Reactive religiosity? The longitudinal relationship between ethnic harassment, religious identity and wellbeing in the UK

ABSTRACT

Ethnic harassment is a common experience among ethnic and religious minorities in the UK and can reduce wellbeing. However, previous research on ethnic minorities suggests that experiences of harassment could increase identification with religious minority groups, and that a strong religious identity may buffer against some of the negative effects of harassment. Using panel data from the UK Household Longitudinal Survey (2009–2021), we investigate whether people who experience ethnic harassment have a subsequent increase in religious identification or practice, and to what extent this affects their wellbeing. We find evidence for a religious buffering effect which reduces distress following ethnic harassment for Christians. Similarly, we observe an increase in religiosity among Christians, but not other religious groups following ethnic harassment. Finally, we examine whether stronger religious identification would increase the likelihood of harassment experiences, and find only partial evidence for this reverse effect.

KEYWORDS:

• Religious identity
• ethnic minority
• harassment
• wellbeing
• UK
• longitudinal

Introduction

Ethnic harassment, discrimination and hate crimes are common experiences, particularly among ethnic and religious minorities, and being the target of such behaviour can lead to insecurity, stress, lower wellbeing and self-esteem, and higher stress and risk of chronic illness (Nandi, Luthra, and Benzeval 2020; Schmitt et al. 2014; Wallace, Nazroo, and Bécares 2016). However, there may be social and psychological mechanisms that ameliorate the negative effects of such experiences, and specifically, there is reason to believe that religion may play a role in this relationship.

According to social identity theory, threats to a person’s self- or group interests could result in their increased identification with their social group (Correll and Park 2005; Fritsche, Jonas, and Kessler 2011; Tajfel and Turner 1979). This could in turn increase a psychological sense of belonging (Baumeister and Leary 1995), opportunity for collective action (van Zomeren, Postmes, and Spears 2008) and social support from other group members (Chen 2010). The rejection-identification model (Branscombe, Schmitt, and Harvey 1999) similarly argues that a minority group member’s experience of rejection by the majority group could cause stronger in-group identification, thus buffering against the negative effects of discrimination.

A number of studies in the sociology of religion and related fields, similarly find that insecurity or threats to the individual or their social group could result in increased religious identification, belief of practice (Chen 2010; Immerzeel and van Tubergen 2013; Norris and Inglehart 2004; Sibley and Bulbulia 2012; Storm 2017). Religious identity and practice are in turn associated with reduced stress, higher than average wellbeing, and an increased sense of inclusion (Inglehart 2010; Lim and Putnam 2010; Park 2005).

We would thus expect that experiences of harassment and other negative experiences attributed to religious or ethnic group membership would result in increased religious identification, as well as increased religious practice. Previous research on similar questions has typically focused on the implications for ethnic, rather than religious identity (Nandi, Luthra, and Benzeval 2020; Ramos et al. 2012). Those that have investigated the relationship with religiosity have largely relied on cross-sectional data (Fleischmann, Phalet, and Klein 2011; Stuart and Ward 2018), which does not allow them to account for the order of events.¹

In this study, we use longitudinal panel data in order to investigate whether people who experience ethnic harassment have a subsequent increase in religious identification or practice, and to what extent this affects their wellbeing. We also investigate differences between Christians and Muslims, and the potential reverse effect, namely that increased religious identification would increase the likelihood of experiencing harassment attributed to ethnic, racial or religious identity.

The protective effect of religion

Previous literature shows that experiences of ethnic and racial harassment and discrimination can reduce mental wellbeing over time and this has been cited as an explanation for systematic ethnic health inequalities (Nandi, Luthra, and Benzeval 2020; Schmitt et al. 2014; Wallace, Nazroo, and Bécares 2016). In addition to the effects of direct experiences of harassment, perceptions of racism, and anticipating harassment and discrimination based on one’s own or others’ experiences are similarly associated with reduced mental health (Karlsen and Nazroo 2002).

There are three main reasons why religiosity might be a source of resilience against the negative effects of ethnic and racial harassment. Firstly, all forms of social identity and belonging could increase wellbeing. Secondly, religious rituals and beliefs could buffer against a variety of different forms of insecurity and stress. Thirdly, religiosity could be a symbol of the specific ethnic or religious ingroup identification which is being threatened or insulted by the harassment experience.

Belonging to a social group is considered a human need, and is associated with higher wellbeing and better health (Baumeister and Leary 1995; Correll and Park 2005; Tajfel and Turner 1979). Religiosity is one form of group identity, which has been associated with increased wellbeing, and reduced stress from adverse circumstances and life events (Storm 2017). Religion has also been associated with a reduction in physical stress responses such as high blood pressure (Chen and Contrada 2007). In a recent meta-analysis of longitudinal studies, Garssen, Visser, and Pool (2021) conclude that there is a modest, but significant positive effect of religiosity on mental health. There are two key mechanisms which could explain this relationship. Firstly, religion may fulfil the same function as other group identities by increasing participation in social networks and groups, and thus increase a sense of belonging and social support (Chen and Contrada 2007; Lim and Putnam 2010). Secondly, there may be added value from religion’s supernatural dimension. Religious beliefs could potentially influence meaning-making in coping with stress (Park 2005), for example by the belief in the afterlife providing assurance (Bradshaw and Ellison 2010) and the authority of God compensating for personal lack of control (Stanley and Kay 2022).

The rejection-identification model (Branscombe, Schmitt, and Harvey 1999) argues that a minority group member’s experience of rejection by the majority group could cause stronger in-group identification, thus buffering against the negative effects of discrimination. This has been supported by several empirical studies in social psychology, across a number of different geographic locations and minority groups including homosexuals (Bourguignon et al. 2020), international students (Ramos et al. 2012) and Muslims (Fleischmann, Phalet, and Klein 2011). Portes and Rumbaut (2001) use the term “reactive ethnicity” to describe the persistence or increase in ethnic identification among second-generation immigrants in the US. In the UK, Nandi, Luthra, and Benzeval (2020, 343) found that “the negative association between [ethnic harassment] and mental health was reduced for UK born minorities with strong ethnic identities and co-ethnic friendship networks”. It is possible that the persistence of religious identity in the face of wider processes of secularisation is in part a reaction to racism, discrimination and perceived injustice (Connor 2010; Fleischmann, Leszczensky, and Pink 2019; Voas and Fleischmann 2012), and this process could be described as “reactive religiosity” (Platt 2014).

Religious congregations may be more attractive communities for descendants of immigrants than communities based on the ethnic or national origin of their parents or grandparents. For example, a wealth of evidence suggests that young Western Muslims in recent decades perceive and define themselves in terms of their religious affiliation rather than their ethnicity and country of origin (e.g. Peek 2005; Platt 2014; Ysseldyk, Matheson, and Anisman 2010). However, ethnic and religious communities are likely to fulfil similar functions in providing social insurance and a sense of belonging. A potential additional benefit of religiosity for minorities who face racial harassment is that the supernatural beliefs may be a psychological resource for coping with insecurity and lack of control (Stanley and Kay 2022).

Together this evidence leads us to the first hypothesis H1: Religious identification and participation reduces psychological distress following experiences of ethnic and religious harassment.

The risk effects of minority religion

There is, however, reason to believe that the positive effect of religion is context-dependent. Specifically, religion would be less effective as a buffer against insecurity when it is not shared with others. Social identity increases within-group assimilation and is also most effective when the group is internally similar on key characteristics (Richerson and Boyd 2005, 212–213; Tajfel and Turner 1979). As Molteni (2021, 147) points out, “religion in modern times is considered less and less as the preferred way through which to cope with threatening situations”. When a specific religious worldview and accompanying rituals are not part of the larger social norm, nor ties the believer to a community of likeminded others, it should not have as much power to console.

The majority of the UK population identify as either nonreligious (52 per cent) or Christian (38 per cent), and weekly church attendance is at a mere 11 per cent according to the 2018 British Social Attitudes Survey (Voas and Bruce 2019, 21–25). However, among non-Christian religious minorities, 40 per cent attend religious services at least once a week (Voas and Bruce 2019, 25), and the majority of these attend ethnically homogenous places of worship (Sobolewska et al. 2015, 279).

In their review of literature on group identity and harassment, Wolfram, Linton, and McDuff (2018) argue that while group identity may protect against the negative effects of harassment experiences, it could also intensify these effects, because a strong group identification may make such experiences appear more personally relevant. In a longitudinal study, Sellers and Shelton (2003) find that racial centrality among African Americans increased perceptions of discrimination. In an experimental study of Latin Americans in the US, McCoy and Major (2003) conclude that “for highly group identified individuals, prejudice against the ingroup is a threat against the self”. Ysseldyk, Matheson, and Anisman (2011) find similar effects on religion, namely that intrinsically religious people react more negatively to threats targeting their religious identity, undermining the otherwise positive effects of religiosity.

Thus, while on the one hand, we would expect minority communities’ religiosity to be effective in reducing psychological distress, religious practice could simultaneously be a source of distress for minorities when combined with threat from the majority. Wolfram, Linton, and McDuff (2018) found that “high ethnic identity salience” impaired wellbeing in black and ethnic minority individuals who experienced harassment, whereas it buffered the negative experiences for white respondents. If this is due to majority/minority status, we should expect to find a similar distinction between minority and majority religions. Christians, whose religious traditions are a more accepted part of the majority culture, are thus expected to benefit more from the protective effect of religion than adherents of minority religions. Muslims, on the other hand, should experience increased distress from harassment, the more religiously active they are.

This can be formulated as a second hypothesis H2: Religiosity increases psychological distress following experiences of ethnic and religious harassment among Muslims.

Does ethnic harassment increase religiosity?

The potential benefits of religious group identity as a social and psychological coping mechanism should also be expected to increase religiosity in response to distress. According to rejection-identification theory (Branscombe, Schmitt, and Harvey 1999), identification with a disadvantaged group should increase in the face of discrimination and prejudice. Such increased identification could manifest as increased participation in group activities, symbolic commitment in the form of clothing, diet, etc. or even in in-group bias, outgroup-prejudice and reduced identification with other social groups. For example, Fleischmann, Leszczensky, and Pink (2019, 971) showed that minority youth in Germany who report more frequent discrimination “lower their (German) national identification over time, which in turn predicts increased minority identification”. Platt (2014, 65–66) found that perceptions of religious discrimination among ethnic minorities in the UK, increased feelings of commonality with co-religionists.

A number of studies in the sociology and social psychology of religion, find that an array of adverse life events and measures of deprivation are associated with both subjective religiosity and religious attendance (Chen 2010; Immerzeel and van Tubergen 2013; Norris and Inglehart 2004; Sibley and Bulbulia 2012; Storm 2017; Ysseldyk et al. 2014). The psychological mechanism could be the same as for other social identities (Tajfel and Turner 1979), but it is also possible that religion may be particularly appealing in situations of insecurity (Stanley and Kay 2022).

That said there are also critics who question both the size and universality of this effect. One criticism is that religious change takes place largely as by generational replacement and that the importance of individual religious change over the life course is negligible (Bruce and Voas 2016; Voas and Chaves 2016). Moreover, in secularising contexts, religiosity is an increasingly less preferred way of dealing with adversity (Molteni 2021, 147), and as a result, the relationship between insecurity and religion is much less clear on the individual than the national level (Höllinger and Muckenhuber 2019). Critics also point out that the effect is asymmetrical; while religious people may become non-religious in response to increased security, the nonreligious are unlikely to become religious in response to insecurity (Storm 2017; Voas and Chaves 2016, 1531). A reactive increase in religious belief or behaviour is thus only likely among those who are already religious or have been socialised into a religious tradition (Molteni 2021, 139). This describes much of the ethnic and religious minority population in the UK.

Ethnic and religious harassment can be regarded as “identity relevant stressors” (Wolfram, Linton, and McDuff 2018) insofar as ethnic and/or religious identity is a salient part of the victim’s identity. In such cases, the harassment experience represents a double threat to both personal and group identity, which is expected to elicit a particularly strong response in the form of increased in-group identification (Fritsche, Jonas, and Kessler 2011). For example, Ysseldyk et al. (2014, 353) found that individuals who experienced religious or ethnic discrimination identified more with their religious group, and reported more religious engagement. In many countries, including the UK, minority religious identity is closely associated with ethnic or national identity. For example, islamophobia can be experienced and understood as a form of racism (Allen 2005). Hence, experienced or anticipated harassment on the basis of either ethnic or religious identity should have the potential to strengthen religious identity and practice.

This leads us to the third hypothesis, H3: Experiencing ethnic or religious harassment increases subsequent religious identification and practice.

Does religiosity increase harassment?

In her study of reactive identities in Britain using cross-sectional data, Platt (2014, 68) cautions that although she found that perceived religious discrimination increased feelings of commonality with co-religionists, “there is the possibility of reverse causation in that association”. There are two possible causal pathways for religiosity to increase experiences of harassment.

Firstly, being actively religious may increase the physical risk of experiencing ethnic or religious harassment. For example, participating in Muslim services and events, wearing Muslim clothing such as the hijab, and socialising with other Muslims may make individuals more exposed and identifiable targets for islamophobic harassment (Ramos et al. 2012, 654).

Secondly, those whose religious identity are more salient to them may be more likely to perceive negative experiences as discrimination or harassment, or to interpret incidents of harassment or discrimination as being due to their religious identity. For example, Obaidi et al. (2018) found that the strength of religious group identification was associated with a higher experienced level of threat, which in turn impacted on expression of hostility to religious outgroups. Sellers and Shelton (2003) similarly find that the centrality of racial identity was positively associated with subsequent perceived racial discrimination among African-American college students.

This leads us to the fourth hypothesis, H4: Religious identification and practice increases the likelihood of subsequent ethnic or religious harassment experiences.

Data and methods

The data comes from the UK Household Longitudinal Survey (UKHLS), also known as Understanding Society, and follows a panel of households in 12 waves of data, collected annually from 2009 to 2021 (University of Essex, Institute for Social and Economic Research 2022a). We analyse data from individual participants who were asked about experienced harassment in waves 1, 5, 7, 9, and 11 and questions about religiosity, asked in waves 1, 4, 8 and 12. Questions about wellbeing were asked in every wave. Data is drawn from the general population sample (GPS) a nationally representative sample of around 26,000 UK households, and the ethnic minority boost sample (EMBS), of around 4,000 households with at least one person from an ethnic minority background. The EMBS was drawn from high ethnic minority concentration areas where 80 per cent of the UK’s five major ethnic minority populations live (University of Essex, Institute for Social and Economic Research 2022b, 10).

Harassment variables

In Wave 11 (2019–2020) questions about harassment were asked of the general population sample (GPS), whereas in the previous waves they were asked of the so-called “extra 5 minutes”-sample (E5MS), meaning the ethnic minority boost sample (EMBS), along with ethnic minorities in the GPS living in low ethnic minority concentration areas and 500 households randomly selected from the GPS. The questions about experienced harassment asked about being physically attacked, insulted, called names, threatened or shouted at. The questions about anticipated harassment asked about feeling unsafe and avoiding a public place. The question was phrased as “In the last 12 months, have you been [physically attacked] in any of these places? If so, which ones?”, with twelve options including, “school”, “work”, “public transport”, “at pub/disco/club” as well as an “other places” category.² When someone answered they had been attacked, insulted, called names, threatened or shouted at, had avoided somewhere or felt unsafe in the last 12 months, they were asked why: e.g. “Were you physically attacked for any of these reasons? If so, which ones?” with 8 options including “your sex”, “your age”, “your ethnicity” and “other reason”.

These variables were combined into two variables: one that records all experienced harassment that was due to ethnicity, nationality or religion in the past 12 months, and one that does the same for anticipated harassment. We then combine the measure for the previous waves together, such that the final variables used in the analysis measure whether the participant experienced or anticipated harassment due to their ethnicity, nationality or religion (ENR) in the past 4 years.

We see a great deal of difference between waves depending on whether the general population sample was asked the question (wave 11) or only the extra 5 min sample (waves 1, 5, 7 and 9). The latter, which consists primarily of ethnic minorities, had higher rates of experienced ethnic or religious harassment (around 6 per cent compared to around 2 per cent in the general population sample). There was also a substantial difference between different religious groups, with Muslims being much more likely to both experience (consistently above 8 per cent) and anticipate (consistently above 18 per cent) ENR harassment. Figure 1 shows the rates of ENR harassment for different religious groups in waves 7 and 11 respectively. Under half of respondents who experienced ENR harassment, attributed this to their religion, but the figure was higher in waves 7 and 9 (45 per cent) than in wave 11 (38 per cent), and was higher for anticipated harassment.

Figure 1. Percentages of experienced and anticipated ENR harassment by wave and religious affiliation/Wave 7 N = 8,104 (Extra 5 min sample), Wave 11 N = 26,769 (General population sample).

Religion variables

Questions about religious affiliation, attendance and importance in life were asked in waves 1, 4, 8 and 12.

Regarding affiliation, participants were asked whether or not they belong to a religion, and if answering yes, they were asked which one from a list of 15 religious denominations and traditions. Most of these groups are very small, and for the purposes of the analysis these were reduced to four categories: None, Christian, Muslim and Other. In additional analysis of the waves 11 and 12 data, we divided the Christian category into Anglican, Catholic and Other Christian (see Appendix 4).

Participants were also asked how often they attend religious services, where answers were recoded into a four-point scale from (1) Practically never to (4) Once a week or more.³

A third question was about religion’s importance in life: “How much difference would you say religious beliefs make to your life? Would you say they make a great difference, some difference, a little difference or no difference?” The variable is recoded such that higher values mean greater difference.

Wellbeing/distress

Wellbeing was measured by the total 12-item General Health Questionnaire (GHQ-12) score which was included in every wave of the UKHLS. This is a measure of anxiety and depression and varies from 0 to 36, with a higher score meaning poorer mental health. Scores of 10–12 are considered typical. Each question is scored on a Likert scale and was asked in the self-completion part of the questionnaire to minimise social desirability bias. The GHQ has been widely used and tested for validity and reliability (Griffith and Jones 2019), including for inter-ethnic comparisons (Bowe 2017), and can be used to measure both positive and negative mental health (Hu et al. 2007).

Sociodemographic variables

We also included sociodemographic variables in all the models which might account for some of the variance in harassment experience, religiosity and wellbeing, including a dichotomous variable for participants’ sex (Female), a continuous variable for age, educational qualification ranging from (1) No qualification, to (6) Degree⁴ (treated as a continuous variable), a dichotomous variable indicating whether the respondent immigrated to the UK after age 16, and a categorical variable for ethnicity distinguishing between White, Asian, Black and Mixed/Other.⁵ The age variable also doubles as a measure of time in the longitudinal models. Table 1 presents means and standard deviations for all measures, as well as correlations between them.

Table 1. Descriptives and correlations for all variables.

Variable	N Obs	N Ind	Mean	SD	Correlations
					1	2	3	4	5	6	7	8	9	10	11	12	13	14	15	16	17	18
1) ENR exp harassment	24 805	4 134	0.07	0.25
2) ENR ant harassment	24 805	4 134	0.13	0.33	0.54
3) Rel exp harassment	24 805	4 134	0.03	0.16	0.60	0.36
4) Rel ant harassment	24 805	4 134	0.06	0.24	0.30	0.66	0.56
5) No religion	47 560	3 963	0.22	0.41	−0.03	−0.09	−0.07	−0.12
6) Christian	47 560	3 963	0.28	0.45	−0.02	−0.06	−0.08	−0.12	−0.33
7) Muslim	47 560	3 963	0.35	0.48	0.05	0.13	0.15	0.24	−0.39	−0.46
8) Other religion	47 560	3 963	0.15	0.36	0.00	0.01	−0.02	−0.03	−0.22	−0.26	−0.31
9) Religious attendance	17 366	4 342	2.41	1.29	0.04	0.06	0.07	0.08	−0.47	0.16	0.20	0.09
10) Religious difference	17 343	4 336	3.03	1.16	0.05	0.09	0.09	0.11	−0.59	0.11	0.39	0.04	0.56
11) GHQ-12	38 440	3 203	11.24	5.92	0.11	0.12	0.09	0.08	0.02	−0.04	0.03	−0.02	−0.10	−0.03
12) Sex	50 850	4 238	1.55	0.50	−0.01	0.02	0.00	0.02	−0.06	0.09	−0.01	−0.02	−0.11	0.08	0.08
13) Age	50 855	4 238	42.55	16.78	−0.08	−0.10	−0.08	−0.10	0.00	0.20	−0.21	0.03	0.00	−0.02	0.02	−0.02
14) Edu qualification	50 650	4 221	3.94	1.76	0.06	0.05	0.02	0.00	0.08	0.06	−0.17	0.06	−0.01	−0.07	−0.07	−0.05	−0.20
15) Immigrant	50 855	4 238	0.41	0.49	0.00	0.02	−0.03	−0.02	−0.22	−0.01	0.10	0.13	0.24	0.22	−0.05	0.00	0.28	−0.10
16) White	50 755	4 230	0.17	0.38	−0.08	−0.13	−0.06	−0.10	0.32	0.15	−0.29	−0.17	−0.30	−0.38	0.00	0.00	0.21	0.01	−0.29
17) Asian	50 755	4 230	0.51	0.50	0.04	0.13	0.09	0.16	−0.30	−0.51	0.51	0.30	0.18	0.28	0.02	−0.05	−0.15	−0.06	0.19	−0.47
18) Black	50 755	4 230	0.21	0.41	0.01	−0.03	−0.05	−0.07	−0.04	0.44	−0.27	−0.16	0.13	0.09	−0.05	0.06	0.04	0.02	0.07	−0.24	−0.54
19) Other	50 755	4 230	0.10	0.30	0.02	−0.01	−0.01	−0.04	0.14	0.05	−0.11	−0.08	−0.10	−0.11	0.04	0.01	−0.06	0.05	−0.04	−0.15	−0.35	−0.18

Extra 5-min sample: N Obs = responses to the variable across all waves and individual respondents. N Ind = number of respondents who answered the question, averaged across the waves where the question was asked. All correlations larger than 0.01 have P values smaller than 0.05.

Models

The analysis is primarily conducted by random effects regression models, with lagged variables to account for the longitudinal structure of the data. All analysis was weighted to account for sampling effects (University of Essex, Institute for Social and Economic Research 2022b, 48–51).⁶

There was not enough variation over time on the religiosity variables to run either fixed or random effect longitudinal models with these as dependent variables. Instead, the lagged response models are fitted by separate regression analysis of each wave where these questions were asked. We focus on wave 8, which took place in the year directly following questions about harassment (wave 7) to examine the relationship between experienced harassment in the previous year and current religiosity when controlling for lagged religiosity.

Findings

Effects on distress

To test the first hypothesis, analysis was conducted with distress (GHQ-12) as the dependent variable. Although a wealth of previous research showed a positive relationship between ethnic harassment and subsequent distress and reduction in wellbeing, we started by testing that this assumed relationship was evident in this dataset.

The analysis was conducted on the extra 5-min sample by a maximum likelihood random-effects regression estimator, and shown in Tables 2 and 3. The null model had an intraclass correlation of 47.7 indicating that about half of the variance in the GHQ-12 is due to between-individual differences, and half due to within-individual differences. The results can be seen in Table 2.

Table 2. Random effects regression (MLE) model predicting Distress (GHQ-12).

	Model 1		Model 2		Model 3		Model 4		Model 5
	Coeff.	SE	Coeff.	SE	Coeff.	SE	Coeff.	SE	Coeff	SE
Lag Experienced ENR harassment	1.039**	0.375	0.972*	0.384	0.968*	0.391	0.912	0.847	1.086**	0.393
Lag Anticipated ENR harassment	0.112	0.300	0.111	0.310	0.049	0.312	0.029	0.312	1.647**	0.544
Female	0.368	0.255	0.281	0.258	0.170	0.195	0.165	0.195	0.165	0.194
Age	0.019*	0.007	0.017*	0.007	0.007	0.006	0.007	0.006	0.006	0.006
Educational qualification	0.143*	0.069	0.118	0.070	0.024	0.056	0.023	0.056	0.018	0.055
Immigrant (after age 16)	−0.782*	0.365	−0.845*	0.374	−0.763*	0.310	−0.730*	0.310	−0.741*	0.309
Ethnic group (Ref: White)
Asian	2.274**	0.540	2.606**	0.570	2.072**	0.470	2.046**	0.472	2.023**	0.469
Black	1.354*	0.554	1.917**	0.593	1.547**	0.458	1.569**	0.458	1.614**	0.456
Mixed/Other	2.736**	0.608	2.572**	0.626	1.751**	0.481	1.745**	0.481	1.709**	0.478
Religious affiliation (Ref: None)
Christian	−0.530**	0.166	−0.398*	0.172	−0.493**	0.163	−0.487**	0.163	−0.393*	0.164
Muslim	−0.291	0.544	−0.228	0.595	−0.313	0.514	−0.439	0.518	−0.360	0.524
Other	−0.623	0.403	−0.535	0.416	−0.699	0.391	−0.697	0.399	−0.667	0.401
Lag religious attendance			−0.252**	0.069	−0.175*	0.070	−0.180*	0.070	−0.178*	0.070
Lag religion makes difference to life			0.024	0.084	−0.006	0.083	0.000	0.083	−0.012	0.083
Lag GHQ-12					0.262**	0.013	0.263**	0.013	0.266**	0.013
Christian* Experienced harassment							−0.588	0.970
Muslim*Experienced harassment							1.143	1.045
Other*Experienced harassment							0.062	1.317
Christian* Anticipated harassment									−3.218**	0.696
Muslim*Anticipated harassment									−1.239	0.729
Other*Anticipated harassment									−1.706	0.943
Constant	8.779**	0.689	9.449**	0.710	7.766**	0.571	7.775**	0.571	7.792**	0.568
/sigma_u	3.826	0.097	3.768	0.097	2.676	0.098	2.673	0.098	2.657	0.098
/sigma_e	4.084	0.030	4.097	0.031	4.099	0.033	4.099	0.033	4.096	0.033
Rho	0.467	0.013	0.458	0.014	0.299	0.017	0.298	0.017	0.296	0.017
N obs	8,831		8,579		7,918		7,918		7,918
N individuals	994		918		913		913		913
LR chi2	67.62		76.91		521.07		525.44		543.49
Prob > chi2	0.000		0.000		0.000		0.000		0.000
Log likelihood	−29,665.7		−28,886.9		−27,121.3		−27,119.1		−27,110.1

**P < 0.01, *P < 0.05. Average no of observations per individual = 8.5 (min 1, max = 11).

Table 3. Random effects regression (MLE) model predicting Distress (GHQ-12) among Christians and Muslims separately.

	Christians						Muslims
	Model 1		Model 2		Model 3		Model 1		Model 2		Model 3
	Coeff.	SE	Coeff.	SE	Coeff.	SE	Coeff.	SE	Coeff.	SE	Coeff.	SE
Lag Experienced ENR harassment	0.343	0.456	−0.173	0.464	−2.925*	1.443	2.286**	0.735	2.236**	0.753	−2.358	2.436
Female	−0.714*	0.330	−0.557*	0.225	−0.553*	0.225	2.273	1.233	1.411	0.882	1.331	0.888
Age	−0.032**	0.010	−0.026**	0.008	−0.026**	0.008	0.094*	0.045	0.080*	0.035	0.078*	0.035
Educational qualification	0.020	0.099	0.009	0.069	0.012	0.069	−0.334	0.316	−0.260	0.255	−0.257	0.256
Immigrant (after age 16)	−0.253	0.513	−0.569	0.380	−0.613	0.380	−1.919	1.044	−1.742*	0.849	−1.744*	0.851
Ethnic group (Ref: White)
Asian	1.021	1.563	0.972	1.102	0.985	1.102	4.258	2.600	3.586	1.885	3.755	1.894
Black	1.554*	0.681	1.187*	0.477	1.205*	0.477	4.449	4.123	3.758	2.786	3.968	2.805
Mixed/Other	3.113**	0.959	2.315**	0.660	2.322**	0.659	−0.267	3.514	0.156	2.410	0.374	2.428
Lag religious attendance	−0.139	0.074	−0.072	0.071	−0.080	0.071	−0.713**	0.240	−0.608*	0.248	−0.763**	0.259
Lag GHQ-12			0.330**	0.020	0.331**	0.020			0.381**	0.065	0.375**	0.064
Lag religious attendance* Exp harassment					0.895*	0.444					1.419*	0.716
Constant	13.609**	1.030	9.495**	0.795	9.511**	0.795	5.739	4.665	2.895	3.594	3.488	3.620
/sigma_u	3.502	0.123	2.138	0.128	2.136	0.127	4.201	0.409	2.361	0.473	2.396	0.472
/sigma_e	3.605	0.040	3.680	0.044	3.679	0.044	4.781	0.160	4.881	0.184	4.852	0.183
Rho	0.486	0.019	0.252	0.024	0.252	0.024	0.436	0.052	0.190	0.067	0.196	0.067
N obs	2,923		2,747		2,747		1,970		1,725		1,725
N individuals	347		346		346		247		242		242
LR chi2	40.16		363.85		367.91		29.43		70.99		74.9
Prob > chi2	0.000		0.000		0.000		0.001		0.000		0.000
Log likelihood	−12,975.3		−1,2295.5		−1,2293.5		−1,630.1		−1,470.0		−1,468.1

**P < 0.01, *P < 0.05. Average no of observations per individual = 8.5 (min 1, max = 11).

Model 1 shows that experienced ENR harassment within the previous 4 years is significantly associated with higher levels of distress (Coeff = 1.039, SE = 0.375), even after controlling for sex, age (year), immigrant, ethnicity, education and prior religious affiliation. In other words, people who experienced or anticipated ethnic or religious harassment subsequently score about 1 point higher on the GHQ-12 scale (predictive margin 12.030, SE = .388) than people who did not (predictive margin 10.991, SE = .127). For experiences of religious harassment specifically (see Appendix Table A2), the effect was even larger (Coeff 2.553, SE = .682) – a 2. 5 point higher score on the GHQ scale, all else equal.

Table 2 also shows that belonging to a non-white ethnic group, is a strong and clearly significant sociodemographic predictor of distress and that descendants are more likely than immigrants to experience distress. Gender, age and education have smaller effects and are insignificant in later models.

The impact of religiosity on distress is also significant. Belonging to a Christian religion was associated with reduced distress (Coeff. = −.530, SE = .166) in model 1. In model 2, lagged religious attendance and importance of religion were introduced as further control variables, and here we see that the effect of affiliation is reduced slightly (Coeff. = −.398, SE = .172), as some of the variance is explained by prior religious attendance, which also appears to reduce distress (Coeff = −.252, SE = .069). Average predicted scores from the model show that people who attended religious services once a week subsequently score about .9 points lower on the GHQ scale – i.e. less distressed (all else considered) than those who said they never attended in the previous survey. This effect is significant and consistent, and nearly as large as the effect of experiencing harassment. This supports the hypothesis that religious affiliation and attendance would increase wellbeing and protect against distress. However, it is notable that how much difference religion makes in the respondent’s life has no significant effect.⁷

In model 3, we test whether the effects would still hold when including previous levels of distress as a control variable, by introducing lagged GHQ-12 into the model. Since the GHQ-12 includes symptoms of mental illness that may be long lasting it is reasonable to assume that prior distress would be an important predictor of subsequent distress, and it does indeed have a significant positive coefficient (Coeff = .260, SE = .013). Nevertheless, the effects of ethnic harassment and religiosity are overall rather similar in model 3, although affiliation appears more effective, and attendance less effective once previous GHQ-12 scores are taken into account.⁸

Finally, in models 4 and 5, to test whether religiosity protects against the negative mental health effects of ethnic harassment experiences specifically, we included interaction terms of religious affiliation and ethnic harassment. The interaction terms with experienced harassment (model 4) were not significant and did not improve the model. However, in model 5, we find a negative interaction effect (Coeff = −3.218, SE = .696), meaning that belonging to a Christian denomination reduced distress more for those who had anticipated ethnic harassment compared to those who had not. Anticipated harassment also has a significant direct effect on distress in this model, meaning it increases distress for the nonreligious, while it decreases distress for Christians. The results are shown as predicted scores in Figure 2. In other words, we do find some evidence for a buffering effect. Table A2 in the supplementary material shows the same analysis for religious harassment specifically, and the results are largely similar, but here the interactions were not significant, possibly due to the smaller numbers who experience religious harassment.

Figure 2. Predicted scores on GHQ-12 by prior ENR harassment and religious affiliation/Waves 1–12, 5extra minutes sample (N obs  = 7,918, N ind 913).

Together the findings from this analysis indicate that people who experience or anticipate harassment due to their ethnic, national or religious identity have a subsequent reduction in wellbeing, but this is not as steep if they belong to a religion, which broadly supports hypothesis 1. However, it appears that this protective effect of identity is specific to Christian denominations, and while we do observe a general effect of religious service attendance in reducing distress, there is no significant effect of the importance attached to religious beliefs (difference to life).

Differences between ethnic and religious groups

To further examine the impact of experienced harassment on majority and minority religious adherents, and to answer whether the protective effect of religion is specific to Christians, we conducted similar analysis on Christians and Muslims separately.

In Table 3, we examine the impact of experienced ENR harassment and religious service attendance on distress (GHQ-12) in model 1, control for previous GHQ-12 in model 2, and add an interaction term between ENR harassment and religious service attendance in model 3.

In models 1 and 2, there is a direct effect of ENR harassment on GHQ-12 among Muslims, but not among Christians. Similarly, the protective effect of religious attendance which was observed in Table 2, is only significant for Muslims. In model 3, the interaction effects (see Figure 3), show that ENR harassment increases distress among Muslims with higher levels of religious attendance. For Muslims, then, religious attendance reduces distress overall but also increases vulnerability to distress from ENR harassment.

Figure 3. Predicted scores on GHQ-12 by prior experienced ENR harassment, religious service attendance and religious affiliation/Based on waves 1–12, 5 extra minutes sample. Christians (N obs  = 2,747, N ind 346), Muslims (N obs  = 1,725, N ind 242).

It is possible that some of the variation between religious groups is in part an effect of ethnicity, because Christians are more likely to be White or Black, whereas Muslims are more likely to be Asian or Black African. Although the models in Tables 2 and 3 control for ethnicity, additional analysis was conducted on each ethnic group separately (see Appendix 3, Table A3), and this analysis shows that harassment as a predictor of distress is only significant for the Asian group, but unlike the White and Mixed/Other ethnic groups, they are not significantly protected against distress by religious affiliation or attendance. In other words, this analysis does not support a general “buffering effect” against ethnic harassment. Instead, religion has a general protective effect against distress for those who are least likely to experience ethnic harassment, namely those with a majority ethnic or religious identity.

To further examine the impact of religion on the majority ethnic group, and to distinguish between different Christian denominations, additional analysis was done on waves 11 and 12, where the general population sample (and not only the extra 5 min sample) were asked the questions about harassment (see Appendix 4, Table A4). In this model, Catholics and other Christians experience less distress from both experienced or anticipated harassment compared to the nonreligious, but this is not the case for Anglicans. Predicted scores are shown in Figures A1 and A2.

In the analysis of the whole extra 5 min sample (Table 2) we found initial support for hypothesis 2, that the effect of religious identification on subsequent psychological distress was more protective for Christian, than minority religious denominations. This was further nuanced when analysing this separately by religion and ethnicity. For Muslims, religious attendance reduces distress overall but increases vulnerability to distress from ENR harassment.

That said, the distinction between Christian denominations in the general population sample indicates that this is not the same for all religious minorities, and Catholics and other non-Anglican Christians, seem to benefit more from religious identity than Anglicans do. This could indicate that the impact of religious “identity centrality” (Wolfram, Linton, and McDuff 2018) and minority status varies between religious and ethnic groups. Christian minorities may benefit from having a distinctive group identity in relation to the majority, without the low status and risk of discrimination and harassment that Muslims experience.

Effects on religiosity

To address whether experienced harassment increase subsequent religiosity, lagged response models were fitted as OLS regressions in wave 8 (2016–2017) where questions about religiosity were asked in the year directly following a wave with questions about harassment (wave 7).⁹ The models were restricted to the extra 5-min sample, and control for prior religious affiliation. The dependent variables were the difference religious beliefs make to life, and religious service attendance, and these were also included as lagged variables. The results can be seen in Table 4.

Table 4. OLS regression with lagged responses predicting religiosity (Wave 8).

	Religious difference to life				Religious service attendance
	Model 1		Model 2		Model 1		Model 2
	Coeff.	SE	Coeff.	SE	Coeff.	SE	Coeff.	SE
Lag Experienced ENR harassment	0.520**	0.189	−0.075	0.238	1.023**	0.243	−0.083	0.234
Female	0.010	0.082	0.012	0.082	0.009	0.091	0.013	0.091
Age	0.005	0.003	0.006*	0.003	0.001	0.003	0.001	0.003
Educational qualification	−0.035	0.025	−0.033	0.025	0.071**	0.025	0.074**	0.026
Immigrant (after age 16)	−0.105	0.168	−0.122	0.171	−0.244	0.188	−0.280	0.186
Ethnic group (Ref: White)
Asian	0.223	0.176	0.272	0.185	0.200	0.213	0.296	0.221
Black	0.387*	0.171	0.404*	0.174	0.553**	0.187	0.586**	0.189
Mixed/Other	−0.212	0.213	−0.196	0.216	−0.263	0.195	−0.233	0.199
Religious affiliation (Ref: None)
Christian	0.347**	0.129	0.343**	0.129	0.275**	0.103	0.266**	0.102
Muslim	0.558**	0.158	0.565**	0.161	0.515*	0.201	0.545**	0.203
Other	0.480*	0.201	0.471*	0.200	0.644*	0.258	0.627*	0.260
Lag religion makes difference to life	0.646**	0.061	0.643**	0.061
Lag religious attendance					0.452**	0.063	0.445**	0.063
Christian* ENR harassment			0.750*	0.348			1.442**	0.347
Muslim*ENR harassment			0.181	0.269			0.096	0.309
Other*ENR harassment			−0.123	0.328			−0.161	0.476
Constant	0.428	0.228	0.413	0.231	0.228	0.232	0.197	0.233
N obs	1,537		1,537		1,546		1,546
F	138.53		116.27		64.19		67.66
P	0.000		0.000		0.000		0.000
R squared	0.660		0.661		0.435		0.441

**P < 0.01, *P < 0.05.

Experienced ENR harassment in the previous year increases both religious service attendance and how much religion makes a difference to the respondent's life.¹⁰ The effect is particularly large for attendance. To illustrate the size of the effect: someone who was not harassed for their ethnicity, nationality or religion attended on average “practically never” or only for special occasions, whereas someone who was ENR harassed the year before attended on average more than “several times a year”, and close to “once a month”. The results for anticipated harassment were similar, but with smaller effects.

In model 2, we include interaction terms between harassment and religious affiliation and find that it is primarily Christians who are likely to increase their attendance and identification following experiences of harassment. This is consistent with the findings that Christians in this sample are more likely than other religious groups to experience a protective effect of their religious affiliation (although the evidence for a protective effect of attendance for Christians was mixed – see Table 3 and A4).

In sum, there is some evidence for a relationship between ethnic or religious harassment and subsequent increase in religiosity, although it seems to be applicable mainly to Christians, and the exact form it takes varies depending on the year and the sample studied. Overall, participants who anticipated or experienced ethnic harassment subsequently increased their estimation of how much difference their religion makes to their life, indicating increased religious group identification. We also see that religious service attendance increased following harassment experiences. Hypothesis 3, Experiencing ethnic or religious harassment increases subsequent religious identification and practice, is thus partially supported.

Effects on harassment

Although harassment appears to increase subsequent religiosity, this does not preclude the reverse association. People who identify strongly with their ethnicity or faith may also be more visible targets for perpetrators of ethnic, racial or religious harassment. For example, they may practice their religion more fervently or frequently, or give more visible signals of their ethnic or religious belonging, for example by wearing traditional religious dress, displaying religious symbols or avoiding certain foods (Ramos et al. 2012, 654).

In addition, increased ethnic or religious identification could also increase the likelihood that instances of harassment or discrimination are interpreted or experienced as being due to one’s religious, ethnic or racial background. In self-reports from the victims, we can only infer the perpetrator’s motivation as perceived by the victim, and as such there is a strong probability of bias.

To understand whether religiosity increases subsequent harassment experience, we conducted analysis on the extra 5-min sample for all waves using population-averaged logit models (GEE) with lagged responses. There are four models, distinguishing between experienced and anticipated harassment attributed to either the respondent’s religion, or the respondent’s ethnicity, nationality or religion. The results are shown in Table 5.

Table 5. GEE population averaged models with lagged responses.

	Experienced Rel harassment		Anticipated Rel harassment		Experienced ENR harassment		Anticipated ENR harassment
	Coeff.	SE	Coeff.	SE	Coeff.	SE	Coeff.	SE
Lagged rel affiliation (Ref: None)
Christian	2.141	1.253	3.161**	0.828	0.878	0.745	0.588	0.851
Muslim	3.992**	1.234	4.721**	0.893	0.746	0.642	0.740	0.690
Other	2.019	1.190	3.169**	0.914	−0.778	0.624	−0.663	0.614
Lagged rel difference to life	0.482	0.346	−0.010	0.225	0.191	0.207	−0.043	0.201
Lagged rel attendance	0.124	0.176	0.607**	0.132	0.026	0.187	0.227	0.196
Female	0.980*	0.495	1.087**	0.322	−0.065	0.396	0.158	0.336
Age	−0.036*	0.018	−0.034**	0.011	−0.044**	0.011	−0.034**	0.011
Educational qualification	0.329*	0.158	0.155	0.097	0.189	0.102	0.100	0.092
Immigrant	0.075	0.467	−0.486	0.299	1.099*	0.447	−0.154	0.322
Ethnic group (Ref: White)
Asian	3.242*	1.617	1.898*	0.959	1.133	0.701	1.773**	0.425
Black	2.696	1.537	1.720	1.066	0.253	0.784	0.949	0.495
Mixed/Other	2.100	1.553	−0.310	1.003	0.365	0.664	0.284	0.495
Lagged harassment	1.530*	0.751	1.772**	0.346	1.701**	0.474	0.694	0.435
Constant	−13.629**	2.443	−11.427**	1.156	−4.486**	0.993	−3.639**	1.073
N obs	4,202		4,202		4,202		4,202
N individuals	916		916		916		916
Wald chi2(10)	190.63		351.01		327.05		166.09
Prob > chi2	0.000		0.000		0.000		0.000

**P < 0.01, *P < 0.05. Average no of observations per individual = 4.5 (min 1, max = 5).

As the first model shows, we find a significant reverse effect where previous religious affiliation increases subsequent experiences of harassment on religious grounds even after controlling for previous religious harassment. Specifically, Muslims seem more at risk of religious harassment than the rest of the population, as are Asians. Other religious affiliations, and religious attendance, only increases anticipated, and not experienced religious harassment. As the other models show, religiosity does not increase ENR harassment, or indeed the overall risk of experiencing or anticipating harassment. This analysis thus provides very limited support for Hypothesis 3: Religious identification and practice increases the likelihood of subsequent ethnic or religious harassment experiences.

The findings are an indication that being Muslim not only increases the negative effects of harassment but might also increase the risk of having such experiences in the first place. In addition, the results for anticipated harassment indicate that religious people are more likely to interpret the harassment and harassment risks that they are exposed to, as being due to their religious identity.

Discussion

This study aimed to examine the relationship between experienced ethnic harassment and religious identity and behaviour among ethnic minorities in the UK, using longitudinal panel data. Our descriptive analysis finds that harassment experiences based on ethnicity, nationality or religion (ENR) are vastly more common among religious minorities, and Muslims in particular, than they are for non-religious and Christian respondents, even when these belong to an ethnic minority group (see Figure 1).

By analysing the data longitudinally, we find that experienced and anticipated harassment based on ENR harassment significantly increases distress in subsequent years. However, religiosity, and particularly religious attendance, appears to reduce subsequent distress. These findings are in line with previous research on religion and mental health, which tend to find small, but significant effects of religiosity and religious service attendance on wellbeing and mental health (see Garssen, Visser, and Pool 2021).

Secondly, while we observe a small protective effect of religious affiliation against psychological distress, this does not mean it is protective against the effects of harassment specifically. Indeed, for some minorities, religious identity and practice both seem to increase, rather than protect against distress, following experiences of harassment. This is contrary to the rejection-identification model (Branscombe, Schmitt, and Harvey 1999), but consistent with Wolfram, Linton, and McDuff (2018) and Ysseldyk, Matheson, and Anisman (2011) and, who argue that that strong group identification could intensify the negative effect of harassment, as they make such experiences appear more personally relevant. Thus, for Muslims in the UK, the centrality of their religious identity, combined with a sense of threat from the majority, could undermine the otherwise beneficial effects of religiosity and group identification.

A third finding of this study is that people who have experienced ENR harassment, are more likely to increase their religious identification and practice afterwards. There was evidence of significantly increased religiosity for this group in the waves where we have available measures of both religiosity and prior harassment. However, closer examination reveals that it is primarily Christians who react to harassment by increasing their Christianity, not Muslims and other religious minorities. The analysis offers some support for the rejection-identification model (Branscombe, Schmitt, and Harvey 1999) of increased minority group identity in reaction to rejection by the majority, and it also supports other literature which suggests that persistent religious involvement among ethnic minority populations could represent a form of “reactive religiosity” in response to hostility from the majority population (Connor 2010; Platt 2014; Voas and Fleischmann 2012). However, it does not seem to account for the persistent levels of Muslim identification.

Finally, we also observe a reverse effect, where religious attendance and affiliation appear to increase the risk of subsequent experiences of harassment on grounds of religiosity. Specifically, Muslims are at higher risk of religious harassment. However, it is notable that religious involvement does not increase subsequent experiences of harassment overall, only those attributed to religious identity. It is thus difficult to ascertain to what extent increased religious involvement increases the risk of harassment per se, and to what extent it shifts the attribution of harassment that would have taken place in any event, to religious identity or practice, rather than for example race, ethnicity, gender or language. Nevertheless, the finding that religiosity increased after ENR harassment experiences, while ENR harassment experiences did not generally increase after increases in religious behaviour, suggests that the relationship between ethnic harassment and religiosity is primarily a result of reactive religiosity and that the reverse causation cautioned against by Platt (2014, 68) and others, does not account for much of the relationship, at least not in the UK.

In summary, the longitudinal analysis of UKHLS data provides evidence for increased religious identity and practice as a reaction to harassment experiences, but not as uniformly protective against negative health implications of harassment, discrimination and racism.

Limitations

There are some significant limitations to note when evaluating the results of this research. Firstly, the survey only asks about harassment in the last 12 months, and the question is only asked in some waves. As a consequence, there may be prior and intervening instances of harassment that the respondents experienced, but that we are unable to account for. Similarly, as the questions about religion are also asked only in some of the waves, there may be short-term increases and decreases in religious identification and behaviour in some respondents, that are not recorded in the data. In this sense, the findings presented here are conservative estimates. Some interesting research questions that could be examined more thoroughly with more fine-grained longitudinal or experimental data would be whether there are any short-term spikes in religious identification or behaviour after individual experiences of harassment, and for how long we can expect the effects of harassment to last.

A second limitation in the data is that by asking where the harassment took place, the questions about harassment invite respondents to focus on harassment in physical places. Although there is an “other places” category, it seems likely that the question wording would lead to an undercounting of online harassment experiences. An interesting question for future research would be whether ethnic and religious harassment that takes place on social media or over the phone has similar effects on religious behaviour and identification.

Finally, this research suffers from the same limitation as much other research on minority groups, namely the difficulty of disentangling the effects of religion, ethnicity and other group identities. There is a great deal of overlap between racial, national and religious minorities in the UK. Moreover, as pointed out in the discussion of hypothesis 4, experiences of harassment can sometimes be both racially and (anti-)religiously motivated, and how it is experienced may depend on which identity is most salient to the respondent. Similarly, the effect on religiosity will depend on the salience of religiosity in the respondent’s upbringing, as well as what other identities and communities are available to them. While controlling for prior religiosity, as we did, will capture some of this variation, research in specific communities could examine more thoroughly to what extent, and under what circumstances the reactive religiosity that we observe, is an addition or alternative to “reactive ethnicity” (Portes and Rumbaut 2001) observed in other studies.

Conclusion

Analysing panel data from the UK, we find evidence that experienced and anticipated harassment based on ethnicity, nationality or religion, significantly reduces subsequent wellbeing. We also find evidence of reactive religiosity, as people who experience ethnic, national or religious harassment are more likely to increase their subsequent religious identity and practice. However, we do not find that increased religiosity protects them against the negative mental health effects of harassment. On the contrary, religiosity seems to increase vulnerability both in terms of increased risk of experiencing and anticipating religiously motivated harassment, and increased distress following harassment experiences.

By examining religion as a protective mechanism against insecurity in the specific situation of ethnic minorities experiencing racial harassment, this research adds to the large body of research on reactive ethnicity, religion and mental health, as well as the insecurity hypothesis in the sociology of religion. The longitudinal data allows us to expand on previously conducted studies, and to clarify causality in the relationship between harassment and religiosity.

Crucially, it is Christians, who are most likely to experience the protective effects of their religious affiliation, and to increase their religiosity following harassment experiences. This is a population which, although many are also ethnic and religious minorities, belongs to a group privileged and protected by the majority. On the other hand, while we see some positive effects of identifying as Muslim, the more they participate religiously, the more distressed they are by harassment experiences. While the evidence for the “reactive religiosity” effect is mixed, what this does suggest is “religious reactivity”, that is, a stronger psychological reaction among religiously active and marginalised minorities to harassment on the grounds of ethnicity, nationality or religion. The findings also support earlier research that suggests differential effects of identity centrality in responses to harassment among ethnic minorities and majorities.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

1 An exception is Fleischmann, Leszczensky, and Pink (2019), who investigate the impact of discrimination on religious identity among young people in Germany, but their study does not include measures of religious behaviour.

2 The category “at home” was excluded, as our interest was primarily in the effects of public harassment.

3 The original variable had five categories: Once a week or more, At least once a month, At least once a year, Never or practically never, and Only at weddings, funerals etc. The last two categories were combined as it was difficult to ascertain which of these represented the higher frequency.

4 The full values were: 1 “No qualification” 2 “Other qualification” 3 “GCSE etc.” 4 “A level etc.” 5 “Other higher” 6 “Degree”.

5 We considered more detailed ethnicity variables, but these were too correlated with the religious affiliation variables to allow both to be included in the same model.

6 For most of the analysis, we used the extra 5-min weight, and restricted the analysis to the extra 5-min sample. For the additional separate analysis of the data from waves 11 and 12 (Appendix 4), the individual response weight for wave 12 was used.

7 Due to concerns about multicollinearity, alternative models introduced religiosity variables one by one. However, the results were largely similar, and thus the combined model is presented here for parsimony. Due to concerns about missing values, alternative analysis including only the waves directly preceded by questions about religion (2, 5 and 9) were also conducted, with very similar results.

8 Rabe-Hesketh and Skrondal (2012, 273) caution against using lagged responses in random intercept models as they are correlated with the random intercept, and thus risk producing inconsistent estimates. In this instance, a satisfactory solution could not be found that would allow the appropriate weights to be used. Hence, the model is shown with and without the lagged response included. A fixed effects model (in supplementary material) also produces similar results.

9 There was not enough variation over time on the religiosity variables to run longitudinal models with this limited data, while using appropriate sampling weights.

10 Similar analysis (available on request) was conducted using data from waves 1, 4 and 12. The results are in the same direction. We show results from Wave 8, as in wave 1 both questions were asked in the same wave; in wave 4, measures for harassment are only available from 3 years before (Wave 1), and data for Wave 12 was collected partly during the COVID-19 pandemic and government-imposed lockdown.