The Segregation Effect: How Residential Isolation Shapes Ethnic Minority Representation in England

Abstract

Extant literature argues that ethnic minority representation in plurality systems will benefit from the presence of sizable co-ethnic populations. I argue that the threshold for election depends not only on a minority population's group size but a district's level of segregation. I show that residential segregation can facilitate the increased representation of ethnic minority populations. Contrary to the prevailing literature, however, I find that increased segregation levels in cities with sizable minority populations decrease the percentage of co-ethnics elected to office. I support this argument with evidence from an original dataset on the local representation outcomes of Muslims in England between 2011 and 2021, which covers 434 district council elections. Using threshold modelling, I introduce the concept of the population threshold, above which increases in segregation level decrease Muslim representation. This article contributes to the electoral geography literature on ethnic minority representation.

Residential segregation has become a persistent reality for ethnic minorities in Europe originating from Muslim-majority countries (Body-Gendrot and Martiniello 2000; Musterd and Ostendorf 1998). Research has suggested that segregation may benefit the election of ethnic minorities to local office as a result of increased co-ethnic concentration, particularly in plurality first-past-the-post (FPTP) systems. These studies contribute to robust literature on the electoral geography of plurality systems, which is primarily concerned with explaining minority representation outcomes as a function of group size. This literature tends to treat group size and segregation level as positive influences on minority representation. It concludes that the electoral leverage afforded by sizable minority populations living in high levels of segregation will lead to increased representation outcomes.

Absent from these studies is a consideration of how sub-district patterns of residential concentration aggregate at the district level and interact with the proportion of minority residents, particularly in cities and towns with large minority populations. Drawing on an original dataset of local election outcomes across England, I demonstrate how segregation interacts with population demographics to shape the representation of English Muslims.¹ Contrary to the prevailing literature, I show that there is a non-linear relationship between Muslim population size in a given district and their representation outcomes. The segregation level conditions the effect of minority population sizes and increases the likelihood that Muslims will be elected to local office. However, the effect of segregation on representation outcomes is reduced in districts with sizable Muslim populations. I propose the concept of a ‘population threshold’, in which election to local office depends not only on the proportion of Muslims in a given district but on its level of segregation. I provide evidence for my findings by drawing on over 33,000 candidate observations across more than 11,000 ward-level elections. Results are aggregated at the local authority level to produce an original dataset of 434 district-level election outcomes between 2011 and 2021.

This article contributes to the literature on the electoral geography of political representation. First, it shows how the geography of co-ethnicity matters for representation outcomes. Secondly, it casts doubt on the expected linear relationship between an ethnic group's size and its electoral power. By demonstrating the conditioning effect of segregation, the findings call for a re-evaluation of this relationship.

Group Size, Segregation, and Minority Representation

Scholarship on representation in plurality systems rests on the assumption that there is a positive relationship between a group's size and its election outcomes (Katz 1997; Powell 2000; Powell 2004; Rae 1967). Within this branch of literature, research has shown that co-ethnicity is a powerful determinant of voter behaviour (Chandra 2007; Cutts et al. 2007; Dancygier 2017; Wolfinger 1965). Minority candidates that share an ethnic background with voters are likely to receive a large proportion of votes from their co-ethnics, known as ethnic affinity voting (Bird et al. 2016; Matson and Fine 2006). Here, the size of the ethnic minority population serves as the mechanism that determines the share of co-ethnics elected to local office. If ethnicity is a salient predictor of voting behaviour, then the parties will be more likely to nominate Muslim candidates in wards where there are sizable Muslim populations, in anticipation of the co-ethnic vote (Hypothesis 1a).

Scholarship has shown that minority communities also influence the political preferences of their co-residents and their ability to mobilize following candidate nomination (Barreto 2007; Cutts et al. 2007; Fieldhouse and Cutts 2008; Fisher et al. 2015; Landa, Copeland, and Grofman 1995; Maxwell 2012; Zingher and Farrer 2016). As group identification is shown to shape voter behaviour, then the impact of a group's political preferences should rise as group size increases. Therefore I expect a positive correlation between the size of a district's Muslim population and the percentage of Muslim candidates elected to local office (Hypothesis 1b).

Segregation and Representation

The spatial distribution of minority groups, particularly their level of residential segregation, can also shape their representation outcomes. Measures of segregation account for the distribution of groups across a given spatial unit, typically at the city level. The electoral geography literature examining the link between segregation and representation has primarily focused on understanding outcomes in the United States, although research has demonstrated similar patterns of residential segregation in European cities (Costa and de Valk 2018; Malmberg et al. 2018; Murie and Musterd 1996; Rogne et al. 2020).

This branch of scholarship suggests a positive relationship between segregation and minority representation (Olsen 1970; Shingles 1981; Verba and Nie 1972). The spatial dynamics of community membership are facilitated by the FPTP system, which privileges residentially isolated groups. According to this research, residentially segregated neighbourhoods serve as foundations for the creation of group-specific interests (Olsen 1970; Schlichting, Tuckel, and Maisel 1998; Verba and Nie 1972). Segregation unites residentially isolated populations to create group consciousness (Shingles 1981), shape political preferences (Cho, Gimpel, and Dyck 2006; Fisher et al. 2015; Johnston et al. 2000; Johnston et al. 2004), and heighten mobilization capacity (Maxwell 2012), thereby increasing minority representation outcomes (Dancygier 2014). Based on these findings, segregation should bind residents together, resulting in proportionally more co-ethnic candidates being elected to local office as segregation increases (Hypothesis 2).

The literature reviewed above suggests that in England, wards with sizable Muslim populations should have increased numbers of co-ethnic candidates elected to office due to co-ethnic clustering.² However, it would be illogical to conclude that Muslim representation necessarily benefits from high levels of segregation in the aggregate. England's plurality FPTP system is composed of sub-district, ward-level electoral contests, wherein one to three candidates are elected. The Muslim population size and segregation level may interact to create a ceiling effect; in districts with sizable Muslim populations, increases in segregation levels should further concentrate co-ethnics within a select number of wards, reducing the electoral scope of Muslim candidacy. Given the small number of seats available within a given ward, the prospects of a Muslim being elected to a local office are limited once the ceiling is reached.

Put simply, in local authorities with sizable Muslim populations, Muslim residential confinement to a select number of wards (that is, high segregation) should dampen their overall council representation. Conversely, local authorities with sizable Muslim populations that are spread out across several wards (that is, low segregation) should increase the electoral scope of their candidacy, thus displaying higher aggregate levels of Muslim representation as a result of residential dispersion. Therefore I expect that in districts with sizable Muslim populations, increases in segregation will reduce the overall percentage of Muslims elected to local councils (Hypothesis 3).

Data, Measures, and Methods

England is a fitting case study to examine the impact of segregation on ethnic minority political representation. It has Western Europe's second-largest Muslim population, and the local representation of English Muslims remains an important indicator of their political integration.³ At the same time, segregation has occupied a visible place in political discourse given relatively high levels across the country (Appendix A).

In the following analyses, the outcome of interest is the share of Muslims elected to office in local authorities. The dependent variable is measured as the percentage of Muslim candidates elected to a local council out of all candidates, both Muslim and non-Muslim.^{4, 5} To obtain the percentage of Muslims elected to local office, I first code each candidate running for district council in a given ward. Next, I code the outcome of each ward-level election before aggregating the results at the city level. The data covers over 33,000 candidate observations across more than 11,000 ward-level elections in ninety-five local authorities between 2011 and 2021, for a total of 434 local authority election outcomes. The sample includes election data from both urban and rural districts to obtain the most representative illustration of Muslim election outcomes. The ninety-five local authorities included in the sample contain nearly 78 per cent of England's Muslim population.

For the purposes of this analysis, identification as a Muslim is grounded in a sociological indicator of ethnoreligious origin rather than assumed personal piety. To identify a candidate's ethnic background, I rely on an onomastic approach that involves hand coding first and last names. Although name coding software exists, it is challenging to identify names on an ethnic basis. Hand coding awards greater certainty regarding the candidate's ethnic identity. In order to ensure valid coding outcomes, I consult secondary sources and, when available, newspaper articles and candidate websites. These sources frequently mention the candidates’ ethnic and religious backgrounds, as well as their countries of birth (Appendix B).

To understand the factors influencing Muslim election outcomes, I include a Muslim population variable, which is measured as the percentage of Muslims in a given local authority using data from the 2011 census. To conduct the initial ward-level analysis analyzing the relationship between Muslim group size and the likelihood of a Muslim being nominated for local office, I use the percentage of Muslims in a given ward.

In addition to the Muslim population percentage, I theorize that the level of segregation should matter for Muslim representation outcomes. To operationalize segregation I use the dissimilarity index, which serves as the standard segregation measure (Duncan and Duncan 1955; Massey and Denton 1988; Taeuber and Taeuber 1976; White 1986). The index ranges from zero (complete integration) to 100 (complete segregation) and measures the proportion of a given group that would have to relocate to achieve a uniform population distribution. I compute dissimilarity indices from Lower Layer Super Output Area (LSOA) census data of Muslim and white populations in a given English district. The index is calculated as follows:

$$D = \displaystyle{1 \over 2}\mathop \sum \limits_{i = 1}^n \left\vert {\displaystyle{{w_i} \over {W_T}}-\displaystyle{{m_i\;} \over {M_T}}} \right\vert $$

where W _T and M _T are white and Muslim populations in a district where w _i and m _i are the Muslim and white populations in LSOA i, respectively. In total, the dataset includes over 50,000 LSOAs that comprise the ninety-five local authorities. Geographic and population data are collected from the 2011 census. LSOA results are aggregated to produce a district-level dissimilarity index.

I also include several control variables. First, I include an economic deprivation variable, which is operationalized as the percentage of households in a given English district that fall below the 60 per cent median standard of living using 2011 census data. Second, I include a dummy variable to denote a Labour-dominated council. Third, I include a district magnitude variable, calculated as the average of the number of councillors elected by district size, ranging from one to three. Fourth, I create a diversity variable using a diversity measure using the Herfindahl index.⁶ Fifth, I include a voter turnout variable, operationalized as the percentage of voter turnout in the given local election year. Finally, I control for urban and rural distinctions. I operationalize the Urban variable using local authority population density data from the 2011 census. Additional explanations of these controls can be found in Appendix C.

I begin by using logistic regression to investigate whether ward-level Muslim population size influences the likelihood of a Muslim being nominated for local office. The predictor variable % Muslim population was tested a priori to verify that there was no violation of the assumption of linearity of the logit. Second, to analyze the relationship between the independent variables and the local authority-level election outcomes, a series of multiple linear regressions were conducted, which are appropriate when the dependent variable is continuous. The reported regression coefficients measure the relationships between the independent variables and the percentage of Muslims elected to office in a given local authority.

Following the initial series of regressions, I used the Johnson-Neyman (J-N) technique as a robustness test. This allows for an evaluation of the interaction's effects in the model relative to the other variables' means. I then construct a threshold model based on the interaction to further investigate the relationship between segregation and Muslim representation on local councils.

Results

I first analyze whether the likelihood of nominating a Muslim candidate to political office is influenced by the size of the Muslim population in a given ward using logistic regression. Figure 1 visualizes the output of the logistic regression and shows a fitted line plot of the observed and predicted values of the Muslim candidate's nomination. As Figure 1 indicates, the likelihood of a Muslim candidate nomination rises as the size of a ward's Muslim population increases, consistent with Hypothesis 1a.

Figure 1. Muslim nomination given population percentage, ward-level.

Notes: The dependent variable is a binary variable measuring whether or not a Muslim is nominated as a candidate. Zero indicates no Muslim candidate nomination and ‘1’ indicates a Muslim candidate nomination. The fitted line shows the predicted probability of a Muslim candidate nomination. Tower Hamlets outlier excluded.

Table 1 presents the results of the OLS models estimating the share of Muslims elected to local councils. For each model, the dependent variable is the percentage of Muslims elected to office in a given local authority. In Model I, I establish the significance of the control variables before demographic variables are included. In Model II, I add the demographic variables % Muslim population and Diversity prior to considering the spatial dynamics of population distributions. Model III includes the variable Segregation level along with an interaction term between the Segregation level and % Muslim population to test whether the effects are reduced when the relative group size of Muslims is larger.

Table 1. The election of Muslims to local councils in English local authorities

Notes: Models I-III show the results of the OLS regressions. The dependent variable is the percentage of Muslims elected to office in a local authority.

***p < 0.001, **p < 0.01, *p < 0.05. Robust standard errors are in parentheses.

The results shown in Model I indicate that the control variables, Economic Deprivation and Voter Turnout, are negative and are significant predictors of Muslim representation, whereas the Labour variable is significant and positive. However, only Voter Turnout remains significant following the addition of % Muslim population and Diversity in Model II, after including the Segregation level and the interaction term in Model III.⁷

The results displayed in Model II show that Muslim population size, measured as the percentage of Muslims in a given district, is a statistically significant and positive predictor of Muslim representation, as predicted by Hypothesis 1b. It remains significant following the inclusion of the Segregation level variable in Model III. This finding supports the electoral geography thesis, which suggests that increases in a minority group's population size will facilitate its representation outcomes. The insignificance of the remaining control variables in Models II and III suggest that demographic and spatial indicators are more powerful determinants of Muslim representation outcomes than economic predictors, electoral rules, and the presence of a majority Labour council.

The results displayed in Model III indicate that segregation has a positive and significant effect on the share of councillors who are Muslim in a local district, which is consistent with Hypothesis 2. Despite the positive and significant effects of the Segregation Level and % Muslim population, the results in Model III show that the interaction term is negative. As the Muslim population rises, increases in segregation reduce the positive and significant impact of the main effects, which is consistent with Hypothesis 3.⁸

Although the interaction term displayed in Model III is significant, the model may fail to reflect the majority of the dataset's values. The interaction may occur at the end of the dataset or represent a theoretical interaction in an area of the model where no data points exist. To establish the model's robustness, I used the J-N technique and restricted the generalization of the results to the sample data. Rather than testing for significance at fixed values of segregation level, the J-N technique solves for values of segregation level for which the effect of Muslim population percentage on Muslim representation becomes non-significant.

Figure 2 illustrates the results of the J-N technique by visualizing the effect of the moderator (Segregation Level) on the coefficient of the focal predictor (% Muslim population). It indicates that the effect of segregation on the conditional expectation of the percentage of Muslims elected to local office diminishes as the Muslim population grows. The average marginal effects are non-zero for the majority of the dataset, except for local authorities with Muslim populations between 12 and 15 per cent.

Figure 2. Average marginal effects of segregation level on the conditional expectation of Muslims elected, J-N technique.

Notes: The solid line outlines the conditional effects. The shaded areas cover the 95 per cent confidence interval surrounding the effects. The rug plot depicts the values of the Muslim population included in the sample.

The Population Threshold

The findings displayed in Table 1 suggest a non-linear interaction between segregation level and a district's Muslim population size.⁹ Table 2 shows the results of the threshold model given the interaction effect displayed in Table 1. When segregation levels remain constant, increases in the Muslim population raise the predicted percentage of Muslims elected to office in a given local authority. However, as Figure 3 illustrates, there is a marked change in the effect of segregation as the percentage of a district's Muslim population moves from 5 to 10 per cent, and increases in the district's segregation level now decrease the predicted percentage of Muslim councillors. The results indicate a ‘population threshold’ wherein the impact of segregation shifts in districts where the Muslim population is 6 per cent and above, leading to reduced levels of Muslim representation (full results of the threshold are shown in Appendix H, Table H1).

Figure 3. Predicted Muslim representation given segregation level and Muslim population.

Notes: The solid lines trace the expected linear prediction at the minimum and maximum segregation levels. The dependent variable is the percentage of Muslims elected to office in a local authority.

Table 2. Predicted percentage of Muslims elected to office in a local authority, threshold model

Notes: The estimated percentages presented in the table are derived from the interaction effect in Model III of Table 1. Segregation levels of zero and 100 are excluded given their unlikelihood.

How can we interpret the finding that districts with Muslim populations of 6 per cent and above will display reduced levels of Muslim representation as segregation levels rise? Given that minorities are disproportionality likely to live in conditions of segregation (Logan, Stults, and Farley 2004; Massey and Denton 1993; Trounstine 2016), Muslims are increasingly confined to a few segregated wards in districts with sizable co-ethnic populations and high segregation levels. In districts without a co-ethnic presence, Muslims have little chance of being elected. Districts with sizable Muslim populations and moderate levels of segregation thus exhibit higher representation levels; the population is spread out across several wards, increasing the number of electoral opportunities and, with it, the chance to become elected. As the level of segregation rises, however, Muslims are increasingly concentrated in a select number of wards and, outside of these wards, have little chance of election. Conversely, when the segregation level is held constant, increases in a district's Muslim population lead to increased co-ethnic representation.

The mean values of the sample's Muslim population size indicate that nearly 50 per cent of cases fall above the population threshold of 6 per cent. In slightly less than half of the districts included in the sample, increases in the segregation level will lead to reduced levels of Muslim representation.

The threshold of this interaction becomes important, from a comparative perspective, for Muslims running for local office. Take the examples of the London boroughs of Ealing and Brent, both of which have Muslim populations that are above the population threshold, and which both displayed significant variation in the percentage of Muslim candidates elected to local office in 2018. Ealing has a segregation level of 56 and Muslims constitute 16 per cent of the population.

Following the 2018 elections, Muslims represented 2 per cent of all candidates elected to office. In nearby Brent, which contains the same share of Muslims (16 per cent) but where the segregation level is lower (38.3), Muslims represented 12 per cent of all candidates elected to local office in the same year.

The ceiling effect created by the interaction between segregation and Muslim population size can explain these divergent outcomes. In Brent, 50 per cent of the Muslim population was spread across nine wards out of twenty-one, whereas in Ealing, nearly 50 per cent of the Muslim population was concentrated in five wards out of twenty-three. High levels of segregation in Ealing decreased the overall electoral power of the co-ethnic vote despite high levels of candidate nomination in the five wards with sizable Muslim populations. Conversely, as a result of the increased population dispersion in Brent, more Muslims were elected to local office. In 2018, Muslims were represented in twelve wards in Brent, but only six in Ealing.

Discussion

This article clarifies the role of minority population size and segregation level for minority representation. The electoral geography literature has overwhelmingly looked at population size to explain representation outcomes (Katz 1997; Powell 2000; Rae 1967). For minorities, co-ethnicity is a powerful determinant of representation (Chandra 2007; Cutts et al. 2007; Dancygier 2017; Wolfinger 1965), particularly in areas with sizable co-ethnic populations. Their geographic concentration is expected to encourage ethnic affinity voting, thereby increasing group representation (Bird et al. 2016; Matson and Fine 2006; van der Zwan, Tolsma, and Lubbers 2020).

The results confirm the significance of group size and co-ethnicity for minority representation. However, the findings cast doubt on the presumed positive relationship between minority population size, spatial concentration, and representation. Increases in segregation do not necessarily produce increased co-ethnic representation. Rather, above the population threshold, rising segregation decreases minority election outcomes. The interaction between segregation and population size creates a ceiling effect; in local authorities above the threshold, increases in segregation decrease the electoral scope of minority candidacy and limit Muslims' aggregate election outcomes.

The findings provide insight into the future of minority representation across Europe and invite further investigation into the interaction between segregation levels and group size in other institutional contexts. Similar patterns may exist across plurality systems, although the threshold will likely differ between countries and across periods of time. Segregation also has the potential to shape minority representation in other electoral systems. While it may not offer the same electoral incentives outside of FPTP systems, it could allow minorities to leverage their residential isolation to push for increased co-ethnic candidates in systems that use proportional representation, including party lists and mixed-member proportional voting. Taken together, the results suggest that segregation may shape minority representation across a variety of electoral systems.