Kees Nijhoff (10262385) Supervisor: dr. Bram Lancee
Co-supervisor: dr. Levi van den Bogaard Graduates School of Social Science: Sociology 26-06-2017
Universiteit van Amsterdam
Bridging the Gap
The Effects of Social Capital on the Labour Market Success of Turkish and Moroccan Immigrants in the Netherlands
Abstract
Using a combination of fixed- and random effect analyses, this thesis analyses the effects of bonding and bridging social capital on the labour market outcomes of immigrants of Turkish and Moroccan descent in the Netherlands. Longitudinal data from the Netherlands
Longitudinal Lifecourse Study is used in answering the question to what extent the differing
forms of social capital have an effect on income, ISEI score and the likelihood of employment. The results are inconclusive; bridging social contact with other ethnic minorities seems to have a positive effect on ISEI score. While bonding social capital can have both a positive and a negative effect on ISEI score, depending on the location that the social contact takes place. Conversely, contact with other ethnic minorities reduces the likelihood of being employed, while intra-ethnic bonding is found to have a positive effect on the likelihood of employment. Contact with the native Dutch population seems to have no effect on the labour market outcomes of Turks and Moroccans. Regarding the effects of ethnic bonding and bridging social capital on income, due to inconsistencies across the different analysis methods, the results are deemed inconclusive.
Content
Introduction………3
Theoretical Background……….5
Methods………12
Results………..22
Conclusion and Discussion………..28
Introduction
Migration is a topic that has always been of great interests to researchers, politicians and the general public alike. However, lately it seems that wherever one goes in the public domain the issue of migration is unavoidable. It is estimated that from 1980 to 2005 the foreign-born population of the world almost doubled to a staggering 191 million. In Europe there was an even larger relative increase, during the same period the amount of immigrants tripled from almost 21 million to 64.1 million (Longhi, Nijkamp, & Poot, 2010).
In more recent years the amount of migrants in Europe seems to have grown to even more impressive numbers. Even a small country like the Netherlands has seen great influx of new settlers. An article in het Algemeen Dagblad tells us; the number of migrants who settled in the Netherlands last year (88,000) exceeds the natural birth-rate in the Netherlands (22,000) (van der Mee, 2017). This great influx of migrants has brought with it interests in the economic and social consequences of this development; how can host countries properly integrate these newcomers into their society and labour market?
This thesis will look at migration from the crossroad between economic and social science. In particular, it will examine the relationship between migrant „labour market success‟ and their social capital. Taking a closer look at what factors might have a possible effect on the labour market success of migrants is especially relevant since immigrants are a group of people that has traditionally struggled a great deal in regards to their economic situation. According to an OECD report the unfavourable labour market position of immigrants in the Netherlands is not a recent phenomenon. Significant gaps between the native Dutch and immigrants, in regards to their respective labour market positions, have been observed since the recession of the early 1980‟s, which hit migrants disproportionately hard (OECD, 2008).
In this paper the focus will be on two of the largest migrant groups in the Netherlands, namely; Turks and Moroccans. The first major waves of Turkish and Moroccan immigrants arrived in the Netherlands in the 1960‟s, when the Dutch government made a labour agreement with both countries. These immigrants were mostly men with a poor educational background, recruited to preform unskilled labour. It was expected that these migrants would return home after a period of work, however many elected to stay. Due to family reunification processes and other family arrangements the amount of immigrants from these countries continued to increase, even after the labour agreement was ended (Lancee, 2012a).
The Netherlands is one of the European OECD countries with the highest naturalisation rates among immigrants. However, the benefits of this naturalisation in terms of
better labour market outcomes are not distributed evenly among immigrant groups. Turkish and Moroccan immigrants in particular do not enjoy a naturalisation premium in terms of higher wages or employment probabilities (OECD, 2008). Among other reasons, this lower premium of naturalisation among these two ethnic minority groups, makes them an even more salient object of study.
This research will approach social capital from an individualistic position, as opposed to a community property. This individualistic view on social capital sees it primarily as ties in a person‟s network. We will use longitudinal data from a set of Dutch surveys (Netherlands
Longitudinal Lifecourse Study 2009 & 2014) on migrants, in order to examine what is distinct
about the networks of some migrants which allow them to prosper and thrive in the Dutch labour market, while other migrants seem to struggle. The current Dutch labour market system is one that is primarily a market-based system, which tends to disfavour immigrants as they accumulate multiple disadvantages compared to the native Dutch population. Among these disadvantages are language problems, lack of knowledge about the local labour market functioning etc. (OECD, 2008). Social capital could offset some of these disadvantages, as immigrants could possibly use their connections with the native population to gain specific knowledge about the Netherlands. As such the central question around which this thesis will be structured reads as follows:
What are the effects of (ethnic-) bonding and -bridging ties on the labour market success of immigrants from Turkish or Moroccan descent in the Netherlands?
In this study labour market outcomes will be measured with the following three variables;
income, employment status and the International Socio-Economic Index (ISEI). This thesis is
not the first paper on this subject; many researchers before us have tried to answer this question or a similar one (e.g. Ryan et al. 2008; Kanas et al. 2012; Lancee, 2012a). These related studies, and more, will be discussed more in depth in the upcoming section. However, these previous studies do not make additional research on this topic redundant, this thesis hopes to contribute to the field of knowledge in several ways; firstly the data used is relatively recent (the latest wave of the survey being measured in 2014). In this globalized age of the internet our society seems to be changing at a rapid pace, using more recent data to test „older‟ theories is thus a worthwhile endeavour. Much of the previous research on bonding and bridging social capital only investigates one form of interethnic bridging social capital, namely interethnic contact with natives of the host country. Our data and operationalization allows us to differentiate between two forms of interethnic bridging contact, bridges with
natives and other ethnic minorities. This is relevant since, per the resource argument (Lin, 1999a) which will be outlined in the upcoming section, these two forms of bridging social contact can have very different effects. Additionally, this thesis differentiates between two forms of bonding social contact; intra-ethnic bonding and familial bonding. Lastly, whereas much previous research only uses one type of analysis; this research employs two methods of analysis, namely random effects and fixed effects. By using two methods of analysis, instead of one, more robust results will hopefully be generated. In these ways this paper hopes to contribute to the theoretical landscape regarding the effects of social capital on immigrant labour market success.
The next section of this thesis will outline some possible ways to answer the question of how certain network ties might lead to better labour market outcomes, as well as outline some of the results of previous research on this topic. In the methods section, we will give a description of how this research was operationalized, as well as give a detailed description of the variables used in the models. Next the results will be presented, followed by the conclusion.
Theoretical Background
In the late 1980‟s Coleman wrote the influential article Social Capital in the Creation of
Human Capital (1988), in which he explained the different ways in which a person‟s social
capital can be used to the benefit of that individual or their community. According to Coleman there are three forms of social capital that can be mobilised by an individual or community: Obligations, Expectations and Trustworthiness of Structures (1), Information
Channels (2) and Norms and Effective Sanctions (3) (Coleman, 1988).
These forms of social capital work in conjunction with two specific social structures that facilitate social capital, namely: Closure of Social Networks and Appropriable Social
Organizations (Coleman, 1988). When applying Coleman‟s theoretical lens of social capital
to recent migration trends we are confronted by a seeming paradox. Migrant communities are often vibrant communities, dense with a high amount of social closure, in which the members have frequent contact with each other (Waldinger, 1997). In other words, members of migrant networks typically have a high amount of social capital on which they can draw. As these are networks with a high amount of closure, they should, according to Coleman‟s theory of social capital, be successful and yet often they are not. (Coleman, 1988). A possible answer to this question can be found in Granovetter‟s work. In 1973 he wrote an article called The Strength
community) to get „ahead‟, but instead it‟s the weak ties that bridge a person to a different network from which people can stand to gain the most (Granovetter, 1973). To illustrate; A large network that is strongly interconnected, but has little contact with different networks on the outside, will not diffuse information generated within the network to the rest of the wider community. But more importantly, new information from the wider community will not reach the network. In contrast, a network that is perhaps less interconnected internally, but instead has ties reaching out to different networks will receive this new information (Granovetter, 1973).
The power of weak ties might explain how migrant communities, while rich in social capital, can still be disadvantaged. While migrants might have a vast network of social capital at their disposal, the social capital they have available does not often bridge to other social strata, thus it will not generate new information. This phenomenon of bridging social capital being more advantageous in nature, than bonding social capital, has also been studied by other scholars. A classical work in this field is by Ronald Burt who focuses on structural holes in networks. Structural holes are gaps in network-nodes between certain demographics (e.g. ethnicity, social class etc.). Agents who obtain network ties that bridge these structural holes stand to gain from them, as they are in a prime location for new information (Burt, 2001). A practical example of an advantage that a bridge across a structural hole can give you regarding the topic of this thesis; a migrant might get some insight into the ebbs and flows of the local labour market by having a tie to a native citizen, thus gaining an advantage over other migrants.
Some recent studies into this phenomenon seem to corroborate this assertion that bridging social capital has certain advantages. For example, a (qualitative) research into Polish migrant networks in England concludes that while ethnic migrant networks help provide basic requirements such as a (low paying) job and housing, they may lock migrants into ethnic specific niches (Ryan, Sales, Tilki, & Siara, 2008). Migration networks function as “personal information fields”, which lead to newcomers possessing incomplete information about their options. This incomplete information in turn leads to channelization, the results of which are diminishing geographical diffusion and eventually, clustering in a small set of occupations and industries (Waldinger, 1997). Because migrants often possess a low amount of human capital the industries in which they end up clustering will often be those on the lower end of the pay grade or sectors in which they are subject to highly exploitative employment (Leonard, 2004). A way to escape this network of possibly strong, but „low quality‟ bonds is through the use of weak ties or bridges across structural holes. A migrant
could, for example, use his connection with a native to breach into a new sector. In the words of Robert Putnam; bonding social capital (networks with people who are like us in some important way) is good for „getting by‟, but bridging social capital (networks with people are
not like us in some important way) is essential for „getting ahead‟ (Putnam, 2000).
The utility of social capital
Nan Lin is an influential scholar in the area of social capital as a resource. His work will be used in theorizing how different kinds of social capital and ties can be used as a resource by agents and thus lead to better labour market outcomes. Lin defines social capital as follows: “Resources embedded in a social structure which are accessed and/or mobilized in purposive
actions” (Lin, 1999a). Lin, taking a methodological individualistic position, thus views social
capital primarily as the resources embedded in social networks (Lin, 1999a; Lin, 1999b). In general there are three ways in which the resources embedded in social networks lead to better outcomes; (1) it facilitates the flow of information, (2) ties may exert influence & (3) social ties might act as an individual‟s social credentials (Lin, 1999a).
(1) Social ties located in certain strategic locations (such as structural holes between networks, which need to be bridged (Burt, 2001)) and/or hierarchical positions, can provide an individual with useful information about opportunities and choices which he would not have access to otherwise. (2) Social ties can exert influence on actors who have a crucial role in the decision making process (for example, job recruitment). Due structural power imbalances or simply due to strategic positioning some actors are more influential and thus a more valued resource. (3) Social ties and their acknowledged relationships to the individuals may be conceived as a certification of an individual‟s social credentials, thus reassuring possible employers. In short, the status of a social contact is used to influence one‟s own status (Lin, 1999b). Following Lin‟s line of reasoning it is clear how some social capital can be more useful than others, as certain kinds of actors possess more information, can exert more influence to the right kind of people etc.
Keeping this framework of Lin in mind, it is highly likely that, due to structural power imbalances in society those immigrant networks, rich in bonding social capital as they might be, are poor in social capital that allows them to make gains on the labour market. As migrants in general are found on the lower rungs of society and thus have little (or not the right kind) of information, can exert little (or not on the right persons) influence and don‟t have the status themselves to vouch for an individual‟s social credentials. Earlier research seems to confirm this; research done in Sweden, on the differences between the mobilisation
of social capital between migrants and natives, concludes: “Even though use of informal
job-finding methods increases the likelihood of job-finding any job, it is not the job-job-finding method per se that leads to obtaining better jobs: it is, rather, the composition of social networks that determines job quality […] In other words, ONW [outside of North-West Europe and North America] immigrants have a social capital deficit which arises because they are embedded in social networks that limit their ability to gain valuable social resources” (Behtoui, 2008).
Isolated closure and bridges to nowhere
I call the dual nature of bonding social capital isolated closure, combining two, often cited, competing arguments regarding the nature of bonding social capital; these two arguments being; the isolation argument and the closure argument (Lancee, 2012a). The isolation argument states that within-group connections do not result in new information reaching the immigrant and thus does not allow them to make headway on the labour market. This harkens back towards the importance of weak ties or structural holes as theorized by Grannovetter (1977) and Burt (2001). Additionally the high amount of social control might impede successful mobility due to social obligations or „downward levelling norms‟ (Lancee, 2012a). In opposition to this stands the closure argument, which states that a network with high solidarity, as is theoretically often the case with regard to ethnic bonding social contact, results in in positive gains on the labour market success of immigrants, Coleman was a prominent proponent of this argument in the late 1980‟s (Lancee, 2012a).
Combining the two, isolated closure thus refers to the fact that while it is true that high amounts of ethnic bonding social contact might lead to more chances to be employed, these jobs aren‟t the best paying or jobs that are very highly regarded; thus leading to a possible dual nature of bonding social capital. Migrant bonding social capital might be positively related to employment chance, but negatively with other measures of labour market success; such as, income or occupational status.
Following the isolated closure argument formulated above we expect the following relationships between income & ISEI and ethnic bonding social capital. Hypothesis 1 reads as follows:
1. Ethnic bonding has a negative effect or no effect on both ISEI and income. (Hypothesis 1)
As theorized, migrants deeply embedded in „migrant networks‟ might become „trapped‟ in these networks, funnelling them into certain sectors of the industry which are often low-skilled and low-paying. As such, ethnic bonding social capital is thought to have a negative or no effect on ISEI and income (hypothesis 1), however the isolation of these ethnic
communities might lead to migrants having an easier time getting a job, low paying as it might be. This same argument can be made for familial contact, often thought to be the „purest‟ form of bonding social capital. The so called isolated closure effect mentioned earlier. Thus, regarding employment status, the second hypothesis of this thesis is as follows:
2. Ethnic bonding has a positive effect on the likelihood to be employed. (Hypothesis 2)
Bridging social capital in its turn might also not have one universal effect. Following the resource argument made by Lin (1999a) it is likely that, for immigrants, ties to the native population might be beneficial. These ties allow the immigrants access to resources they might otherwise not have had, such as; specific knowledge about the local labour market, language skills or an increased access to local labour market channels, to name a few. As such the third and fourth hypotheses of this thesis are the following:
3. Bridging ties to the native Dutch population have a positive effect on ISEI and income. (Hypothesis 3)
4. Bridging ties to the native Dutch population have a positive effect on the likelihood to be employed. (Hypothesis 4)
However, not all bridges are equal. Interethnic bridging ties to other ethnic minorities very likely do not confer on the immigrants the same amount of resources that ties to the native population do. One might call these interethnic bridging ties to other minorities,
bridges to nowhere. These bridges to nowhere are not very likely to lead to the same benefits
that bridging ties to natives do. Even though, technically, they are bridging spanning structural holes, with regards to the labour market they are not likely to lead to any significant advantage; as other migrant groups also do not possess great amounts of information about the local labour market. In fact, they might even be more detrimental than ethnic bonding ties, as these bridges to nowhere are both resource poor and not associated with a high amount of network closure. As such the last two hypotheses of this thesis are the following:
5. Bridging ties to other ethnic minorities have a negative effect or no on ISEI and income. (Hypothesis 5)
6. Bridging ties to other ethnic minorities have a negative or no effect on the chance to be employed. (Hypothesis 6)
Social capital, labour market and immigrants; results from a selection of previous studies
Above we have already shown some results of previous research on the effects of bonding and bridging social capital on the labour market success of immigrants (Behtoui, 2008; Ryan, Sales, Tilki, & Siara, 2008). However, in this section we wanted to give a short run down of
the results of some more research done in this field; in order to give the reader a sense of the theoretical landscape around social capital and immigrant labour market performance. It is beyond the scope of this section to address all, or even most, research done in this area of interest. Instead this short and limited overview is used to give the reader a short impression of previous results, and what might be expected of this thesis.
Kanas et al. used panel data and a longitudinal research design to determine what the effects were of bonding and bridging (with natives) social contact on the labour market success of immigrants in Germany (Kanas, Chiswick, van der Lippe, & van Tubergen, 2012). They conclude that bridging social contact (with natives) has a positive effect on occupational status (and on income using a separate model with sheaf coefficients). They measured bridging social capital as having a German partner and frequency of contact with German‟s. They did not include a measure of bonding social capital per se, but did have a measure of „general‟ social contact. On this front they concluded that social contact had a positive effect on several measures of labour market success.
In a couple of previous studies, Aguilera (2002) and Aguilera & Massey (2003), confirmed that, in regards to social capital returns, African Americans received a greater return on social capital than whites, an increase in social capital gave this ethnic minority a relative bigger benefit in regards to some forms of labour market outcomes (Aguilera, 2002). The 2003 study done by Aguilera and Massey shows that having friends and family with migratory experiences improves the labour market outcomes for Mexican immigrants in America (Aguilera & Massey, 2003). This last study seems to confirm the benefit of bonding social capital for migrants labour market succes, while the study of Aguilera (2002) shows that social capital does not have a static effect across all ethnicities.
In a study preformed on Asian immigrants in Los Angeles Sanders et al. (2002) conclude that family- and ethnic-based social networks, through their closure, influence the incorporation of these immigrants into their host societies. Providing groupbased resources that facilitate the advancement of immigrants in their new society (Sanders, Nee, & Sernau, 2002). Showing that for immigrants bonding social capital might also have advantages on the labour market.
Lastly, a qualitative study preformed by Iosifides et al. (2007), focussing on the incorporation of Albanian immigrants in Greece, concludes that; bonding social capital has a impact on the labour market incorporation of immigrants but enhances the phenomenon of „ethnic specialication‟, or as ethnic niches as they were called earlier in this thesis. No results are reported of the effects of bridging social capital on labour market outcomes, but they do
mention that a increase increase in inter ethnic contact with Greek natives, led to reduced discriminatory practices at work (Iosifides, Lavrentiadou, Petracou, & Kontis, 2007).
The purpose of this section was to give the reader some insight into previous research done in the field of social capital and migrant labour market preformance. The selected studies were chosen because they represent the migrant experience in a few different countries. As to give the reader a idea of the scientific landscape beyond the area under study in this thesis, we will return to some of these observations in the discussion at the end of this thesis. Before turning to the method section this thesis will go into detail on one more study, namely Lancee‟s (2012a).
Social capital, labour market and immigrants; previous research in the Netherlands
The topic of this thesis has been researched before, namely by Lancee (2012a). His research into the effect of social capital on the labour market success of immigrants in the Netherlands yielded the following results; bridging social capital allows immigrants to make headway on the labour market, being associated with a higher income & occupational status, as well as a higher likelihood of being employed. On the other hand bonding social capital is generally not associated with better labour market outcomes (Lancee, 2012a).
This thesis builds on the work by Lancee in several ways; firstly, it use more recent data. Secondly, it expands the operationalization of bridging and bonding social capital in several ways. Whereas Lancee measured bonding social capital as mostly familial bonding social capital, this thesis employs multiple measures of bonding social capital; familial bonding and intra-ethnic bonding. The concept of bridging social capital is also expanded upon; this thesis makes a distinction between bridging social contact with Dutch natives and bridging social contact with other ethnic minorities. This distinction is especially relevant since even tough contact with other ethnic minorities is bridging social capital; this form of social capital usually does not come with the resources that contact with the native Dutch population provides. As such, it is conceivable that these two forms of bridging social capital have very different effects.
Lastly, whereas Lancee‟s research regarding the Netherlands was cross-sectional, this thesis borrows an analytical technique used by Lancee (2012b) in a subsequent research, regarding social capital; namely the combination of both fixed effect and random effects analyses. This combination of different analytical techniques, only possible when using longitudinal data, allows us to get a more complete picture.
However, when compared to Lancee‟s (2012a) research, this thesis also has several limitations. Lancee differentiates between two forms of social capital; structural and
cognitive. Structural social capital refers to the wires in the network, the ties between people
that make up a network of connections. While cognitive social capital is the nodes in a network; attitudes and values that facilitate the exchange of resources (Lancee, 2012a). A very useful distinction as these different forms of social capital can have differing effects. Secondly, Lancee‟s measures of social capital are more thorough, where this thesis only uses self-reported frequency of contact as a measure of social capital, Lancee combines more variables. His scale forms a possibly more robust measure of social capital than the one employed in this thesis. However, as said above, the methods employed in this thesis do allow us to expand the concept of bridging and bonding social capital.
All in all, the differing operationalization of this and Lancee‟s research both have merits, as well as their own strengths and weaknesses. These strength and weaknesses allows the two studies on the same topic to complement each other, without making either redundant.
Methods Data
To answer the research question lined out in the first part of this thesis, data from the
Netherlands Longitudinal Lifecourse Study (hereafter; NELLS) will be used (Tolsma,
Kraaykamp, de Graaf, Kalmijn, & Monden, 2014). The NELLS is a panel dataset, which includes two waves (one in 2009 followed by a second one in 2014). The survey focuses on three main themes; social cohesion, norms & values and inequality. Due to these central themes and the longitudinal nature the NELLS survey is an ideal dataset for answering the central question of this thesis, namely; how does the social contact of migrants affect their labour market success? The NELLS data was recorded using a mixed-mode format, combining face-to-face and internet methods. In addition to this the number of respondents in the, unaltered, dataset is over 5,000 (Tolsma, Kraaykamp, de Graaf, Kalmijn, & Monden, 2014). However, for this study the sample will be reduced significantly, since we are only interested in Turkish and Moroccan respondents, which resulted in a total of 1,946 lost observations. Additionally, for the ISEI and income analyses the sample is reduced even further by only including respondents who have a job, which resulted in a further loss of 471 observations. In addition to this, respondents who had missing values on any of the relevant variables were excluded, which resulted in a loss of 168 observations for the ISEI & income sample and 358 for the employment sample. Students, housewife‟s and other respondents who
are possibly not active on the labour market are included in the employment status sample. This concession had to be made due to the fact that only wave 1 of the NELLS included a question asking the respondents if they were actively looking for a job. Additionally, the sample also suffered from attrition, which will be discussed more in depth later. This leaves us with a final sample of 244 individuals in the income & ISEI sample and 525 individuals in the employment sample.
The longitudinal character of this panel dataset is relevant, since it allows us to employ methods of analysis that look at within (respondent) variation, as opposed to cross sectional data which only allows us look at between (respondent) variations. This thesis will employ both fixed effects analyses and random effects analyses. In the next section we will outline the benefits and limitations of both methods.
Fixed effects versus random effects
As stated above fixed effect models can only take into account within respondent variation. In practice this means that variables which are time invariant, or in which there is very little within variation, can‟t be used in these models. For example, the effects of variables like gender & educational attainment (most people don‟t change their educational level after a certain age) can‟t be taken into consideration in these models. At first glance this might seem like a major detriment to using these types of models. However, only looking at within variation has the benefit of automatically controlling for all types of time-constant variables. Thus, removing any form of unobserved time-constant respondent heterogeneity. Additionally, employing FE models allow us to get more insight into causality. In addition to automatically controlling for all time constant variables, by only looking at within variation we know that a change in X goes together with a change in Y and that they are not caused by some exogenous circumstance that occurred before the measurements took place. FE models ability to remove any form of heterogeneity bias, by automatically controlling out time constant characteristics, has allowed this type of modelling to become the „gold standard‟ in many disciplines (Bell & Jones, 2015).
In random effects models the unobserved variables are assumed to be statistically independent of all the observed variables (Williams, 2016). In the social sciences however this is rarely the case, as controlling for all possible confounders is mostly an insurmountable task. Still a great number of publications in the social sciences rely heavily on random effects models. There are many reasons for this, but one of the most prevailing seems to be that, due to using cross-sectional or non-panel data, using a fixed effects model is impossible. This is
because repeated observations from the same individual are needed to employ fixed effect models. However, random effect models also have several advantages; the available N is most likely larger, as panel attrition and the like won‟t be of influence on your useable data. In addition to this, random effect models produce a smaller standard error (Williams, 2016). Additionally, RE models have a greater flexibility and generalizability (Bell & Jones, 2015). Some researchers even argue that in controlling out context, FE models cut out much of what is going on, and these goings-on are usually of interest to the researcher, the reader and the policy maker. Arguing that models that implicitly control out time constant context variables (gender, education etc. etc.) offer overly simplistic results (Bell & Jones, 2015). Lastly, for the great many papers that are interested in the estimation of the effects of time constant variables, using FE models is not an option. This thesis will make use of both random- and fixed effects models, not only to show the difference between the two methods. By employing them both we can get more insight into the relationship between our variables, as well as having the benefits of both methods, while keeping their limitations in mind.
Variable description & transformation
To measure labour market success this study employs three dependent variables; ISEI, income
& employment status. The International Socio-Economic Index of occupational status (ISEI)
is a socio-economic index score that refers to an occupation‟s main antecedent (education) and main consequence (earnings) at its core; it is not just a measure of occupational prestige. The ISEI score can thus be thought of as the cultural and economic resources that are typically associated with certain occupations (Ganzeboom, 2010). The NELLS dataset included variables which associated an ISCO-08 score with the occupation of the respondents. Using a transformation scheme these ISCO-08 scores were transformed into ISEI-08 scores (Ganzeboom & Treiman, 2010).
Income is measured in the NELLS dataset as the total household income of the
respondents and was recorded as a categorical variable. For the purposes of the analyses this categorical variable was recoded into a numerical scale variable using the median of each income category. To get a measure of the effective purchasing power that the familial income represents the economies of scale have to be taken into account. The same amount of money will result in differing in purchasing powers, as the needs for housing, electricity etc. grow in a non-proportional manner with an increase in the amount of household members. Following recent OECD publications, this thesis will employ a simple scale which divides the total
house hold income by the square root of the total number of household members (OECD, n.d.).
Due to an error in the dataset the variable directly measuring the employment status of the respondents contained a lot of missing values. To counteract this error, a new variable, measuring the employment status of the respondents, had to be constructed. The respondents who were employed at the time of the interview were asked if they were self-employed or on a payroll. Respondents who answered this question with either self-employed or on a payroll were classified as employed. While, respondents who were not asked this question were classified as not employed.
Social contact, as measured in the NELLS, takes the following form; respondents were
asked how often they were in personal contact with a person from X ethnicity at Y location. The following ethnicities were represented in X: Dutch, Turkish, Moroccan,
Surinamese/Antillean and Other non-Western. The different locations (Y) for which this,
self-reported, social contact was measured were the neighbourhood, work/school,
voluntary/leisure associations. To indicate how often they were in personal contact with X
ethnicity at Y locations, respondents could give the following answers for each XY combination; Never, About once a year, A few times each year, About once a month, A few
times each month, Once or a few times each week & (almost) Every day
From these measures it was possible to differentiate between the three forms of social contact of interest of this thesis; intra-ethnic bonding, interethnic bridging & Dutch bridging. As well as allowing us to differentiate between social contact at different locations. Intra-ethnic bonding measures the contact that the two Intra-ethnic groups, Moroccan and Turkish, have with their own ethnicity, so Moroccans with other Moroccans and Turks with other Turks. Interethnic bridging measures contact with all other ethnicities but the respondents own and the Dutch. Dutch bridging measures contact with the native Dutch population. The choice was made to differentiate between interethnic- and Dutch bridging social contact, since these concepts are theoretically distinct and are hypothesized to have different effects. Contact with the native Dutch population is thought to come with benefits that interethnic bridging contact with other minorities does not.
In order to standardize the effects of interethnic bridging and get its values in line with the other two constructed contact variables, the value of the variable is the mean of all interethnic (with the exception of Dutch) contact. Respondents who were not a member of leisure or voluntary organisation were coded as having the lowest amount of contact there (never).
Lastly, a variable measuring familial ties was included in the models. A question was posited to the respondents asking them how often they were in contact with any of their (non-distant) family members. The answer categories were identical to the answer categories of the social contact variables listed above.
Control variables
In both waves the respondents were asked to indicate how well they could read and write Dutch, thus giving us a good way to measure (self-reported) Dutch language proficiency. They could give the following answers; I can‟t (read/write Dutch), A little, Decent, Good &
Excellent
In wave 1 the respondents were also asked to indicate how well they could speak and understand Dutch, sadly these questions were not repeated for wave 2. So the choice was made to just include reading and writing proficiency. Next, Dutch reading & writing proficiency were combined into a scale variable, using the mean proficiency of both. PCA, KMO and the Chronbach Alpha indicated that that the scales were fit for use. (results not shown; available upon request)
A deficiency of this variable is the fact that the language skill is self-reported. The NELLS does include a variable in which the interviewer had to give his or her assessment on the language skills of the respondents. However, this variable was just a dichotomous variable asking if the interviewer thought the respondent did or did not speak Dutch well. Using this variable a lot of detail will be lost, on top of this the number of respondents who got their Dutch language skills judged as insufficient by the interviewers was rather small. Including a control for host country language skills is needed due to the likely fact that Dutch language skills correlate with labour market outcomes, a better grasp of the Dutch language can allow a migrant to appear more professional, preform his job more efficiently etc. In addition to this, Dutch language skills likely correlate with certain forms of social capital, such as bridging contact with the native Dutch, as a better grasp of the Dutch language facilitates ease of communication. For these reasons Dutch language skills can act as a confounder and thus needs to be included in the models as a control (Chiswick & Miller, 2002).
We will also control for the following three demographic characteristics ethnicity,
gender and age; gender and ethnicity are dichotomous variables, labelling the respondents as male/female and Turkish/Moroccan respectively. The age variable was constructed by
subtracting the respondents year of birth with the year the interview was taken. These variables are all relatively common control variables included in many researches. Due to the
fact that they correlate with a great many subjects of interests and thus can act as possible confounders. For example, it is a known fact that in general women have worse labour market outcomes than men. Additionally, previous research shows that there are differences in the social network of men and women, due to for example the burdens of childrearing (Munch, Mcpherson, & Smith-Lovin, 1997). Combined the correlation between gender & labour market outcomes and the correlation between gender & social capital enables the variable to act as a possible confounder and thus needs to be controlled for. Similar arguments can be made for the variables age and ethnicity, as they both likely correlate with both labour market outcomes and social capital.
The percentage immigrants living in district variable is constructed by the NELLS research team, which uses municipality data to measure the percentage of non-western immigrants that live in the same district as the respondent (Tolsma, Kraaykamp, de Graaf, Kalmijn, & Monden, 2014). We control for this variable seeing as how the percentage of immigrants living in the respondents district likely has an effect on the amount of possible social contact that a respondent can have with immigrants and natives. A higher percentage of immigrants, most likely, leads to more contact with other migrants and less contact with the native Dutch. In addition, neighbourhoods with a high amount of immigrants living in them are most likely neighbourhoods with a high concentration of people with a low social-economic status. Thus we find the need to control for this possible confounder.
The ISEI score of the respondent‟s first job was recoded from an ISCO-08 score to an ISEI-08 score, using the same transformation scheme mentioned earlier in this methods section. Education is a measure of the highest level of education completed by the respondent, for which they have received a degree. We refer to the descriptive analyses in table 1 for a breakdown of the different educational levels. Both education and previous occupational achievements allow individual job seekers to utilize more high-status contacts, which in turn has a positive effect on future labour market outcomes (Lin, Ensel, & Vaughn, 1981). For these reasons controlling for these two variables is pertinent.
The respondent‟s „marital‟ status is measured by three dummy variables, constructed for this research, which indicate if a respondent either cohabitates with a partner, has a partner but does not live together or is single. Previous research shows that marital status has an effect on the frequency of contact that people have with their network ties (Ajrouch, Blandon, & Antonucci, 2005). The literature is divided upon the question if marital status has an effect on labour market outcomes, with some researchers arguing that is it not marital status but the presence of children that effects labour market outcomes (Dolton & Makepeace, 1987).
Regardless, it seems fruitful to control for the „marital‟ status of the respondents, as it might have an effect on both the primary predictor variables and the dependent variables.
Lastly, we will control for weekly working hours. This variable was coded differently between wave 1 & wave 2 of the NELLS. In wave 1 it was recorded as a scale variable, while in wave 2 the researchers opted for coding the variable categorically. The choice was made to recode the values of wave 1 to the same categories as wave 2. These are as follow; Up to 12
hours, 12 to 16 hours, 17 to 24 hours, 25 to 32 hours, 33 to 40 hours & More than 40 hours.
There is a logical connection between weekly working hours and labour market outcomes, as an increase in the amount of hours worked usually means a higher amount of monetary income. However, and increase in the amount of working hours might mean a reduced ability to socialize and thus possibly a reduced social capital. Because this variable possibly correlates with both social capital and labour market outcomes it is pertinent to include it in the models as a control.
Lastly, to control for any population-wide changes over time a control was added for the wave of the interview, the first wave taking place in 2009 and the second one in 2014.
Income & ISEI Sample N =488 (244 Individuals) Employment Sample N = 1050 (525 Individuals) Mean SD Dependent Variables ISEI 42.067 20.135 Income 1324.426 691.198 ( % employed ) Employed 69.33 Neighbourhood Contact Mean SD Ethnic Bonding 4.770 2.344 5.163 2.275 Ethnic Bridging 3.094 1.884 3.471 1.953 Dutch Bridging 5.884 1.460 5.888 1.482 Work/School Contact Ethnic Bonding 5.805 1.849 - - Ethnic Bridging 4.361 2.150 - - Dutch Bridging 6.704 .876 - - Leisure Contact Ethnic Bonding 2.959 2.415 3.128 2.510 Ethnic Bridging 2.346 1.937 2.539 2.010 Dutch Bridging 3.664 2.516 3.700 2.573 Familial Contact Family Ties 5.246 1.533 5.570 1.529 Control Variables Language Proficiency 4.309 .883 4.417 .808 % Immigrants District 25.059 16.412 25.556 17.093
ISEI First Job 37.805 19.750 - -
Age 36.738 6.501 31.208 9.477 Weekly Working Hours 4.461 1.2682 - - Percentage Percentage Ethnicity Turkish 53.28 52.76 Moroccan 46.72 47.24 Gender Male 56.97 49.52 Female 43.03 50.48 „Marital‟ Status Cohabiting Partner 80.12 55.62 Partner 13.94 6.48 Single 5.94 37.90 Education No Schooling 0.82 2.10 Primary School 6.15 5.33 Lower Vocational SE 7.79 8.19 Vocational SE 10.25 9.52 General SE 8.61 10.48 Scientific SE 2.46 2.67 Lower Vocational TE 11.07 11.62 Vocational TE 23.77 27.05 General TE 19.67 15.62 BSc 2.05 2.10 MSc/PHD 7.38 5.33
Descriptive analyses
Due to only including working people, who are of Turkish or Moroccan descent and who participated in both waves of the NELLS, as well as dropping observations who had missing values, the original sample size of over 5000 is dropped down to N = 488 observations (244 individuals) for the Income and ISEI models. The models with employment status as a dependent variable were a little more lenient in regards to sample restrictions, as jobless respondents are included in this sample. This leaves us with a total N of 1050 (525 individuals) in the employment status models.
A complete breakdown of all descriptive statistics can be found in table 1. However, it seems fruitful to go over some of the important variables to clarify the meaning of the values. In regard to differences between the two samples on the social contact variables, the mean and SD of all variables are rather similar. The largest difference of the mean being just under 0.4, this can be found at the neighbourhood ethnic bonding variable. As such, in order to streamline this methods section, both samples will be discussed together.
Ethnic bonding has a mean of around 5 in the neighbourhood context, around 6 in the work context and around 3 in the leisure context. Which means that bonding occurs, on
average, a few times each month in the neighbourhood, A few times each week on work and a
few times each year on leisure/voluntary associations. The mean value of the leisure context is
much lower due to the fact that all respondents who were not a member of a leisure or voluntary organisation were given a value of 1, meaning never having any form of contact there. This is a pattern we will see repeated for each leisure social contact variable.
Ethnic bridging has a mean of about 3 in the neighbourhood context, about 4 in work
context and about 2 in the leisure context. Meaning that, on average, bridging contact with other ethnic minorities is made about a few times each year in the neighbourhood, about once a month on work and only about once a year on leisure locations.
Dutch bridging has a mean of almost 6 in the neighbourhood, 7 at work and 4 at
leisure locations. Which amounts to an average contact with Dutch people of once or a few times each week in the neighbourhood, almost every day on work and about once a month on leisure locations. The mean for familial ties is around 5, meaning that responds on average talk to their families a few times each month.
We can see several patterns emerge from these descriptive statistics; contact with Dutch natives seems to be the highest scoring variable on all locations. This is not a very surprising result, as the overwhelming majority of the Dutch population is native Dutch; as
such their presence is ubiquitous. Secondly, migrants seem to have most of their social contact at work, followed by the neighbourhood and the least at leisure locations.
When we take a look at the other variables we can see some stark differences between the two samples; namely, in regards to marital status, age and gender. As the employment sample includes nonworking people we can see an increase in the amount of female respondents, from this we can most likely infer that men are more likely to work than women. We also see a decrease in the mean age, hinting to the fact that older immigrants are more likely to be gainfully employed than younger migrants. Lastly, we see a stark difference between the marital statuses of the two samples. Compared to the ISEI and income sample, the employment sample shows a huge increase in the amount of single people and a decrease in the amount of people with a live-in partner. Most likely this means that people with a cohabiting partner are more likely to be employed.
Panel attrition
To check if the panel attrition was random or if certain demographics were more likely to attrite, logistic regressions were carried out with attrition as its dependent variable. These checks gave the following three results, significant on at least p<0.05; firstly, migrants with a high amount of contact with Dutch neighbours were less likely to attrite. Secondly, respondents who have higher Dutch language proficiency are less likely to attrite. And lastly, women are less likely to attrite than men. The results from these logistic regression analyses are not shown in this thesis, but are available upon request. The results of the attrition check are corroborated by the codebook of the NELLS research team, who also find that, for the entire sample, women are less likely to attrite than men (Tolsma, Kraaykamp, de Graaf, Kalmijn, & Monden, 2014). None of the principal dependent variables (ISEI, income & employment status) show signs of selective attrition. As none of the dependent variables suffer from selective attrition we can continue using these models. However, the few variables that do suffer from selective attrition might bias the results, we will return to this in the conclusion. The total panel attrition was quite significant however; of the 5312 respondents of the first wave of the NELLS only 2829 completed the second wave questionnaire (Tolsma, Kraaykamp, de Graaf, Kalmijn, & Monden, 2014). The relative attrition in the sample relevant for this research was even more severe, as the attrition rate among ethnic minorities was higher than the native Dutch. In total there was a possible relevant sample of 2301 individuals who participated in wave 1 of the study. Of these 2301 only 883 returned for wave 2.
Post-estimation checks
The results of all post-estimation checks are not shown in this thesis, but they are available upon request. In order to check if multicollinearity was a problem, several of the most important predictors, regarding social capital, were tested independently. This did not impact the results in a majorly significant manner (e.g. flipping an effect from a positive value to a negative one), thus we can reasonably conclude that multicollinearity was not a major problem. To combat possible heteroscedasticity all models were estimated using robust standard errors (with the exception of the logistic models), which clusters the standard errors around respondent ID. This is especially relevant for our purposes, as the nature of panel data and the oversampling of ethnic minorities by the NELLS research team makes our data prone to violating the assumption of homoscedasticity (Williams, 2015). A histogram of the residuals was generated which shows that, within reason, the residuals follow a bell curve, meaning that they, more or less, follow a normal distribution.
Results ISEI
In table 2 the results from both the random effects (hereafter; RE) and the fixed effect (hereafter; FE) models with ISEI scores as their dependent variable can be found. The standard error in all models is made robust by clustering them around respondent ID. We can see that a few social contact variables have a significant effect on ISEI score in both the RE and the FE models. These are Neighbourhood ethnic bonding, Leisure ethnic bonding and
Leisure ethnic Bridging. Ethnic bonding in the neighbourhood seems to have a positive effect
on ISEI scores; this is true for both the RE and the FE models. From the RE models we can learn that having a higher value on neighbourhood ethnic bonding at one point in time is associated with having a higher ISEI score, this same fact we could have learned from running a regular regression. But since this thesis employs FE models, which only take into account within respondent variation, we can see that an increase in individual neighbourhood bonding contact has a positive effect on the ISEI score. This ability to calculate within respondent variation is a major asset of using panel data and FE analyses, as this greatly improves our ability to talk about casual effects. FE analyses remove all instances of time-constant unobserved heterogeneity between the respondents. In short, neighbourhood bonding social contact seems to have a positive effect on migrant ISEI scores.
However, bonding social contact does not seem to have a universal positive effect. In fact, when taking a look at the leisure social contact the opposite seems to be true. As in both
the FE and the RE models, bonding social contact at voluntary/leisure associations seems to have negative effect on ISEI scores. Whereas bridging contact, at leisure locations, with other ethnic minorities seems to have a positive effect on ISEI scores. This seems more in line with Putnam‟s theory on social contact which states that: „‟Bonding social capital is good for
getting by, but bridging social contact is essential to get ahead” (Putnam, 2000). However,
these results have one oddity, this bridging social contact is with other ethnic minorities and not the native Dutch population. Usually contact with the native population is seen as the most beneficial, as these peoples possess potential resources (language skills, network contact etc.) that other migrants do not. We will return to this observation in the conclusion.
ISEI N = 488 (244 Unique) RE Model 1 B (SE) RE Model 2 B (SE) FE Model 1 B (SE) FE Model 2 B (SE) Neighbourhood Contact Ethnic Bonding .507 (.421) 1.101 (.421) ** 1.128 (.450) ** 1.256 (.453) ** Ethnic Bridging -.761 (.511) -.990 (.509) * -.911 (.563) -.727 (.554) Dutch Bridging .686 (.544) .550 (.544) .732 (.641) .607 (.597) Work/School Contact Ethnic Bonding -1.376 (.486) ** -.657 (.463) -.347 (.549) -.388 (.541) Ethnic Bridging .030 (.439) .005 (.419) -.372 (.499) -.395 (.469) Dutch Bridging .522 (1.223) -.023 (1.139) -.483 (1.396) -.346 (1.382) Leisure Contact Ethnic Bonding -1.537 (.481) ** -1.660 (.460) *** -1.283 (.563) ** -1.366 (.548) ** Ethnic Bridging 2.030 (.628) ** 1.907 (.606) ** 2.010 (.763) ** 1.995 (.721) ** Dutch Bridging .200 (.334) .095 (.330) -.183 (.379) -.063 (.402) Familial Contact Family Ties .371 (.545) .633 (.522) .386 (.651) .540 (.611) Control Variables Language Proficiency - 3.535 (1.052) ** - -3.534 (1.654) ** % Immigrants District - -.061 (.056) - -.174 (.167)
Weekly Working Hours - 1.163 (.080) - -1.100 (1.270)
Partner (ref. single) - -.549 (4.970) - 2.019 (6.515)
Cohabiting Partner (ref. single)
- 3.728 (2.477) - 7.391 (5.324)
Turkish (ref. Moroccan) - -1.518 (1.871) - -
Age - -.0124 (.153) - -
Female - 3.618 (2.010) * - -
Education - 1.574 (.423) *** - -
ISEI First Job - .355 (.060) *** - -
Wave - 1.513 (1.272) - .559 (1.120)
Constant 39.456 -9.600 39.827 55.452
R^2 .067 .479 .074 .113
*** P >.001 **P >.05 *P >.1 (two-tailed tests, robust standard errors).
Table 2 – Random and fixed effect models predicting ISEI scores among Turkish and Moroccan immigrants in the Netherlands
One more observation regarding the relationship between social contact and ISEI needs to be addressed. In the RE models ethnic bridging contact in the Neighbourhood has a negative effect on ISEI scores. From this we can infer that, at one point in time, migrants who have more contact with other ethnic minorities have a lower ISEI score. However, we cannot find this same effect in the FE models, meaning that an individual change in ethnic bridging social contact is not associated with a change in ISEI score. It seems likely that there is some unobserved heterogeneity between individuals who have a high amount of interethnic neighbourhood contact, leading to a spurious result.
When we take a look at the control variables we can find two more noteworthy results. Being female seems to have a positive effect on ISEI scores, which seems to contradict conventional and scientific wisdom. However, this could be explained by our sample selection. As only working people are included in this sample, it might be that within migrant communities women have a far higher unemployment rate than men (this is corroborated by the results of the employment analyses, see table 4). As a result, the women that do have a job might be the most independent and ambitious individuals, thus biasing our sample of women towards those with a higher ISEI score. Lastly, the results of Dutch language proficiency seem perplexing and rather contradictory. The RE models show that, at one point in time, higher Dutch language proficiency is associated with a higher ISEI score, which is the expected result. However, in the FE model we can see that an individual positive change in Dutch language proficiency is associated with a negative change in ISEI score. Again this result seems to contradict conventional and scientific wisdom.
All in all, the consistent results across all models regarding neighbourhood ethnic bonding and leisure ethnic bonding and bridging seem to indicate that social contact at these locations has some sort of impact on the ISEI scores of migrants. The consistency across the models, disregarding language proficiency, also points to a correct operationalization of the social contact variables, meaning that they most likely measure what they intend to measure.
Income
Whereas the ISEI models give a rather clear and consistent picture across all the different models, the income models, sadly, do not. As we can see in table 3 there are no social contact variables which have a consistent effect across all four models. In the RE models we can see that both ethnic bonding contact on the work floor and familial ties have, at a single point in time, a negative effect on income. However, we cannot find these same effects when looking at the within respondent variation in the FE models. This leads us to conclude that the results
found in the RE models, regarding ethnic bonding on the work floor and familial ties, are caused by unobserved heterogeneity between the respondents. Regarding the negative effect of work floor ethnic bonding, it seems likely that this is caused by the fact that jobs which have a high amount of ethnic minority homogeneity are also the jobs on the lower end of the salary spectrum (Ryan, Sales, Tilki, & Siara, 2008). In addition, it seems like this low salary is not caused by the ethnic bonding social contact, but due to some exogenous cause. If the bonding social contact was (partially) the cause for this lower salary found at jobs with high amount of ethnic homogeneity, we should have found a similar effect in the models looking at within respondent variation. As this is not the case, some factor not included in our models seems to be the cause for this lower salary (e.g. discrimination or type of industry).
Income N = 488 (244 Unique) RE Model 1 B (SE) RE Model 2 B (SE) FE Model 1 B (SE) FE Model 2 B (SE) Neighbourhood Contact Ethnic Bonding -3.795 (15.088) 10.222 (16.359) 31.241 (16.827) * 34.575 (18.995) * Ethnic Bridging -22.848 (19.377) -26.780 (18.140) -22.551 (21.797) -23.236 (21.460) Dutch Bridging 22.490 (23.249) 24.804 (21.811) 1.174 (32.934) 1.667 (32.663) Work/School Contact Ethnic Bonding -56.154 (19.693) ** -44.715 (18.187) ** -1.208 (21.985) -2.399 (23.685) Ethnic Bridging -6.644 (12.707) -9.385 (12.617) -15.509 (14.748) -16.938 (15.091) Dutch Bridging 61.498 (34.730) ** 37.915 (34.270) 7.124 (36.845) 7.200 (38.340) Leisure Contact Ethnic Bonding -17.436 (18.846) -23.405 (19.039) 3.515 (19.607) 5.049 (19.640) Ethnic Bridging 22.539 (26.981) 28.837 (27.978) 2.107 (27.625) -.701 (27.854) Dutch Bridging 3.922 (13.880) -1.647 (13.716) 4.876 (16.583) 3.688 (17.110) Familial Contact Family Ties -45.445 (22.530) ** -47.523 (21.268) ** -29.394 (32.121) -34.760 (32.357) Control Variables Language Proficiency - 104.599 (34.310) ** - 51.803 (50.479) % Immigrants District - -.511 (1.829) - -4.753 (7.943)
Weekly Working Hours - 151.750 (28.107) *** - 90.379 (45.118) *
Partner (ref. single) - -52.292 (129.797) ** - -126.648 (165.611)
Cohabiting Partner (ref. single)
- 128.294 (96.996) - 68.402 (206.382)
Turkish (ref. Moroccan) - 169.750 (28.107) - -
Age -- 3.753 (6.101) - -
Female - 233.768 (85.777) ** - -
Education - 35.524 (14.798) ** - -
ISEI First Job - 5.140 (2.193) ** - -
Wave - 3.180 (49.541) - 24.437 (45.660)
Constant 1447.176 -432.382 1386.174 825.963
R^2 .142 .363 .027 .057
*** P >.001 **P >.05 *P >.1 (two-tailed tests, robust standard errors).
Table 3 – Random and fixed effect models predicting income among Turkish and Moroccan immigrants in the Netherlands
The negative effect of familial ties on income could be explained by a number of different cases of unobserved heterogeneity. For example, it could be that this effect is caused by migrants who work in a family company, which in the Dutch migrant context, are often companies in the lower-end service industry. We can‟t be sure about the true nature of this unobserved heterogeneity but, due to not finding a repeat effect in the FE analyses, we can say that this effect is most likely not caused by the familial ties themselves, thus making the results spurious.
In turn in the FE analyses we find a significant effect (only on p<0.1), which is not found in the RE models. An individual increase in neighbourhood ethnic bonding has a positive effect on income. This results matches well with the positive effect of neighbourhood ethnic bonding on ISEI score mentioned earlier. However, it is rather curious that we did not find this same effect repeated in the RE models. The opposite case of unobserved heterogeneity causing spurious effects could be at play here; unobserved heterogeneity which supresses an effect. It could be that, while neighbourhood bonding contact has a positive effect on income, due to some other exogenous factor negatively impacting income, shared by migrant with high amounts of neighbourhood bonding contact, this positive effect of bonding contact gets supressed.
Some observations from the control variables; once again we find a positive relationship between being female and income, repeating the observation from the ISEI models. This seems most likely caused by the fact that, due to only including working people, our sample of women is biased towards the more ambitious and successful. Language proficiency loses its significant effect in the FE models, as opposed to the RE models. Once again this could be caused by unobserved heterogeneity.
In conclusion, the rather clear picture that arises from the ISEI models is sadly not repeated in the income models. Where we could say, with some certainty, that certain kinds of social capital had a positive or negative effect on ISEI scores, we cannot do this for income. Next we will turn our eye towards the models predicting the chance of being employment.
Employment Status
When measuring employment status we could not use FE analyses, due to a very small N (~120), as within respondent variation is needed for this method. To keep the N at acceptable levels the choice was made to just model employment status using RE analyses. Additionally some variables had to be dropped from the models, these being; work social contact & ISEI
