1
Group Consciousness and Political Participation in
Amsterdam: Do Ethnic Group Attachment and Perceived
Discrimination have a Mobilising Effect on Inhabitants
with a Migration Background?
Master Thesis M.Sc. Research Master Social Sciences Hannah Schwarz
UvA ID: 10436839
Thesis Supervisor: Dr. Floris Vermeulen Second Reader: Dr. Tom van der Meer Submitted: August 15, 2014, Amsterdam
2 Abstract
In this paper, I investigate factors impacting the political participation of ethnic minority group members in Amsterdam. Lending from the group consciousness literature, I will specifically look at the effect of ethnic group attachment. I hypothesise that, in the presence of perceived discrimination and a sense of political efficacy, ethnic group attachment increases the voting intention among members of Amsterdam’s largest three minority groups: Moroccans, Turks, and Surinamese. This hypothesis will be tested in a multilevel logistic regression model using data from several waves of the Amsterdam citizen monitor survey. My findings do not confirm the hypothesised interaction effect but perceived discrimination and efficacy show to have significant main effects on voting intention. Migrants in Amsterdam are more likely to intend to vote when they have a sense of political efficacy and, unexpectedly, they are less likely to vote when they feel discriminated. Furthermore, I find a positive effect of ethnic attachment in my sample which approaches significance.
Keywords: Group Consciousness, Political Participation, Voting Intention, Ethnic Attachment, Perceived Discrimination, Migrants in Amsterdam
Introduction
In Amsterdam, the political participation of migrants1 has received substantial amounts of attention from researchers as well as in public debate in recent years. Differences in turnout between groups as well as fluctuations over time have inspired discussion about potential causes of migrants’ political (non-)participation time and again among researchers, policy makers and journalists (see e.g. Tillie, Fennema, and van Heelsum 2000; Van der Heijden and Van Heelsum 2010; Het Parool 2014; Kranendonk et al. 2014). While research investigating the political participation of migrants in Europe has frequently relied on a ‘civicness approach’ according to which the degree of civicness of an ethnic group is an important explanatory factor for an individual member’s political participation (Bloemraad and Vermeulen 2014), a prevalent strand of literature about the political participation of African-Americans has looked for explanations pertaining to ‘group consciousness’. The concept of group consciousness has mostly been used as consisting of minority group members’ sense of belonging to their ethnic group which can have a mobilising potential when individuals are
1
The terms migrant and immigrant will be used interchangeably both denoting people who have migrated to the Netherlands or who have at least one parent who has migrated to the Netherlands. When referring to the situation in the Netherlands, the term ethnic minority group member also refers to this same group of people.
3 also aware of the disadvantaged status of their group and, at the same time, feel that political action could help advance this status (McClain et al. 2007). My aim in this paper is to translate this theoretical reasoning to the political participation of ethnic minority groups in current day Europe, particularly in the city of Amsterdam. My research question will be: Is there a group consciousness effect on political participation among Amsterdam’s largest three ethnic minority groups?
The fact that my research question and the focus of the analysis are inspired by debates in the local policy-research nexus already indicates the societal relevance of this paper in the Amsterdam context. Here, as well as in other urban areas, migrants’ formal political participation is growing in importance as the number of people who have a migration background (and a right to vote) is higher than in rural areas and continues to rise (Bureau Onderzoek en Statistiek 2013). In Dutch local elections, many more migrants are entitled to vote than in the national elections as formal citizenship is not a requirement. Rather, after five years of legal residence, or six month for EU-citizens, migrants are granted the right to vote. Local elections in urban centres, such as Amsterdam, thus provide a good opportunity to study migrants’ participation in conventional politics. Furthermore, the general argument has been made that studying migrant integration on the local level is of special relevance as integration processes are essentially taking place locally (Borkert and Caponio 2010).
The topic of voting behaviour of migrants and minority groups has been deemed worthy of investigation by many in Amsterdam and the Netherlands, and beyond. Voting is the most important legitimation of democratic governments and their policies. Unequal turnouts between ethnic groups have thus frequently been described as worrying (Schönwälder and Bloemraad 2013). Van der Heijden and Van Heelsum (2010) have studied the relatively lower turnout of different groups of migrants in the Netherlands, also specifically at local elections. In Amsterdam, exit polls conducted since 1994 have consistently shown a lower turnout
4 among residents with a migration background than among people of Dutch descent as well as among certain migrant groups as compared to others. Van der Heijden and Van Heelsum (2010, 8) comment that this is problematic in terms of legitimacy as a systematic underrepresentation of migrants’ interests and a lack of consciousness of their preferences in the political sphere are likely under such conditions. This normative concern of studies investigating “quantity, quality and equality of participation” is also stressed by Campbell (2013, 34). He, furthermore, attributes a tradition of advocating public policies to affect participation to this literature. Clearly, a thorough understanding of who participates is necessary to advise such policy efforts.
A typical feature of the scholarly field of political participation is the fact that it is dominated by empirical studies (Geys 2006). It is then also an important contribution of this study to add further empirical insights to this already existing pool of findings enabling the accumulation of knowledge, for example in the form of meta-analyses (see e.g. Geys 2006). Yet, the academic relevance of this paper is not only high in terms of its empirical results. The development of theory on the electoral participation of ethnic minority group members in Europe has stagnated recently. During the last two decades, work on the topic has mostly stressed social capital and civic network approaches to understanding the differential political participation of different groups (see e.g. Fennema and Tillie 1999). Notably, the finding that a group’s civicness can partly explain differences in political participation between groups has become rather well established. While this approach proved useful in the city of Amsterdam, studies have also shown it to be of less value elsewhere (Bloemraad and Vermeulen 2014). However, other strands of literature suggest a promising addition to this theoretical approach focusing on factors that have been neglected by researchers in the field so far. The literature which evolved against the context of African-American political participation in the 1960s emphasises the aspects of subjective identification with the ethnic group and individual
5 perceptions of group disadvantage. My contribution in this paper will be to take a new look at the electoral participation of migrants in Europe from a perspective inspired by the group consciousness literature on the political participation of the African-American minority in the US in the 1960s and see to what extent this can give us new insights adding to the ones provided by the civicness literature.
I will use multilevel logistic regression analyses to test whether mechanisms found among African-Americans also exist among the largest three migrant groups in Amsterdam, namely people with Moroccan, Turkish, or Surinamese background. For this, I will use data from the cross-sectional survey ‘Burgermonitor’ (citizen monitor) conducted with a representative sample of Amsterdam inhabitants. This survey is conducted and results are reported on a yearly basis by the Amsterdam Office for Research and Statistics (Bureau Onderzoek en Statistiek) but the survey data have rarely been used for other research. The citizen monitor contains indicators for political participation as well as an item on feeling attached to one’s ethnic group and one on feeling discriminated on an ethnic basis. Identical questions have been posed to different respondents across several years and, as the sample has been drawn randomly, it is representative for the whole population of Amsterdam inhabitants. As becomes clear, the citizen monitor data is quite rich and has many favourable qualities. It is time to make use of its potential in advanced statistical analyses to investigate questions concerning citizen participation in Amsterdam such as the electoral participation of migrants.
Theoretical Framework
The Socioeconomic Core Model
The research field of voter turnout has been said to be dominated by empirical studies (Geys 2006). Instead of focusing on developing theory further, recent research has often relied on classical and widely accepted works. One of these is the monograph ‘Participation in
6 America’ by Verba and Nie (1972). According to their approach, inclinations to vote importantly result from a feeling of civic duty. Whether this is present, in turn, crucially depends on an individual’s socialisation and the socioeconomic resources present in this socialisation. In particular, the authors stress the role of individual resources and motivations. Education, wealth and high status occupation hereby provide resources that can be converted into political activity. As the key factor, education provides motivations or ‘civic orientations’ that will make an individual become active, such as efficacy, political interest, information, and a sense of obligation. Many scholars have described similar mechanisms positing that education enhances norms or attitudes connected to a sense of civic duty (Wolfinger and Rosenstone 1980; Verba et al. 1993). Furthermore, the effect of education has often been claimed to work through increasing political knowledge (see Junn 2000) or providing relevant skills to process political information and make political decisions (Wolfinger and Rosenstone 1980; Verba et al. 1993). Verba et al. (1993) here name the examples of organisational and communicational skills. The presence of these skills can, moreover, be mediated by income as high income jobs can be expected to provide more opportunity to practice the skills. Income can thus be seen as a mediator variable between education and participation. Especially for the specific activity of voting where, as opposed to other political activities, monetary resources do not play a role, I expect no separate effect of income (Brady et al. 1995)2. Another factor that has been seen to increase relevant skills is age (Wolfinger and Rosenstone 1980). However, age has also been argued to increase electoral participation through another mechanism. Namely, older people have been argued to be more likely to vote because they tend to be less mobile and more integrated into the communities where they live (Leighley 1995). The role of different mechanisms in creating the effect of education on participation is not yet clear and has remained understudied (Junn 2000). Such investigations might have
2
A preliminary statistical analysis in which sex, age, education and income level dummies were included also showed no significant effect of income.
7 been neglected as the abundance of empirical evidence for the existence of the relationship seems to have settled the question. A further predictor for electoral participation that has traditionally been found to matter is sex, with women having been less likely to vote than men. However, differences have been decreasing and it is questionable whether they are currently still present (Leighley 1995).
Already Verba and Nie (1972) started their research on the relationship between socioeconomic status and participation from a firm empirical basis as this relationship had been established by numerous studies before. They replicated the former findings using National Population Survey data from the US in 1967 showing that people with upper social status are overrepresented amongst those who participate in various political activities, one of which was voting. A wide array of empirical studies has found similar results with education consistently proving to be the most consistent and powerful predictor3 (Campbell 2013; Gallego 2010; Wolfinger and Rosenstone 1980; Leighley 1995; Caínzos and Voces 2010). A model based on indicators for socioeconomic status has become widely accepted as a ‘socioeconomic core model of political participation’ with which a large share of the variance can usually be explained (Campbell 2013; Leighley 1995).
H1: Having a higher level of education increases a person’s likelihood to intend to vote in the next elections.
Additions to the Model in the Case of Migrants
Research on the political participation of minority group members shows that the common individual level predictors work similarly well here as among the majority group (Junn 2000; Bird, Saalfeld, and Wüst 2011). However, studies have also found that even when controlling for socioeconomic variables, differences between groups tend to persist. Summarising
3
Whereas claims about the effect of income have ranged from declaring it almost as persuasive as that of education (Leighley 1995) to declaring it largely absent (Wolfinger and Rosenstone 1980).
8 findings from studies in 11 liberal democracies in the Western World (among which the Netherlands), Bird, Saalfeld and Wüst (2011) conclude that the gap in voting turnout between citizens with migration background and the native majority groups is narrowed when socioeconomic status is accounted for but is not closed. Hence, next to socioeconomic status, other variables specific to minority groups seem to make a difference. Next to citizenship and length of residence, Bird, Saalfeld and Wüst (2011) discuss the social capital of ethnic groups to be an important factor. Fennema and Tillie (1999) as well as Van Heelsum (2005), similarly, work with the framework of group social capital to explain differences in the political participation of migrants, between ethnic groups, across time and between the local and the national level.
H2: Amsterdammers with a migration background are less likely to intend to vote than members of the Dutch ethnic majority group.
Very different approaches to explaining such group differences in political participation appear when one looks at scholarly work on minorities in a different context, namely that of African-Americans in the 1960s. Verba and Nie (1972) observed differences in their political participation as compared to members of the ‘white’ majority group, with ethnic minority group members participating more than would be expected on the basis of their socioeconomic status. Such observations have led Verba and Nie, as well as other scholars, to theorising about group-based mobilisation processes and, particularly, to formulating the hypothesis that a sense of group consciousness can increase political participation. More concretely, the consciousness of blacks to belong to an ethnic community with generally low status has been seen to motivate its members to strive for social changes that could benefit the black community, also by means of participation in conventional political activities such as voting (Guterbock and London 1983, 440).
9 This strand of research offers an interesting basis to come to terms with the political participation of minority groups in current day Europe, in this particular case in the city of Amsterdam. The most important parallel that makes this transfer of theory seem useful is that here, many ethnic minority groups also have lower average social status and used to do so historically. Furthermore, hostility towards these groups from members of the majority society (also expressed in the political sphere) is a constant condition against which the (political) integration of ethnic minority group members has to be understood. Chong and Rogers (2005, 28) have similarly argued that it makes sense to apply the group consciousness concept to the study of the political participation of ‘new minorities’ in the United States (notably Latinos and Asian Americans) as they have a “racial minority status, non-European backgrounds, vulnerability to discrimination, and in some cases, socioeconomic disadvantage”. In the following, I will give a brief introduction into the literature that developed the group consciousness hypothesis.
While Verba and Nie’s (1972) approach has been termed social psychological, they do not discuss the concept of group consciousness and the mechanisms through which it could affect the political participation of ethnic minority group members as explicitly as later literature that more formally refers to social psychological theories. This literature is often ultimately based on the notions of collective identity and the role of people’s group membership in social behaviour as presented in the works of Tajfel and Turner (1979;1986).
Miller et al. (1981) use both Verba and Nie’s (1972) work in the political science literature and the social psychological literature based on Tajfel and Turner when theorising differences in the political participation between demographic subgroups. From different descriptions of the group consciousness concept in the literature, they derive, on the one hand, an identificational component (group identification) and an affective component resulting from that (polar affect, a sort of in-group favouritism). On the other hand, they derive a component
10 consisting of the awareness of materialistic disadvantage (polar power) for which blame is attributed externally (system blame). This allows them to go beyond Verba and Nie’s (1972 as cited by Miller et al. 1981) definition of group identification which assumes that identification automatically entails a perception of deprivation. Miller et al. (1981) stress the distinction between group identification and group consciousness with the latter also involving a political awareness or ideology regarding the group’s relative position in society, along with commitment to collective action aimed at realising the group’s interests. They thus emphasise that “there is no theoretical reason to expect a simple direct relationship between group identification and political participation” (Miller et al 1981, 495). Rather, this relationship is expected to be moderated by a political awareness of the disadvantage of the group in society. Empirically, Miller et al. (1981) separately investigate the relationships of the different components of group consciousness they identify with voting. The results show that the components separately are only weakly and inconsistently related to voting. Also combining them in an additive model does not result in a significant strengthening of the correlations. However, a model in which the different group consciousness components interact turns out to fit the data much better. The strongest interaction model they find combines group identification with an awareness of materialistic group disadvantage and system blame for this disadvantage.
In the more current and European literature, we find Simon and Klandermans’ (2001) attempt to further develop theorising on politicised collective identities. Coming from a social psychological perspective as well, the authors deal with similar concepts as Miller et al. (1981) when working out three conceptual triads that make up a politicized collective identity. Next to collective identity, other main elements in their approach are awareness of a shared grievance of the group as well as an element of power struggle, and adversarial attribution of the causes of the grievances. Hence, also in this model, a collective identity is only assumed
11 to have an impact on political participation if it plays a political role. Schildkraut (2005, 291) summarises these different specifications of group consciousness effects at a more abstract level stating that they all investigate the effects of group membership and of perceptions of group-level and individual-level treatment on political activity.
Shingles (1981) stresses another important condition for the group consciousness effect. Criticising the sparseness of theorising of concrete mechanisms in Verba and Nie’s (1972) work, he suggests trust and efficacy as important mediators. Based on work by Gamson (1968 as cited by Shingles 1981) and Gurin (1969; 1975 both as cited by Shingles 1981), he theorises that the relationship between African-American group consciousness and political participation is mediated by feelings of political mistrust and a sense of internal political efficacy. However, the roles played by efficacy and trust in this relationship differ according to which type of participation is investigated. For the activity of voting, it is harder to predict their exact roles as, more so than with other activities, difficulty and meaning of the act of voting are argued to vary across individuals. Yet, Shingles states that the consensus among scholars who work on the relationship between efficacy, trust and political activity is that efficacy encourages activity but that depending on the level of trust, it encourages activity of different types. In his core theoretical assumptions, the author posits that efficacy is always beneficial for high initiative forms of participation but of less importance for low initiative forms of participation, such as voting. Yet, he also clearly states that individuals without any sense of political efficacy can be expected to withdraw from conventional politics altogether. While the relationship between political trust and voting thus remains somewhat ambiguous theoretically, there seems to be agreement that efficacy should have a positive effect on voting, even if its role is not as important here as in higher investment forms of political activity. Empirically, this is backed up by findings from Guterbock and London (1983)
12 showing that African-Americans who score high on both trust and efficacy vote significantly more often than expected on the basis of their socioeconomic status.
H3: Migrants are more likely to intend to vote if they feel attached to their ethnic group. This relationship only holds if, on the one hand, they perceive their group to be disadvantaged and, on the other hand, they have a sense of political efficacy.
Contextual Level Factors
Next to individual level predictors, also contextual factors have commonly been theorised to have an impact on whether people decide to vote or not (see e.g. Geys 2006). While such factors are often left out in research focusing on individual level analyses, my data offer the great opportunity to account for contextual influences while investigating individual level relationships. Contextual influences that have been discussed in the literature include factors that are likely to vary across time and such that are more likely to vary across spatial units within the city I am studying. Examples for the former are state-level or city level systemic variables, such as per capita income, the current degree of party competition and, in election years, the intensity of campaign activities (Gaardsted Frandsen 2002; Leighley and Nagler 1992). Examples for the latter are socioeconomic neighbourhood characteristics as well as population heterogeneity and population turnover of a neighbourhood (Geys 2006; Gaardsted Frandsen 2002; Saalfeld 2011). Additionally, the factors varying across neighbourhoods are also likely to vary across time within the neighbourhoods.
Beyond this general importance of accounting for contextual influences on political participation, it is specifically crucial to do this in my particular analysis as the individual level relationship I am looking at has been hypothesised to be directly affected by certain contextual factors. Notably, the salience of a collective identity has been seen to affect the extent to which people’s political behaviour is affected by this identity (Tajfel and Turner
13 1979; 1986). Political contestation of social categories that collective identities are based on has been seen as a factor increasing this salience (Verba, Nie, and Kim 1978). In times where the integration of migrants is a hot topic in public debate it seems plausible to assume fluctuations in political contestation and salience of ethnic minority group identities over time and potentially also across spatial (neighbourhood) units.
Data, Operationalization and Measurement
Data
I draw my data from the ‘Burgermonitor’ (citizen monitor) survey conducted since 1999 on a yearly basis by the Office for Research and Statistics of the city of Amsterdam. This survey is concerned with Amsterdammers’ relation to their city focusing on topics such as contact between citizens and the city administration, political interest and activity, media use, and social networks (Michon et al. 2014). The number of respondents ranges between 2500 to 3000 every year and the survey typically includes around 100 items. The number of respondents with a migration background is around 700 per year and the number of respondents associated with one of the largest three minority groups around 400. Combining the data from the yearly survey waves from 2004 to 2009, and dropping cases with missing values, I end up with an overall sample of 14,592 Amsterdammers and a subsample of 2263 Amsterdammers with Moroccan, Turkish, or Surinamese background.
Operationalization and Measurement
Dependent Variable
The dependent variable discussed in the literature review is voting. I will approach this with a variable on voting intention. Voting intention is not a good indicator for actual voting behaviour, as the gap between intention and turnout has often been found to be substantial (Rogers and Aida 2014). However, for my research question concerning factors mobilising
14 migrants, a measure of actual behaviour is not necessary but measuring effects on people’s voting intentions, capturing their political motivation, is suitable. The dependent variable is a dummy variable that was created based on an ordinally scaled variable. The question underlying the item used was ‘If there would be city- and district council elections tomorrow, how big is the chance that you would go vote?’. Three answer options were possible: ‘I will vote for sure; I will maybe vote; I am sure I will not vote’. In the dummy variable, only the answer option ‘I will vote for sure’ is coded as one (intending to vote), while the other two options are coded as zero (not intending to vote). Considering the fact that it is generally socially desirable to vote, this seems sensible.
Independent Variables
McClain et al. (2007, 476) discuss how group consciousness goes beyond group identity in that it is identification which is politicised by beliefs about the social standing of one’s group and the attitude that this social standing can best be improved by collective action. My operationalisation approaches this definition of group consciousness by translating these three aspects into three separate variables of which I then compute an interaction variable. A slight divergence from this definition lies in the fact that my interest is in political action as a means to improve a group’s status rather than necessarily collective action. In all three cases, dummy variables have been created from the ordinally scaled original variables.
Firstly, a group identity or attachment variable is constructed on the basis of an item asking respondents: ‘We would like to know to what extent you feel attached to the following entities [amongst others: ‘your own ethnic group in Amsterdam’]. You can choose from the following answers: not at all attached, not attached, neutral, attached, very attached.’ The answer options ‘very attached’ and ‘attached’ were coded as one (attached) whereas the other answer options were coded as zero (not attached).
15 To account for people’s belief about whether their group is disadvantaged, the variable perceived discrimination is employed. People who feel that they are being discriminated on the basis of their ethnic background are aware of the disadvantaged status of their ethnic group. For the item underlying this variable, respondents were asked: ‘Do you sometimes feel discriminated on the basis of your ethnic background?’. There were four answer options: ‘often, sometimes, rarely, never’. The dummy variable was created in such a way that ‘often’, ‘sometimes’, and ‘rarely’ were coded as one (perceived discrimination) and ‘never’ was coded as zero (no perceived discrimination).
The aspect whether respondents think that political action is a good means for reaching group goals is captured by a measure of sense of political efficacy. This is based on an item asking for respondents’ opinion on the statement: ‘People like me do have a considerable influence on government politics’ with the answer options: ‘agree; neither agree nor disagree; disagree’. Here, the first two options were coded as one (efficacy) and the others as zero (no efficacy). I decided to also count the ‘undecided’ respondents to the people having a sense of efficacy. This is because I think that the wording of the question sets the threshold to responding with an unconfined ‘agree’ rather high as it seems to imply a quite direct impact of an individual on government politics. For the year 2009, the same item was not included in the survey. Instead, I had to use a different but conceptually similar item asking for agreement with the statement ‘the city councilors and parliamentarians in Amsterdam don’t care about the opinion of people like me’ with the answer options: ‘completely agree; agree; neither agree nor disagree; disagree; completely disagree’. In this case, the last two answer options were coded as one (efficacy) and the former ones as zero (no efficacy).
Moreover, sociodemographic control variables will be used. My education variable is based on an item asking “Which is the highest level of education you completed?”. The answer options are according to the Dutch tracked school system including several options for
16 vocational training at different levels. Those were recoded into the more universally meaningful categories: No formal education, primary education, pre-vocational education, vocational education, advanced vocational education and academic education, and a dummy variable was created for each educational level. Sex was coded such that zero denotes male and one denotes female. Age was indicated by the respondents in years.
The ethnic background of a person is determined according to the rules used for Dutch official statistics. Here, a person who is born abroad is attributed the ethnicity of his or her country of birth. A person is also classified as having a migration background if at least one parent is born abroad. That person’s ethnic background is then automatically the one of the foreign born parent(s). If both parents were born abroad but in different countries, the country of origin of the mother determines a person’s ethnic background (Centraal Bureau voor de Statistiek 2014).
Grouping Variables: Neighbourhoods and Neighbourhood-years
In the literature review it was pointed out that contextual level factors varying across time as well as geographical units (amongst others neighbourhoods) have been seen to impact my dependent variable political participation as well as the particular individual level relationship I hypothesise. It will therefore be valuable to account for the clustered nature of my data. The observations are clustered in years, as I use six survey waves, and in neighbourhoods, as the random sample of Amsterdam inhabitants comprises observations in different geographic and administrational units of the city. The task of defining meaningful neighbourhood units is not straightforward but based on theoretical considerations and variances of the dependent variable at different units, a reasonable choice can be made. The citizen monitor survey offers information that would allow me to construct differently sized spatial clusters. The smallest would be defined by six digit postcodes. The units thus delimited often include just a few streets and are thus not suited from a theoretical perspective if one wants to investigate
17 neighbourhood effects. On the other extreme, Amsterdam has also been divided into 15 large city parts (stadsdelen) resulting in units comprising on average roughly 15 square kilometres or 50,000 inhabitants (Dienst Onderzoek en Statistiek 2005). It cannot be assumed that there is much of a neighbourhood connection between people living several kilometres away from each other. A good solution in the middle are the units called neighbourhood combinations (buurtcombinaties). The city administration divides Amsterdam into around 100 of those units with numbers having slightly changed due to administrational changes over the years. For reasons of data availability, I will work with the neighbourhood combination units defined in 1998 of which there are 94 in the population, 92 in my overall sample and 82 in my migrant subsample. Results from running empty random intercept models with different neighbourhood units4 shows that using those as neighbourhoods increases the variance as compared to using city parts (rho increases from 0.000 to 0.002) but, on the other hand, they are still large enough to assume neighbourhood dynamics to unfold at crucial sites such as schools or neighbourhood organizations (Jencks and Meyer 1990).
Table 1. Variance (rho) of dependent variable at different neighbourhood units
unit (number of clusters) variance (rho) in voting intention at the level of this unit neighbourhood (264) 0.004
neighbourhood-combination (83) 0.002
city part (15) 0.000
As the variance of the dependent variable across years is likely to differ between neighbourhoods, I include a further level of clustering into my analysis, namely the neighbourhood-year level. Each neighbourhood-year combination is hereby attributed a unique value according to which observations can then be grouped.
4
These analyses were run on the subsample of people with a background associated with either of the largest three minority groups as I am interested specifically in their political participation.
18 Methods of Analysis
Before the main analysis, I will model the effect that having a migration background has on voting intention in the case of Amsterdam. To analyse the effects of multiple dummy and continuous independent variables on my dichotomous outcome variable ‘voting intention’, I will use a logistic regression model. This model assumes a binomial distribution for the dependent variable, which is natural for binary outcomes. I will calculate the coefficients in odds ratios to enable a more easily comprehensible interpretation. Odds ratios indicate the percentage of increase or decrease of the estimated odds of falling into one of the two categories of the outcome variable (Agresti and Finlay 2009). I will conduct this first set of analyses on the whole sample, including people with and without a migration background. The main analyses will be conducted on a subsample of people who have a migration background associated with either of the largest three minority groups in the Netherlands (Moroccans, Turks, and Surinamese).
To account for contextual level influences while investigating the individual level relationships, I will construct a hierarchical or multilevel model. In the presence of multilevel data this allows for controlling for the variance in the dependent variable due to clustering. In this case, individuals are modelled as nested within neighbourhood-year combinations which are then again nested within neighbourhoods. As I use data including observations from several years, I also introduce year-dummy variables into the model to control for the variance due to year specific contextual influences. More concretely, that part of the variance in voting intention that can be attributed to variation of voting intention across neighbourhoods and across years within neighbourhoods (neighbourhood specific trends) will be controlled for and thus not disturb the investigation of the investigated individual level relationships. Similarly, introducing the year dummies allows controlling for the variance in voting intention across years in all of Amsterdam (neighbourhood unspecific trends). In the terminology of multilevel
19 models, working with ‘fixed effects’ and ‘random effects’, I will thus use fixed effects for my variables of interest to investigate the individual level relationships while controlling for variance of the dependent variable at higher levels (of clustering) by introducing year fixed effects and random intercepts for neighbourhoods and neighbourhood-years (DiPrete and Forristal 1994).
Analyses and Results
Descriptive Statistics
Table 2. Descriptive statistics of dependent and independent variables for the analyses on the
whole sample N Mean SD Dependent variable voting intention 14592 0.772 0.419 Independent variables sex 14592 0.545 0.498 age 14592 46.365 16.623 Dutch 14592 0.665 0.472 Moroccan 14592 0.062 0.242 Turkish 14592 0.059 0.235 Surinamese 14592 0.07 0.256 other Western 14592 0.104 0.305
other non Western 14592 0.040 0.195
edu none 14592 0.020 0.141
edu primary 14592 0.051 0.220
edu pre-vocational 14592 0.177 0.382
edu vocational 14592 0.239 0.427
edu advanced vocational 14592 0.251 0.433
edu academic 14592 0.262 0.440
The means of the dummy variables can be read as percentages of the sample scoring one on the respective variable. Of the whole sample, 77 percent of the respondents indicated that they would surely go vote if elections were being held tomorrow. In the overall sample, women are slightly overrepresented with 55 percent of the respondents being female. The average age is 46 years. People below the age of 18 were excluded from the sample as they are not entitled
20 to vote. People without migration background make up 67 percent of the sample, followed by people with a Western migration background (10 percent), people with a Surinamese background (7 percent), Moroccan background (6 percent) and Turkish background (6 percent). Most people in the sample fall into the three highest educational categories: academic education (26 percent), advanced vocational education (25 percent), and vocational education (24 percent). However, still 7 percent of the sample have had none or only primary education.
Table 3. Descriptive statistics of dependent and independent variables for the analyses of the migrant
subsample N Mean SD Dependent variables voting intention 2263 0.635 0.481 Independent variables attachment 2263 0.675 0.468 perceived discrimination 2263 0.592 0.492 efficacy 2263 0.544 0.498 sex 2263 0.545 0.498 age 2263 36.553 14.538 Moroccan 2263 0.311 0.463 Turkish 2263 0.313 0.461 Surinamese 2263 0.376 0.484 edu none 2263 0.061 0.240 edu primary 2263 0.089 0.285 edu pre-vocational 2263 0.220 0.414 edu vocational 2263 0.360 0.480
edu advanced vocational 2263 0.172 0.379
edu academic 2263 0.096 0.295
Among the migrant subsample, the number of respondents who indicate that they surely will go voting amounts to 64 percent which is 13 percent lower than in the overall sample. Same as in the overall sample, women are slightly overrepresented (55 percent). The average age in this subsample is only 37, almost 10 years lower than in the overall sample. The three ethnic minority groups Moroccans, Turkish and Surinamese are similarly strongly represented making up between 31 and 38 percent of the sample each. The distribution concerning
21 education looks quite different than in the overall sample. The highest percentage of people can be found in the categories vocational (36 percent), and pre-vocational education (22 percent), and 15 percent of the respondents have no or only primary education. The academic education category accounts for a mere 10 percent of the respondents. Concerning the group consciousness variables, it is shown that 68 percent of the respondents in the subsample indicate feeling attached to their own ethnic group. Furthermore, 59 percent report to feel discriminated on an ethnic basis and 54 percent indicate to have a sense of political efficacy (they do not disagree with the statement that people have a considerable influence on government politics).
Individual Level Effects in the Overall Sample
I will begin my statistical analysis with assessing the value of the socioeconomic core model for predicting the political participation of all Amsterdammers. According to the literature reviewed, the model boils down to looking at the effect of education, age and sex. Sex turns out to have a significant negative effect showing that women are still less likely to intend to vote in elections than men in the early 2000s (OR=0.856, p<0.001). Being a woman makes one, on average, 14 percent less likely to intend to vote. Furthermore, the likelihood to intend to vote seems to increase with age (OR=1.030, p<0.001). Taking university level education as the reference category shows that having any educational status below this significantly lowers one’s likelihood to intend to vote. The odds ratios for the effects of the separate dummy variables range between 0.153 and 0.698 and are all significant at the 0.1 percent level. Hypothesis one thus receives support. Among Amsterdammers, educational status turns out to positively predict the intention to vote in the next elections.
Next, I will look at whether having a different ethnic background than the Dutch majority group has an effect on Amsterdammers’ intention to vote. My results show that this is indeed the case. In a model including only the ethnic background predictor variables, people with any
22 kind of migration background are shown to be significantly less likely to vote than people without migration background. After introducing education, sex and age as control variables into the model, this effect fades for the group of migrants with a Turkish background but is still there for all other ethnic groups, however, with lower effect sizes. Having a Surinamese ethnic background makes one, on average, 42 percent less likely to intend to vote as compared with the Dutch majority group (OR=0.579, p<0.001). Having a Moroccan background has less strong of an impact, making one 20 percent less likely to intend to vote (OR=0.795, p<0.01). Sex, age and education stay significant predictors and their effect sizes do not change substantially when the ethnic background dummy variables are introduced. A likelihood ratio test comparing the model containing only sex, age and education (model 1) with the model containing also the ethnic background predictor variables (model 3) shows that the latter fits the data significantly better (LR chi² (df)=112.343 (5), p<0.001). Hence, hypothesis two, that having a migration background makes Amsterdammers less likely to intend to vote, receives support.
23
Table 4. Regression coefficients in odds ratios (SE) for models 1 to 3
Model 1 Model 2 Model 3
DV: voting intention
sex' 0.856 (0.036)*** 0.863 (0.036)***
age 1.030 (0.001)*** 1.028 (0.002)***
Dutch omitted omitted
Morrocan 0.366 (0.028)*** 0.795 (0.067)** Turkish 0.417 (0.033)*** 0.920 (0.080) Surinamese 0.412 (0.031)*** 0.579 (0.044)*** other_Western 0.724 (0.049)*** 0.679 (0.047)*** other_nonWestern 0.400 (0.037)*** 0.478 (0.046)*** edu_none 0.153 (0.020)*** 0.166 (0.023)*** edu_primary 0.196 (0.019)*** 0.203 (0.021)*** edu_pre_vocational 0.238 (0.017)*** 0.245 (0.018)*** edu_vocational 0.364 (0.023)*** 0.377 (0.024)*** edu_adv_vocational 0.698 (0.047)*** 0.711 (0.048)***
edu_academic omitted omitted
2004 omitted omitted omitted
2005 1.215 (0.084)** 1.294 (0.088)*** 1.239 (0.086)** 2006 1.740 (0.127)*** 1.842 (0.132)*** 1.751 (0.129)*** 2007 1.322 (0.096)*** 1.391 (0.099)*** 1.305 (0.096)*** 2008 0.934 (0.063) 1.036 (0.068) 0.935 (0.063) 2009 1.017 (0.069) 1.152 (0.077)* 1.037 (0.071) % variance at level 1 (individuals) 99.99 99.99 100.00 % variance at level 2 (neighbourhood-years) 0.00 0.00 0.00 % variance at level 3 (neighbourhoods) 0.01 0.01 0.00 N 14592 14592 14592 Wald chi² (df) 1054.91 (12)*** 494.93 (10)*** 1160.51 (17)*** *p<.05 **p<.01 **p<.001
Even though it seems to make a difference for their voting intention which of the three biggest minority groups Amsterdammers are associated with, all three groups will still be at the core of this analysis jointly. Theoretically, membership in any of these groups is considered a similar feature in that membership in one of the largest minority groups brings with it the group processes that are the core of this paper. For the following analyses, I reduce my sample to include only people with a Moroccan, Turkish, or Surinamese migration background.
24
Individual Level Effects in the Migrant Subsample
Looking specifically at the subsample of Moroccan, Turkish, or Surinamese minority group members, I will test whether the socioeconomic core model holds amongst these groups jointly. Subsequently, I will look at the effects of the independent variables of core interest, namely the variables operationalising group consciousness. The results of my analysis show that the general model holds amongst Amsterdam’s largest migrant groups. The effects of age and education are not substantially different from those in the analysis of the whole sample. The effect size of sex is the same as well but it just exceeds the five percent level significance threshold in two of the three models of the subsample analysis displayed in table five (p=0.095 in model 4; p=0.056 in model 6). This can most probably be attributed to the substantially smaller sample size of the subsample. Including controls for ethnic group shows that having a Turkish or a Moroccan background makes one significantly more likely to intend to vote as compared to having a Surinamese background5. More precisely, having a Moroccan background as opposed to a Surinamese one makes one 33 percent more likely to intend to vote (OR=1.326; p<0.05), and people with Turkish background are even 38 percent more likely to intend to vote (OR=1.380; p<0.01) than those with Surinamese background.
In a next step, I will introduce the three group consciousness variables into the model. I first introduce them all separately and then in additive models in different combinations to see how their main effects develop when either or both of the other variables are controlled for. When being introduced into the model separately, ethnic group attachment shows a significant positive effect (OR=1.225, p<0.05), perceived discrimination shows a significant negative effect (OR=0.805, p<0.05) and efficacy shows the strongest significant positive effect (OR=1.720, p<0.001) on migrant’s voting intention (see table six in attachment). When introducing the variables in different combinations, it turns out that the main effects of
5
Switching the reference group to Moroccan shows that there is no significant effect of having a Turkish as compared to having a Moroccan background.
25 attachment to one’s ethnic group and perceived discrimination are largely independent of one another. Their p-values and effect sizes do not change substantially when the respective other variable is added to the model and their interaction term is insignificant (see table six in attachment). However, introducing the efficacy variables does slightly lessen the effect of perceived discrimination and results in the effect of attachment just exceeding the p-value threshold for the five percent significance level (the p-value increases from 0.037 to 0.057). Yet, there is no significant interaction effect of either ethnic group attachment or perceived discrimination with efficacy (see table six in attachment). When all three group consciousness variables are accounted for (model 5), Amsterdammers with a migration background who perceive discrimination are 18 percent less likely to intend to vote than those who do not perceive discrimination (OR=0.817, p<0.05). Furthermore, having a sense of political efficacy unsurprisingly has a positive effect (OR=1.700, p<0.001) with the effect size indicating that someone with a sense of efficacy is 70 percent more likely to intend to vote than someone without it. Furthermore, I can see a positive association between feeling attached to one’s ethnic group and intending to vote in my sample. However, this cannot be generalised to the population with to the common minimum of 95 percent confidence of not committing a type one error but merely with 94.9 percent confidence (OR=1.212, p=0.051).
Lastly, I test the effect of main interest, namely a three-way interaction effect of the three group consciousness variables as hypothesised in hypothesis three. This effect does not show to be significant indicating that the specific combination of factors as formulated in the hypothesis does not seem to play the role I expected. Hypothesis three does thus not receive support.
26
Table 5. Regression coefficients in odds ratios (SE) for models 4 to 6
Model 4 Model 5 Model 6
DV: voting intention attachment 1.212 (0.119) 1.279 (0.141)* perceived discrimination 0.817 (0.078)* 0.872 (0.098) efficacy 1.700 (0.157)*** 1.835 (0.214)*** attachm.*discrim.*effic. 0.832 (0.140) sex 0.857 (0.080) 0.831 (0.078)* 0.835 (0.079) age 1.032 (0.004)*** 1.032 (0.004)*** 1.032 (0.004)*** Moroccan 1.326 (0.160)* 1.333 (0.164)* 1.331 (0.163)* Turkish 1.380 (0.168)** 1.393 (0.172)** 1.399 (0.173)**
Surinamese omitted omitted omitted
edu_none 0.174 (0.045)*** 0.173 (0.046)*** 0.172 (0.046)*** edu_primary 0.316 (0.076)*** 0.299 (0.073)*** 0.300 (0.073)*** edu_pre_vocational 0.341 (0.069)*** 0.332 (0.069)*** 0.333 (0.069)*** edu_vocational 0.379 (0.074)*** 0.378 (0.074)*** 0.379 (0.075)*** edu_adv_vocational 0.520 (0.110)** 0.512 (0.110)** 0.512 (0.110)**
edu_academic omitted omitted omitted
2004 omitted omitted omitted
2005 1.216 (0.190) 1.181 (0.187) 1.171 (0.184) 2006 1.691 (0.283)** 1.709 (0.290)** 1.705 (0.289)** 2007 1.277 (0.209) 1.311 (0.217) 1.308 (0.217) 2008 0.861 (0.122) 0.827 (0.119) 0.829 (0.119) 2009 1.047 (0.156) 1.106 (0.167) 1.104 (0.166) % variance at level 1 (individuals) 100.00 100.00 100.00 % variance at level 2 (neighbourhood-years) 0.00 0.00 0.00 % variance at level 3 (neighbourhoods) 0.00 0.00 0.00 N 2263 2263 2263 Wald chi² (df) 138.00 (14)*** 171.39 (17)*** 172.02 (18)*** *p<.05 **p<.01 ***p<.001 Variance Analyses
The random intercepts introduced at the neighbourhood and neighbourhood-year level in all the models show extremely low variances of the dependent variable at these levels. The rho values indicate that 0.01 percent of the variance in voting intention resides at the neighbourhood level in the first two models but as soon as the socioeconomic predictors and the ethnic group predictors are introduced together (model 3), this variance is actually
27 accounted for by the individual level factors. The variance thus completely resides at the individual level and there is no reason to assume an effect of contextual level influences.
Discussion and Conclusion
The foregoing analysis started out with once again confirming the adequacy of the socioeconomic core model for predicting political participation. The model proved useful, as expected, for the population of all Amsterdam residents as well as for members of ethnic minority groups living in the city. The subsequent core analysis of this paper was based on the idea of transferring models used to explain the voting behaviour of African-American minority group members in the US to the situation of ethnic minority group members in Europe, more particularly in the city of Amsterdam in the Netherlands. The three variables ‘attachment to ethnic group’, ‘perceived discrimination’ and ‘sense of political efficacy’ were hypothesised to, in interaction with each other, produce a group consciousness effect on voting intention. This interaction effect was not found. However, I did find significant main effects of perceived discrimination and efficacy on voting intention and the main effect of ethnic group attachment showed to approach significance.
The positive effect of having a sense of political efficacy is not surprising as many studies before have shown this to be an important factor increasing political participation (Shingles 1981). The effect size of attachment to one’s ethnic group goes in the direction expected on the basis of the literature review. Of my sample of Amsterdammers with a migration background, those who feel attached to their ethnic group are more likely to intend to vote than those who do not have this sense of attachment. This finding cannot be generalised to the population of all Amsterdammers with Moroccan, Turkish or Surinamese migration background as the introduction of the variable efficacy results in an insignificant effect of attachment. Also the effect of perceived discrimination was lowered when efficacy was introduced but also only marginally. This suggests that a small part of these two effects could
28 be mediated by efficacy. Yet, efficacy by no means fully mediates the effects. Rather, they seem to work mainly via other mechanisms.
Also the mechanism posited in hypothesis three does not seem to reflect the actual dynamics as the three-way interaction effect was not even close to significance (p=0.615). Looking at the literature for other ways to make a connection between feelings of attachment to one’s ethnic group and political participation, Olsen’s (1970) early version of the group consciousness effect presents itself. While hypothesising the same relationship as the authors whose theoretical reasoning I primarily worked with (Verba and Nie 1972; Miller et al. 1981; Simon and Klandermans 2001), Olsen outlines a different kind of causal mechanism. Instead of stressing group interests as primary sources of motivation to participate in politics, he sees social pressures within the ethnic groups to participate (or not) as central. Referring back to Lane (1959 as cited by Olsen 1970), he suggests an ‘ethnic community thesis’ according to which members of an ethnic community receive pressure to conform to group norms concerning political activism. If norms are such that the pressure to participate is high, feeling attached to one’s ethnic group might have a positive effect on political participation via this route. If one assumes the effect of attachment to one’s ethnic group to work through a mechanism of ethnic group norms, the group consciousness reasoning becomes very well compatible with the civicness approach that has been prevalent in the European literature on migrant political participation. The individual relationship between group attachment and participation might be moderated by the contextual level variable group civicness. Feeling attached to a group with a high degree of civicness might make one more likely to participate as one adopts the group’s norms of participation. On the other hand, feeling attached to a group with low civicness might not have an effect on political participation or might even have a negative effect. This would imply that the effects of attachment to one’s ethnic group on political participation should differ between groups. I have tested the additive model with
29 the three main effects in the subsamples of the different ethnic groups separately (see table 8 in attachment). According to Fennema and Tillie’s (1999) findings, the Turkish community in Amsterdam shows the highest degree of civicness, followed by the Moroccan and lastly the Surinamese community. The effects of ethnic group attachment found in the separate subgroups are not in line with what this ranking would imply. In the Surinamese subsample, attachment shows to be a significant and positive predictor, similar to the test including all three groups. With an odds ratio of 1.437 (p<0.05) it has a larger effect size amongst this group than among the whole subsample and it is significant also in the presence of the efficacy variable. In neither the Moroccan, nor the Turkish subsample does the effect of ethnic attachment show to be significant (p=0.856 and p=0.176, respectively). What is more, the coefficients of the effects in the sample go in different directions. While the odds ratio of the Moroccan subsample would indicate a slightly negative relationship (OR=0.967), the odds ratio for the Turkish subsample would indicate a positive relationship (OR=1.279).
This finding of potentially differing effects of ethnic group attachment is in line with Chong and Rogers’ (2005) discussion that there might be different ‘forms of group consciousness’ as pertinent ideas and ideologies are likely to differ between ethnic groups. The authors specifically refer to the question whether the beliefs that were measured to study group consciousness effects in one group are relevant for other groups. Group’s ideologies and concerns can be assumed to differ resulting from numerous contextual factors such as “elite messages, contact among group members, common culture or history, and group beliefs about in-group commonalities and out-group differences” (Chong and Rogers 2005, 26). Such considerations call for close-up qualitative research about mechanisms of political mobilisation within the particular ethnic groups going beyond the ‘degrees of civicness’ described in the civicness literature.
30 The results concerning the effect of perceived discrimination are surprising. From my literature review, a feeling of discrimination would be expected to strengthen sentiments of common fate with one’s ethnic group motivating political participation with the aim to fight for group interests. Not only was no interaction effect between attachment and perceived discrimination found, but discrimination turned out to have a separate negative effect showing that Amsterdammers with migration background who feel discriminated are less likely to intend to vote. The negative effect of discrimination would be an interesting aspect for further studies to follow up on, even more so as prior studies in different contexts have also found it (see Schildkraut 2005 on Latinos in the United States). Schildkraut (2005), however, also finds that this negative effect of perceived discrimination can be mitigated by self-identification with the ethnic group. This is contradicting my finding that there is no interaction effect between discrimination and attachment. Furthermore, Schildkraut (2005) discusses that while much theoretical work exists on the somewhat related negative effect of discrimination on well-being that can be mitigated by a sense of ethnic group belonging, scholars in the field have not theorised similar effects on political participation. While the objective presence of discrimination, on the other hand, got more theoretical attention (see e.g. Guterbock and London 1983), an (additional) psychological mechanism through which perceived discrimination inhibits political participation seems to be worthy of investigation.
All of these individual level effects were investigated while controlling for variance of the dependent variable at the contextual level. My literature review suggests that contextual factors could impact individual’s voting intention and should therefore be accounted for. However, I have not found indications for substantial contextual influences as shown by the lack of variance on the neighbourhood and on the neighbourhood-year level. Some of the year dummy variables showed significant effects, particularly the 2006 year dummy was consistently significant in the subsample analyses. Voting intention amongst migrants thus
31 significant differed in this year as compared to the reference category 2004 but without including as an own level in the multilevel model, we cannot know how much of the variance this is accounting for. For this, one would need more than only six comparable waves of the citizen monitor.
While the study presented in this article was successful in deriving new insights on the relationship between group consciousness and the voting intention of migrants in Amsterdam, its limitations should be made explicit so that suggestions for future research avoiding those can be derived. The above discussed assumption by Chong and Rogers’ (2005) that group consciousness is likely to take on different forms is also at the core of an important limitation of this study. Here, the innovative aspect of this article is at the same time a point requiring scrutiny. Applying theoretical work, the establishment of which is firmly connected to a certain group in a certain context, to other groups in other contexts has proven fruitful. Yet, the differences between the environment in which a theory was developed and the one it is transferred to have to be made explicit. To me, the two most important differences seem to be the following: Firstly, the groups have a different degree and legacy of group disadvantage. The disadvantage of contemporary migrants in the Netherlands is of a much less structural nature than that of African-American’s in the 1960s which was very closely tied to a legacy of legal discrimination. Connected to this, the situation of African-Americans in the 1960s was shaped by the civil rights movement which channelled dissatisfactions about ethnic disadvantage into political activity and increased the salience of the political implications of group membership. Expectations about the political self-understanding of immigrants in Amsterdam are much less clear-cut and differences between ethnic groups have shown to be present (Van Heelsum 2005, Michon and Vermeulen 2013).
Amsterdam is certainly a relevant and frequently studied case of immigrant political participation but it is also a very specific case. Compared to other EU cities, it
32 demographically distinguishes itself by its comparatively small but diverse population. Out of its roughly 800,000 inhabitants, 51 percent have some kind of migration background (Bureau Onderzoek en Statistiek 2013). A major limitation of the study is thus its case study character. As my sample is drawn from only one city, generalisations beyond Amsterdam cannot be made. At the same time, as discussed above, there are a number of good reasons to study migrants’ political participation at the city instead of at the national level. While I recommend transferring this study’s approach to different cities, the abovementioned potential for differences in group consciousness between different groups in different contexts should always be kept in mind. Studies should approach cities separately, especially when wanting to work with the ‘civicness of groups’ line of thought, as one cannot automatically assume the association between certain ethnic groups and certain degrees of civicness to be similar across cities (see Van Heelsum 2005). On the other hand, the mechanisms I have focused on are quite general so that, theoretically, they should easily be transferable to other cities. If similar patterns are found across many institutional settings, my assumption that rather general mechanisms are at work would be supported.
Last but not least, a point that future studies could improve on is the relatively rough measurement of my dependent as well as independent variables. For all my key variables, ‘voting intention’, ‘attachment to ethnic group’, ‘perceived discrimination’, and ‘sense of efficacy’, I had to rely on dichotomous measures. Survey data containing (quasi-) continuous scales for such items would substantially enrich the analysis. I thus end with the ubiquitous call for enhancing the collection of relevant survey data.
33 References
Agresti, Alan, and Barbara Finlay. 2009. Statistical Methods for the Social Sciences: Fourth
Edition. Upper Saddle River, NJ: Prentice Hall.
Bird, Karen, Thomas Saalfeld, and Andreas M. Wüst. 2011. “Voter Turnout Among Immigrants and Visible Minorities in Comparative Perspective” In The Political
Representation of Immigrants and Minorities. Voters, Parties and Parliaments in Liberal Democracies edited by K. Bird, T. Saalfeld, and A. M. Wüst, 25-65. Abingdon,
UK: Routledge.
Bloemraad, Irene, and Floris Vermeulen. 2014. “Understanding Immigrants’ Political Incorporation.” In An Introduction to Immigrant Incorporation Studies edited by J. Rath and M. Martiniello. Amsterdam: Amsterdam University Press.
Borkert, Maren and Tiziana Caponio. 2010. “Introduction: The Local Dimension of Migration Policymaking.” In The Local Dimension of Migration Policymaking edited by T. Caponio and M. Borkert, 9-23. Amsterdam: Amsterdam University Press.
Brady, Henry E., Sidney Verba and Kay L. Schlozman. 1995. “Beyond SES: A Resource Model of Political Participation.” American Political Science Review 89(2): 271-294. doi: http://dx.doi.org/10.2307/2082425.
Bureau Onderzoek en Statistiek. 2013. Amsterdam in Cijfers 2013. Amsterdam: Gemeente Amsterdam.
Caínzos, Miguel, and Carmen Voces. 2010. “Class Inequalities in Political Participation and the ‘Death of Class’ Debate.” International Sociology 25(3): 383-418. doi: 10.1177/0268580909360298.
34 Campbell, David E. 2013. “Social Networks and Political Participation.” Annual Review of
Political Science 16: 33-48. doi: 10.1146/annurev-polisci-033011-201728.
Centraal Bureau voor de Statistiek. “Begrippen: Allochtonen, Herkomstgroepering.” Last modified 2014.
http://www.cbs.nl/nl-NL/menu/themas/dossiers/allochtonen/methoden/begrippen/default.htm?ConceptID=31 5.
Chong, Dennis, and Reuel Rogers (2005). “Reviving Group Consciousness.” In The Politics
of Democratic Inclusion edited by C. Wolbrecht and R. E. Hero, 45-74. Philadelphia,
PA: Temple University Press.
Dienst Onderzoek en Statistiek. 2005. Stadsdelen in Cijfers 2005. Amsterdam: Gemeente Amsterdam.
DiPrete, Thomas. A. and Jerry. D. Forristal. 1994. “Multilevel Models: Methods and Substance.” Annual Review of Sociology 20: 331-357. doi: 10.1146/annurev.so.20.080194.001555.
Fennema, Meindert, and Jean Tillie. 1999. “Political Participation and Political Trust in Amsterdam: Civic Communities and Ethnic Networks.” Journal of Ethnic and
Migration Studies 25(4): 703–26. doi: 10.1080/1369183X.1999.9976711.
Gaardsted Frandsen, Annie. 2002. "Size and Electoral Participation in Local Elections."
Environment and Planning 20(6): 853-870. doi: 10.1068/c0228.
Gallego, Aina. 2010. “Understanding Unequal Turnout: Education and Voting in Comparative Perspective.” Electoral Studies 29(2): 239–248. doi: 10.1016/j.electstud.2009.11.002.
Geys, Benny. 2006. “Explaining Voter Turnout: A Review of Aggregate-Level Research.”
