual citizenship. However, the same is true for migrants from countries that have never allowed dual citizenship. Thus, it is not clear to us that this indicates a potential harmful selection by migrants that we are unable to control for.

Before moving on to the result section, it is important to note that when comparing

coefficients, the 2SLS evaluates the effects of naturalization for those migrants that were

affected by the change in citizenship policy in their countries of origin. Changes in country

of origin citizenship policies can be seen as an ‘as if random’ intention to treat. From this

randomization of the intention to treat, we estimate the local average treatment effect

(LATE). It is an indication of how the migrant on the margin is affected, i.e. those

who decided to obtain citizenship only after the change in their country of origin policy.

This is not necessarily a downside as this group could be particularly relevant for potential policy targeting. If naturalization for this group is positively associated with social capital investment, increasing the incentives for migrants at the margin to naturalize might be a good way to increase investment and better integrate migrants in their host societies.

The BP on the other hand estimates the average treatment effect, which is the sample average effect of naturalization on the social capital outcomes of interest (Chiburis et al.

2011). This effect is comparable to the LPM coefficients which also estimate the average

treatment effect.

5 Results

Figure 1: Effect of Naturalization on Social Capital Investment (a) Linear Probability Model

Social Activities

Friends & Family

Member Trade Union

Voluntary Participation

Non−Political Participation

Newspaper

−.02 0 .02 .04 .06 .08

(b) Two-Stage Least Squares

Social Activities

Friends & Family

Member Trade Union

Voluntary Participation

Non−Political Participation

Newspaper

−4 −3 −2 −1 0 1

(c) Bivariate Probit

Social Activities

Friends & Family

Member Trade Union

Voluntary Participation

Non−Political Participation

Newspaper

−.1 0 .1 .2 .3

Note: Standard errors clustered on survey-country level. Confidence Intervals (95 %).

Specifications include survey-country, country of birth and survey-year fixed effects as well as socioeconomic controls. ESS post-stratification weights are applied.

In Figure (1) we show the main results of our three models. One of the most obvious

patterns in the LPM specifications is that naturalization is positively associated, although

not significant in all cases, with all of the social capital outcome variables of interest. The

relationship is relatively small in magnitude for all variables in the LPM. Social activities,

participation in non-political organizations and newspaper reading are positively related

with citizenship in the LPM, and trade union membership and newspaper reading has a

positive relationship with citizenship in the 2SLS. In the 2SLS the estimates and standard errors are however greatly inflated. In the Bivariate Probit in Figure 1 (c), the magnitudes are closer in size to those of the LPM, although the only effect which is significiant in both of the Bivariate Probit regressions (first and second stage) is the effect of non-political participation.

In Tables (3), (4) and (5) more detailed regression results for the three different models are presented.

5.1 Linear Probability Model - Results

In Table (3), we show the coefficient of interest, as well as the socioeconomic controls included in all specifications.

In the LPM model, citizenship acquisition is associated with a 3.5 percentage point increase in the likelihood of spending ‘about the same’ or more time than other people on social activities, statistically significant on the 1% level. ¹¹ Citizenship is also significantly and positively associated with the probability of participating in non-political organiza- tions and newspaper reading, with an increase of 2 and 3 percentage points, respectively.

In the LPM, we find no statistically significant effect of naturalization on spending time with friends, family or colleagues. The coefficients on being a member of a trade union, or the probability of participating in voluntary organizations are also insignificant.

From Table (3) we see that speaking one of the official languages of the host-country at home is positively associated with four of the six social capital measures. The positive sign is expected as speaking an official language at home is a clear sign of integration which should correlate with social capital investment. Less intuitive is the correlation between gender and social capital. Men seem to invest more in social capital than women, correlating positively in five of the six specifications.

Length of stay seems to be important for trade union membership, and participation in voluntary and non-political organizations. The correlation is strongest for those who have stayed the longest in the country, which is what we expect.

Migrant age and social capital outcomes are ambiguously correlated. Some measures, such as spending more time on social activities, friends & family as well as voluntary organization participation are negatively correlated with age. Trade union membership and newspaper reading are however positively correlated with the age of migrants.

11 Compared to how much time the respondent thinks other people in general spend on social activities.

As expected, education is positively and significantly associated with many of the social capital variables. Further, the association seems to be stronger for higher levels of education, with tertiary education having the largest correlation with social capital outcomes.

Income also shows the expected sign and is positively associated with five of the outcomes. Finally, we find no clear relationship between the living arrangement variables and the outcomes of interest, as the coefficients vary in sign and significance over the specifications.

One thing to note is that the numbers of observations differ substantially for some of the outcome variables since some of the questions associated with the outcome variables are not asked in all survey waves. Although fewer observations lowers the statistical power of the analysis, we do not expect any bias to stem from this loss of observations, as the reduction can be considered random after controlling for observable factors.

The R ² ranges from 0.06 to 0.27 depending on the outcome variable. Relatively little (6% - 27%) of the variation in social capital investment is therefore explained by naturalization status and the other independent variables.

We also test whether all LPM results are robust to corresponding Probit specifications.

We do not include the results in the tables, since as expected the average marginal effects

from the Probit almost exactly follow the linear estimates.