Intergenerational and interethnic mental health: An analysis for the United Kingdom

This paper uses a nationally representative data set to examine the extent to which family migration history helps explains interethnic variations in mental health in the United Kingdom. We confirm that there is significant variation in mental health across ethnic group and generation of migration. Furthermore, we show how these dimensions interact. The analysis explores the extent to which neighbourhood, personal characteristics, and migration experience are related to mental health. We find evidence that all are important. Our results are consistent with a dynamic view of migration and settlement whereby individuals' circumstances and how they might contribute to mental health change over time and across generations.


| INTRODUCTION
Poor mental health is a widespread problem. At least one third of all families in England include someone who is currently mentally ill (The Centre for Economic Performance's Mental Health Policy Group, 2012). In addition to personal costs, poor mental health has a negative impact on public finances and on the economy (Layard, 2013).
A large literature has grown to examine various determinants of mental health, focusing on economic, social, and personal influences (Layard, Clark, Cornaglia, Powdthavee, & Vernoit, 2014). Age and income have received particular attention (Gardner & Oswald, 2007), but the increased richness of data has more recently allowed the dynamics of mental health to be considered (Clark, 2014;Clark & Georgellis, 2013), as well as life-cycle (Berner, Cornaglia, & De Neve, 2012), and childhood experience effects (Frijters, Johnston, & Shields, 2014;Layard et al., 2014;Powdthavee, 2012). The conclusion from these studies is that mental health is determined by a combination of adult outcomes, family background, and childhood development.
In recent decades, the U.K. population has been characterised by increasing immigration and, partially as a result of this, has become more ethnically diverse. In view of this, the ethnic and migrant dimensions of mental health are both relevant and intertwined. Both premigration and postmigration experiences have been recognised to play an important role in shaping the mental health of migrants (Arévalo, Tucker, & Falcón, 2015). Understanding the relationship of migration and ethnicity to mental health is important for policy if preventative health strategies are to target population groups most in need. Moreover, because mental health can be associated with severe limitation of economic and social functioning (Johnston, Schurer, & Shields, 2011), being able to intervene effectively has the potential to improve social and economic integration of ethnic groups of different migrant generations.
In this paper, we explore how mental health varies by ethnicity and migrant generation. We use the Understanding Society data that have an ethnic minority booster sample and therefore provide sufficient numbers of observations to allow these dimensions to be considered.
We consider three aspects of mental health, all constructed from the General Health Questionnaire (GHQ). These are anxiety and depression, social dysfunction, and loss of confidence. Another distinctive feature of our analysis is that we distinguish between first-generation migrants, second-generation migrants, and "natives," a shorthand for those born in the United Kingdom and with both parents also born in the United Kingdom. We further distinguish first-generation migrants between "recent" and "established" migrants, according to whether or not they arrived in the United Kingdom within the last 10 years.
Our analysis examines ethnic and migrant variations in mental health. We use regression analysis to assess whether significant ethnic variation exists after controlling for migrant generation and, likewise, whether significant variation by migrant generation exists after controlling for ethnic group. Our results allow us to see the interaction between ethnic and migrant variations. Furthermore, we include additional variables into our regression analysis to examine the extent to which factors relating to migration experience appear to be related to individuals' mental health. We use multilevel regression to allow for spatial clustering (within local authority districts).
Our results document heterogeneity in mental health across ethnic group and migrant generation. Pakistanis stand out as most likely to suffer poor mental health. With regard to variations by migration history, we find that recent migrants experience better mental health, on average, than White natives. The ethnic and migration dimensions interact, resulting in a rich pattern of results. We explore some of the reasons behind this and find that neighbourhood diversity is associated with better mental health for both second-generation minorities and recent minority migrants. For this latter group, living in areas where one's own ethnic group is well represented is also associated with improved mental health. Moreover, the analysis of migration experience shows that the mental health of first-generation migrants declines and converges to that of natives the longer migrants stay in the host country.
Furthermore, although mother tongue and language spoken in childhood does not seem to affect mental health of first-generation migrants, speaking a language other than English in childhood is associated with worse mental health for second-generation migrants. This paper is organised as follows. Section 2 reviews the relevant existing literature. Section 3 describes the data. Section 4 presents some descriptive statistics and regression results. Section 5 concludes.

| EVIDENCE ON HOW MENTAL HEALTH VARIES ACROSS ETHNIC GROUPS AND BY MIGRANT STATUS
Mental health of minority groups can be considered as an indicator of integration and an indicator of the way different ethnic groups assimilate and adjust into the cultural and social life of the largest ethnic group of the U.K. population: White. According to the U.K. census in 2011, Whites represented 87% of the U.K. population.
A recent strand of research has analysed the relationship between migration and health, with a large strand analysing the assimilation of immigrants' health over time, termed the healthy immigrant effect, by focusing primarily on physical health and documenting that immigrants are in better health upon arrival in the hosting country than the natives do, although this health advantage erodes over time (Antecol & Bedard, 2006;Giuntella & Stella, 2017).
The healthy immigrant effect with respect to mental health has instead received less attention. Research for Canada (Lou & Beaujot, 2005) indicates that immigrants' mental health status assimilates to that of the native Canadian population over time; a more recent evidence for Australia (Janisch, 2017) finds that mental health of immigrants deteriorates over time, with that of female immigrants exceeding mental health of natives upon arrival.
Both premigration and postmigration experiences have been recognised to play an important role in shaping the mental health of migrants (Arévalo et al., 2015). Due to the different experiences during the immigration process (Giuntella, Kone, Ruiz, & Vargas-Silva, 2017), the route of entry can explain heterogeneity of health of migrants. Chiswick, Lee, and Miller (2008) show that in Australia, immigrants' self-reported health status varies with visa category, being better among those selected based on their potential for economic success.
In a more recent contribution for the United Kingdom,  looked at reason for migration and found that immigrants who migrated for employment reasons were less likely to report mental health conditions than natives, whereas those who migrated for asylum reasons were more likely to do so. Hatzenbuehler et al. (2017) Stillman, Gibson, McKenzie, and Rohorua (2015) use survey data on successful and unsuccessful applicants to a migration lottery to New Zealand to estimate experimentally the impact of international migration on objective, in terms of incomes and expenditures and subjective well-being. Although international migration improves objective well-being, the effects of migration on subjective well-being are complex, with mental health improving but happiness declining.
Analysing the mental health of Puerto Rican immigrants in the United States, Arévalo et al. (2015) document that the association of neighbourhood ethnic density with depressive symptomatology was significantly modified by sex and level of language acculturation, with men, but not women, experiencing protective effects of ethnic density.
Several studies (see, e.g. Chiswick et al., 2008;Arévalo et al., 2015;Janisch, 2017) have highlighted the importance and role of language proficiency in the process of acculturation since this allows immigrants to navigate their environment effectively to locate social and economic resources and may facilitate adaptation to the host society, reducing adaptation-related stress. Additionally, evidence for the United Kingdom has documented that poor English skills lead immigrants to live in areas with a high concentration of people who speak their same native language (Aoki & Santiago, 2018). As pointed out by Chiswick et al. (2008), knowledge of the language of the destination may be relevant for health status, because it would facilitate communication. Language ability has been emphasised in different studies as one of the main determinants of successful integration (Adserà & Ferrer, 2015;Aoki & Santiago, 2018). Language proficiency is considered a vital component of any migrant's integration process because it facilitates mobility, helps to develop social networks, provides a sense of cohesion, and unlocks access to social connections, enhancing assimilation and integration (McAreavey, 2010). In fact, previous studies (Biddle, Kennedy, & McDonald, 2007) revealed differences in health profiles of immigrants from English-speaking and non-Englishspeaking countries, which were associated with acculturation or environmental effects.
Analysing different aspects of mental health of migrants is crucial for several reasons. Even when they are from the same ethnic background, migrants may differ from natives and from other migrants of different cohorts. Migrants are a subgroup of their original population with characteristics, culture, tradition, and preferences that differ from those of natives and can vary significantly across countries. For example, distance from home, weather changes, and culture shock can all contribute in different ways to shaping the mental health of migrants.
The degree of heterogeneity among migrants may vary with the duration of the migration experience (Simpson, 2013).
Moreover, the integration of minority groups is a complex and long-term process that, across generations, can be hindered or facili- Mothers in minority groups are more likely than White British/Irish mothers to perceive their health as poor and to feel depressed.
Beyond these observed differences, there is the question of why mental health varies. Local area characteristics may be important. In psychiatry, the relationship between mental health and neighbourhood ethnic density has been explored. Under the "ethnic density hypothesis," individuals may have better mental health when living in areas with a higher proportion of people of the same ethnicity (Shaw et al., 2012). Positive ethnic density effects have been found for suicide-related outcomes for Black people in the United Kingdom (Bécares, Nazroo, Albor, Chandola, & Stafford, 2012). Similarly, a study of Black Caribbean people in the United Kingdom shows that increased Black ethnic density was associated with improved health (Bécares, Nazro, Jackson, & Heuvelman, 2012). As suggested by Bécares, Nazro, et al. (2012), ethnic density effects are likely to vary with the reasons for migrating and the length of stay, as well as the socio-economic profiles of ethnic groups and the places where they live.
The aim of this paper is to provide a fuller understanding of how mental health in the United Kingdom varies within migrant generation and ethnic group, by focusing on the differences between and within first-and second-generation migrants. In so doing, we contribute to the existing literature in several ways. First, we analyse three measures of mental health, allowing us to identify which psychological aspect is most affected. Second, we consider how an individual's mental health varies with both the ethnic density of the local population, and what we refer to as "concentration," the degree to which the individual's own ethnic group is represented in the local population.
Third, we jointly consider the role of migration-related characteristics.
As mentioned above, although recent evidence suggests that one of the key aspects of health heterogeneity across migrants is the reason for immigration (Chiswick et al., 2008 and, a key limitation for the current study is that Understanding Society does not provide this information.

| DATA
Understanding Society is a longitudinal survey of households living in the United Kingdom, in which each adult member of the household is interviewed annually. It has been running since 2009 and is a nationally representative sample of around 30,000 households. It is particularly suited to our use because it incorporates a booster sample of approximately 4,000 households where at least one member (or their parents or grandparents) is from an ethnic minority group, with the intention of achieving at least 1,000 adult interviews from Black African, Bangladeshi, Black Caribbean, Indian, and Pakistani ethnic groups.
In line with this and with most of the existing studies (see Dustmann & Theodoropoulosy, 2010), we focus on the six largest ethnic groups defined by the following typology: White, Indian, Pakistani, Bangladeshi, Black Caribbean, and Black African. Mixed and other, representing just below 3.5% of the sample, have also been excluded because they are very heterogeneous groups. Like Longhi (2014) and Knies, Nandi, and Platt (2016), because the measures of diversity are time-invariant, we use Wave 3 only of Understanding Society, with respondents interviewed in 2011 to 2012.
All respondents are asked whether they were born in the United Kingdom and, if not, when they moved to the country. They are also asked about their parents' country of birth. Using this, we categorise each respondent as follows: • recent (first-generation) immigrant-born outside the United Kingdom, parents both born outside the United Kingdom, lived in the United Kingdom for less than 10 years; • established (first-generation) immigrant-born outside the United Kingdom, parents both born outside the United Kingdom, lived in the United Kingdom for 10 years or more; • second-generation immigrant-born in the United Kingdom, parents both born outside the United Kingdom; • native-Whites only, born in the United Kingdom, parents both born in the United Kingdom.
We use a measure of mental health derived from the 12-item GHQ, a self-administered screening test aimed at detecting psychiatric disorders that require clinical attention among respondents in community and nonpsychiatric clinical settings. The GHQ is used to detect disorders of a temporary nature, such as depression or anxiety, but also permanent conditions such as psychotic depression and schizophrenia. The main advantage of the GHQ is that it does not require a subjective assessment by a specialised clinician (Hauck & Rice, 2004)  There are 12 GHQ questions in the Understanding Society. All require a response on a scale ranging from 1 to 4, 1 being the best score. We recode all these indices to range between 0 (least distressed) and 3 (most distressed). We aggregate the 12 GHQ measures into three broader categories: anxiety and depression, social dysfunction, and loss of confidence (see Table A1 for details).
This disaggregation, first adopted by Graetz (1991), is pretty common in existing studies, and it allows identification of the particular dimensions of respondents' psychology that are affected (Dustmann & Fasani, 2015). Each measure is expressed as the average score across the corresponding GHQ measures.
In addition to the measures of mental health, Understanding Society contains rich demographic information. We use the following as control variables in the regression analysis: age, gender, a dummy for working (as employed or self-employed), a dummy for partnership, number of own children in the household (none, 1 child, 2 or more children), and a dummy for living in London. We also include logged household income, equivalised using the modified OECD equivalence scale to take account of household composition.
Moreover, Understanding Society contains variables that capture migration-related characteristics.
We account for various migration-related characteristics that might affect mental health, distinguishing between first-and secondgeneration immigrants. Years since migration provide information on the length of stay in the United Kingdom, and age at arrival in the United Kingdom provides information of the stage in life that an individual arrived in the country.
Following existing literature (Biddle et al., 2007;Chiswick et al., 2008;Janisch, 2017), we control for country of birth in order to capture heterogeneity of migrants' countries of origin. Unfortunately, Understanding Society data only collect detailed information of country of birth for the largest groups in the United Kingdom, with 23% of the first-generation immigrants not reporting the country of birth.
Groups of the country of birth are defined as follows: Europe includes Cyprus, France, Germany, Ireland, Italy, Poland, and Spain; Asia includes Bangladesh, China/Hong Kong, India, Pakistan, and Sri Lanka; Africa includes Ghana, Kenya, Nigeria, South Africa, and Uganda; Caribbean refers to Jamaica. Due to the small sample, we have grouped together United States, Canada, New Zealand, and Australia.
In order to control for the role of English knowledge, we exploit two variables: (a) based on country of birth, we derive a variable for immigrants' mother tongue, specifically deriving a dummy for non-English country immigrants, and (b) we control for language spoken in childhood deriving a dummy for not speaking English in childhood.
Language spoken in childhood is likely to be the first language learned and being determined by parents and is also less likely to be affected by self-reported bias (Janisch, 2017). In addition, individuals exposed to a new language during childhood can learn it more easily than those exposed to it outside of this critical period (Aoki & Santiago, 2018).
We also construct a dummy for having arrived as a child (aged less than 15) and not speaking English in childhood.
To account for migrant history and characteristics of parents, for the second-generation immigrants, we control for whether an individual spoke English in childhood and if either parent arrived from a non-English-speaking country.

| Descriptive statistics
The sample is summarised in Ethnic minorities also tend to live in much more diverse neighbourhoods than Whites. However, this is not driven by specific ethnic groups being concentrated in particular areas. Whereas Whites live in predominantly white areas on average, individuals from other ethnic groups appear to live in areas that, ethnically, are much more mixed.
Table 1b provides information on the migration history of the subsample of first-and second-generation immigrants.
On average, first-generation immigrants have been living in the United Kingdom for 23 years and are 23 years old 2 ; the vast majority of them (79%) come from a non-English-speaking country, and 13% arrived as a child from a non-English-speaking country. The largest first-generation immigrant is from Asia (43%), followed by Europe (14%), and Africa (13%). Only 3% are from Australia, New Zealand, Canada, and United States, and 5% from Caribbean; 42% of the second-generation immigrants did not speak English in childhood, whereas 32% of either parent where from a non-English-speaking country.
Figures 1-3 graphically represent the mean scores for the three measures (anxiety and depression, social dysfunction, and loss of confidence) by ethnicity and by migrant generation. The score varies from 0 to 3. Lines closer to the centre indicate better levels of mental health. However, as can be seen from the charts, the mean levels observed are always closer to zero than they are to their possible maximum.
Looking across Figures 1-3, two points are apparent. First, recent migrants appear to have a better level of mental health than more established and second-generation migrants. This varies by outcome measure and by ethnic group, but as a broad point, it holds true. Second, on average, Pakistanis appear to have a worse mental health compared with the other ethnic groups.

| Regression results
To look deeper into thee descriptive findings, we use regression analysis. Including both ethnic group and migrant generation indicators among the regressors allows us to see whether the dimensions have separate independent associations with mental health. Furthermore,  the specification allows these two dimensions to interact so the possibility that the variation by ethnic group differs across generations can be captured. We allow for random effects of neighbourhoods and follow Bell (2014) by adopting a simple multilevel model: where y i are the scores of the measures of mental health; E ei is an indicator variable taking value 1 when the respondent is a member of ethnic group e (0 otherwise); G gi is an indicator variable taking value 1 when the respondent is categorised as being of migrant generation g (0 otherwise); X i includes individual characteristics, specifically age, age squared, and sex; and u LAD is the Local Authority District random effect. When estimating mental health equations of the type considered here, it is important to recognise the potential for regressors to be endogenous or even dependent on the outcome variable (reverse causality). We are careful to include only exogenous regressors among the X i (age and sex) in order to avoid this source of bias. However, we relax this with our final estimates in order to allow some speculation as to the factors that might contribute to differences in mental health.
Because the dependent variables are coded on a point scale, it is a common practice to estimate Equation (1) Table A3) point to significant variation by ethnic groups for all migrant generation (except for loss of confidence for second-generation immigrants) even after controlling for age and sex differences. Table 2 shows that many recent first-generation immigrants (specifically  Mental illness among second-generation Pakistanis is lower than White natives, depending on the outcome. Established first-generation Pakistani immigrants have the lowest outcomes and lower than that of natives. The pattern for Black Caribbeans is more mixed.

Tests of the variation by ethnic group and generation (reported in
To explore potential factors driving these results, we augment Equation (1) to include additional variables Z i : The Z i variables include several characteristics that are often thought to influence mental health (partnership status, number of children, employment status, and household income). They also include area characteristics that may capture the extent of social isolation and/or integration: whether the respondent lives in London, the proportion of ethnic minorities in their local area (density), and for non-Whites, the proportion of the local population of the respondent's own ethnic group (concentration). We allow the density variable to interact with ethnicity (a White/non-White dummy) and generation dummies and the concentration variable to interact with generation dummies.
We also include variables intended to capture premigration and postmigration experiences that may affect mental health. For firstgeneration immigrants, we control for years resident in the United Kingdom, age of arrival in the United Kingdom, country of birth, whether from a non-English-speaking country, and whether arrived as a child and spoke non-English in childhood. For second-generation migrant, we control for not speaking English in childhood, whether either parent arrived from a non-English-speaking country, and an interaction between concentration index and either parent arrived from non-English-speaking country, capturing that migrants are likely to move in areas with of same race/origins/language.
An important caveat is that the modelling approach does not engage with the issue of causality. All the Z i variables are potentially endogenous. As such, the regression results permit only a description of the extent to which they are associated with variations in mental health. This is itself useful in a diagnostic sense. We therefore discuss the findings in the context of other results in the literature.  Hence, these results provide little support for the finding in psychiatry studies (Shaw et al., 2012) that living in areas with more people of the same ethnicity has a "protective" (i.e., positive) effect on mental health of ethnic minority, due to the enhanced social support and positive identity and higher self-evaluation.
Analysing the migrant-related variables for first-generation, consistent with existing literature, as time spent in the country increases, mental health deteriorates, converging to that of natives. In similar way, age is associated with worse social dysfunction and loss of confidence. This could be due to the fact that older individuals are more likely to have developed stronger social or cultural ties in their country of origin that may make acculturation more difficult compared with those who arrived at younger age. 5 Considering the heterogeneous group of migrants by country of birth reveals that only European and Caribbean immigrants experience worse social dysfunction and anxiety and depression, respectively.
Arriving from a non-English-speaking country and arriving as a child from a non-English-speaking country do not appear to be drivers of mental health. This may be due to the fact that migrants are on average more educated and more likely to have a good English proficiency, so this does not represent a barrier for first-generation immigrants.
When analysing the migrant-related variables for second-genera-

| CONCLUSION
In this paper, we use a large and nationally representative survey to examine how mental health varies with ethnicity and family migration history.
We find significant variation across both dimensions. Our results provide an insight into how generations progress, as captured through mental health, varies across ethnic groups. For some ethnic groups (including Whites, Indians, and Black Africans), recent migrants have better mental health than established migrants and those who were born in the United Kingdom. There are two obvious interpretations of this. One possibility is that the higher mental health among the more recent migrants will persist such that, over time, the nature of generational differences will change. The opposite possibility is that individual mental health is dynamic and, over time, will decline among those who are currently recent migrants, leaving the generational profile unchanged.
In attempting to understand the reason behind the observed differences, our results control for a range of additional characteristics.
There is a well-established literature on the influences on mental health, and it is possible that the ethnic and generational variations can be accounted for by controlling for these factors. In fact, while doing so does change the findings, it does not account for the variation.
The results are mixed. Mental health of recent non-White migrants is better for those living in areas where their own ethnic group is represented well. The reasons behind these findings are likely to be complex and are perhaps suggestive of the importance of dynamic factors. One interpretation of the results is that the "cushioning" effect of density is important in helping migrants adjust to a new country whereas, in longer term, minorities may have less need for the protective environment of the neighbourhood.
Although speculative, such a portrayal highlights the dynamic nature of an adjustment process. Moreover, although not addressed here, another aspect to consider would be how return migration may change the interpretation of the results. Established migrants are net of onward migration and may be compositionally different as a result.    No variation by ethnic group or generation H 0 : δ eg = 0, ∀e,g 0.0000 0.0000 0.0000