Understanding the Spatial Trajectories of Minority Groups: An Approach that Examines their Demographic, Cultural and Socio-economic Characteristics

Population researchers have contributed to the debate on minority group distribution and disadvantage and social cohesion by providing objective analysis. A plethora of new distribution measurement techniques have been presented in recent years, but they have not provided suffi cient explanatory power of underlying trajectories to inform ongoing political debate. Indeed, a focus on trying to summarise complex situations with readily understood measures may be misplaced. This paper takes an alternative approach and asks whether a more detailed analysis of individual and environmental characteristics is necessary if researchers are to continue to provide worthwhile input to policy development. Using England and Wales as a test bed, it looks at four small sub-populations (circa 250,000 at the turn of the century) – two based on ethnic grouping: Bangladeshi and Chinese; and two based on an under-researched area of cultural background, religion: Jews and Sikhs. Despite major differences in longevity of presence in the UK, age profi le, socio-economic progress, and levels of inter-marriage, there are, at a national level, parallels in the distribution patterns and trajectories for three of the groups. However, heterogeneity between and within the groups mean that at a local level, these similarities are confounded. The paper concludes that complex interactions between natural change and migration, and between suburbanisation and a desire for group congregation, mean that explanations for the trajectory of distribution require examination of data at a detailed level, beyond the scope of index-based methods. Such analyses are necessary if researchers are to effectively contribute to future policy development.


Introduction
Population geographers and other social scientists have been investigating and analysing the spatial distribution of minority groups, and the impact of group segregation on the stability of society for many decades.This work has made a positive contribution to academic and political debate in many parts of the world, by providing an objective analysis of the extensive amounts of data that are available (for example, Rugh/Massey 2010; Ãslund/Skans 2010; Shon 2010; Jivraj/Simpson 2015; Catney 2016).Frequently, this has taken the form of assessing and quantifying the pattern of distribution and its change over time through the use of a wide variety of indices of distribution.
The utility in attempting to summarise a complex picture in a readily understood measure that can be transferred from one situation to another clearly has benefi ts.However, given the need to explain, understand, and predict changes in spatial distribution if policy advice is to be developed, a focus on measurement techniques may be misplaced.It may have contributed to an undue prominence being given to residential segregation, and attempts to reduce it per se, rather than considering positive aspects of group congregation, or identifying location-specifi c underlying health, qualifi cations, employment or other socio-economic disadvantages.Indeed, "housing policies have a limited effect on ethnic concentration, not only because they often contradict each other, but also because they fail to address the main causes of segregation" (Bolt 2009: 397; see also van Ham/Manley 2009;Dhalmann/ Vilkama 2009;Münch 2009;Kempen/Bolt 2009;and Holmqvist/Bergsten 2009).
This paper proposes taking an alternative approach.It suggests that a more detailed analysis of individual and environmental characteristics is necessary if researchers are to continue to provide worthwhile input into the discussion of segregation/integration, disadvantage, and social cohesion.As a case study, analysis is presented for four small cultural groups, based on output from the England and Wales 2001 and 2011 censuses.

Recent Measurement Techniques
Numerous indices have been used to measure various aspects of spatial distribution, with in-depth reviews carried out to reach a consensus on their use (Duncan/ Duncan 1955;Massey/Denton 1988;Simpson 2007).Despite a long period in which use of the Index of Dissimilarity (D) predominated, in recent times a wide range of alternative approaches have been developed.However, few of these have been widely adopted; indeed, as discussed below, a number of researchers have revisited their approaches in subsequent studies.Despite its popularity and its easy interpretation as "the proportion of a group that would have to move to be distributed through localities like the rest of the population" (Simpson 2007: 421), three types of weakness in D have long been recognised: • mathematical/theoretical weaknesses (Cortese et al. 1976;Taeuber/Taeuber 1976;Winship 1977;Voas/Williamson 2000), • Variability of results depending on geographic units -the results of index calculations suffer from the modifi able areal unit problem (MAUP) (Openshaw 1984) and zonation issues (Rees et al. 2017).
• The fundamental aspatiality of the measure (Wong 2016) -in that the absolute and relative positions of the spatial units/zones being used have no impact on the resultant value of the index.The checkerboard analogy, presented by White (1983), demonstrates this in a very clear manner.
Examination of a number of papers that have been published since the release of small scale outputs from the 2011 England and Wales census shows that a wide range of alternative approaches have been applied.These attempt to meaningfully summarise levels of integration or segregation of minority groups, and overcome the weaknesses.
Harris and Owen (2018) attempted to enhance D by developing a Spatial Multilevel Index of Dissimilarity.They considered the impact of adjoining areas at different scales, having previously proposed an Index of Discontinuity (based on differences in values of D in adjoining zones) (Harris 2014).Another approach sought to apply D to measure unevenness in combination with Moran's I spatial autocorrelation coeffi cient to measure clustering (Catney 2017).However, in a later paper, Catney (2018) used both D and the Index of Isolation (P) to examine spatial variation.
An alternative methodology was developed by Johnston (Johnston et al. 2015(Johnston et al. , 2016)).They employed a rule-based approach that categorised areas according to the percentage share of the population provided by white, all non-white, and individual non-white ethnic groups.In parallel with this Jones et al. (2015) were developing yet a further technique that used a modifi ed log-Poisson model based on the degree of over-or under-representation of each group compared with its average presence.This was subsequently enhanced by the introduction of a multi-scale element (Johnston et al. 2016).
Of course, these matters are not solely an England and Wales issue.Internationally, there is a considerable body of research that makes use of assessment of the presence in the nearest k-neighbours of members of the same or different ethnic/ racial/nationality group (Johnston et al. 2004;Reardon et al. 2008).The base unit can be an individual, a small census tract, or gridded data.The technique allows the value of k to be varied to examine group concentrations at a variety of scales, addressing the MAUP issue.In recent years, variations of, and enhancements to, the technique have been applied in Los Angeles and Sweden (Östh et al. 2015;Clark et al. 2015), Seattle (Fowler 2016), and Amsterdam, Utrecht, and Groningen (Netherlands) (Petrović et al. 2018).Geographically wider studies include those of several US cities in the context of the USA as a whole (Hennerdal/Nielsen 2017), and a comparative study of Belgium, Denmark, Netherlands, and Sweden (Andersson et al. 2018).
Overall, we can say that a wide range of alternative approaches and enhancements to earlier indices have been tested.In the case of the UK, their lack of adoption by other researchers, and the apparent continuing need for the original authors to revise their techniques, suggests that attempting to measure and summarise spatial distribution is not advancing the understanding of distribution or the forecasting of its future trajectory.The thesis of the current paper, therefore, is that in this context such approaches should be set to one side.

Approach of this paper
A closer examination of the variation in the characteristics of minority groups may improve the explanation of change in spatial distribution and its underlying drivers.
It is important that this is better understood since cultural and "ethnic diversity, which is now a key characteristic of contemporary society … is an issue of public, policy, political, and academic interest."(Catney 2016: 13).
Using England and Wales as a test bed, this paper seeks to broaden the discussion on understanding spatial distribution trends.It takes the view that heterogeneity within groups means that a more detailed exploration of the data is required if a clearer understanding of spatial distribution trajectory is to be achieved.The 2001 census identifi ed four explicitly defi ned groups (that is, excluding "mixed" and "other" groupings) with a population of around a quarter of a million people -Bangladeshis, Chinese, Jews, and Sikhs.All four groups are examined side by side, using data from the 2001 and 2011 censuses, with the majority white British group used as a benchmark where appropriate.
The case for the selection of these four groups (in addition to their similarity in size) is that: • Religion is an under-researched area of cultural identity, compared with ethnic background; presenting detailed analysis of spatial distribution change for Jews and Sikhs for the fi rst time seeks to redress this.
• This size of group may display a different dynamic to the larger ethnic groups that other researchers have examined.
• As described later, despite similarity in size, the groups demonstrate different demographic and socio-economic characteristics, with a notable variation in their length of material presence in Britain, the importance of students, and levels of recent immigration.
The focus is thus on two groups based on religion, and two based on ethnic background.Similarities and differences in their trajectories provide an interesting and varied basis for the investigation of spatial distribution.
This paper examines census data at a detailed geographic level.For a higher level comparison of ethnic groups in recent UK censuses see Jivraj /Simpson 2015;Finney/Simpson 2008;Simpson/Finney 2009, and for the 1991 census see Rees et al. 1995.British studies need to be seen in the context of interest in cultural group spatial distribution across the globe.Recent international studies, making use of census data, include: an analysis of characteristics of foreign nationals in the Czech Republic (Přidalová/Hasman 2018); an examination of religious diversity in Australia (Bouma/Hughes 2014); a wide-ranging and detailed examination of ethnic diversity in Indonesia (Ananta et al. 2015); and a multi-national investigation of ethnic, religious and linguistic diversity in Africa (Gershman/Rivera 2018).
UK censuses are accepted as being of high quality, and undergo a strong validation process (ONS 2009).However, it is worth briefl y considering some aspects of the ethnic group and religion questions and responses on which this paper focuses, and the differences between them.Pre-defi ned response categories were specifi ed for both questions, an approach that can sometimes be regarded as infl uencing the replies given, as can the order of questions (Blaikie 2000;Voas/Bruce 2004).Uniquely, the religion question was voluntary, and ONS accepted blank responses (about 7 percent of the total) -allocating them to a "religion not stated" category.As with all other questions, the ethnic group question was compulsory though 3 percent of respondents did not answer it; in these cases ONS imputed an answer (ONS 2012).Another difference between the ethnic group and religion questions was that ONS provided "advice" as to what the ethnic group question was seeking to elucidate: "tick one box to best describe your ethnic group or background".However, the religion question simply asked "What is your religion?"(ONS 2011), leaving it to the respondent to determine whether this might mean, for example, belief, family affi liation, or membership.There was a specifi c concern that the Jewish response rate might be low because of concerns about biblical prohibition on counting Jews (see Hosea, 2:1), or holocaust-related mis-use of census data (Brasz 2001).Indeed evidence of some undercount amongst strictly orthodox Jews was found, certainly as regards the 2001 census (Graham/Waterman 2005;Voas 2007).
Ultimately, however, the focus of this paper is on changes over time and underlying characteristics of the members of the groups, and not numerical comparisons between the groups or absolute measures of segregation/congregation.On this basis, use of the census outputs without additional manipulation is a suitable basis on which to proceed.

Theoretical background
Questions about the spatial distribution of minority populations and how this might change over time are not new.As long ago as 1926, Park recognised that change in residential location was associated with levels of education, employment and income (Park 1926).Subsequently, Gordon (1964) defi ned various social/cultural levels of assimilation that he believed represented a trajectory along which minority groups might travel.Twenty years later Massey (1985: 316) set out to develop "a modern theory of ethnic residential segregation", under which immigrant groups would initially form concentrations, but would gradually disperse into the wider community as their socio-economic circumstances improved.This approach was later criticised, for example by Nagel (2009) and Wright et al. (2005), and alternative theories/approaches developed (see for example: Portes /Zhou 1993;Iceland/Nelson 2008;Ehrkamp 2005).Attention has also been given to the positive aspects of group concentration or congregation (Peach 1996;Phillips et al. 2007;Dunn 1998).
Whilst internal (and international) migration is important in bringing about change in spatial distribution of groups (Rees et al. 2013), it is not the only mechanism in play.Other processes, such as natural change due to births and deaths, can have a larger impact on segregation and neighbourhood mix (Bailey 2012).
In summary, several factors have confounded the simplicity of traditional theories about spatial distribution: heterogeneity amongst groups; uncertain economic conditions; changing outlook of later generations; and changing political and social circumstances that impact on individuals' aspirations and expectations.However, most researchers continue to fi nd some form of link between socio-economic advancement, cultural preferences and assimilation, and changes in spatial distribution.

Overview of the Four Cultural Groups
Prior to examining the spatial distribution of the four cultural groups at a detailed level (and how this changes over time), this section provides an overview of the characteristics of each group.These characteristics, all linked to theories of spatial distribution, may provide us with expectations about the likely distribution and trajectory of distribution of the groups.
In this section, four aspects are considered: • Longevity of material presence in, and scale of recent immigration to, Britain -which may infl uence the level of continuing presence in immigrant settlement areas; • Age profi le and thus whether groups might be increasing in presence due to an excess of births over deaths; • Socio-economic status -where improvement might be associated with advancement up the property ladder, and relocation to more sought-after areas; • Inter-marriage -as a marker for cultural assimilation -which some commentators have associated with the post-physical integration period.The main sources used by these authors to produce estimates for earlier years were: census country of birth data (Chinese and Bangladeshis); relationships between number of registered places of worship and numbers of worshipers (Sikhs); and a comprehensive database of burials with ages at death (Jews).Figure 1 identifi es the trajectory of population change for each group over the last 80 years; for each group, 100 percent equals the 2011 census population.1 Inevitably there must be a margin of error around the fi gures, but the purpose in including the diagram is only to establish the general shape of the profi les and the extent to which the four groups are similar to or differ from each other.The profi le for the Jewish population is clearly in stark contrast to the other groups.Over the last 60 years, the Jewish group's population has gradually fallen from its peak, which was more than 50 percent higher than the current population.The other groups have all grown from no more than 5 percent of their 2011 fi gures.This confi rms that Jews have been present in the UK in signifi cant numbers for a much longer period than the other groups considered here.

Longevity of material presence and recent immigration levels
The graph can be supplemented by examining information on year of arrival in the UK of those present at the 2011 census by making use of the 2011 Census Microdata Individual Safeguarded Sample fi les (ONS 2014) -a 10 percent anonymised sample of the England and Wales 2011 census returns -see Table 1.
The profi le for the Chinese group differs from the others, with just under half having arrived since 2001, and with students making up almost half of that element.The majority of students are likely to be transient visitors to the UK, located in key university towns.The overall impact on spatial distribution of such a material proportion of the group having arrived since 2001 depends on whether they have settled in areas where the Chinese group is already present (Luk 2008(Luk , 2009;;Lymperopoulou 2013).Traditional theory would lead us to expect that Jews would be highly dispersed amongst the wider British population, but we already know that this is not the case (Simpson 2012).However, we might expect to fi nd that their physical location is away from the settlement areas occupied a century ago.

Age Profi le
Table 2 summarises a number of characteristics taken from 2011 census outputs, including information on age profi le.The proportion of the Bangladeshi community aged below 15, at 46 percent, is twice that of any other group listed here and indicates that the population is expanding at a fast rate through natural change (see also Simpson/Jivraj 2015).The Jewish group has a larger proportion of its population in the aged 65 and over band than all of the other groups, and has the potential for shrinkage through natural change.The absence of a signifi cant elderly presence in the Sikh and Chinese communities will result in an excess of births over deaths.We should expect to see these natural changes refl ected in changing spatial density.

Socio-economic status
We can anticipate that higher incomes will be a pre-requisite for enabling households to live in (or to move to) the more sought-after/affl uent home-owning areas.Indeed, the median weekly wage (for the 2006-2008 period) for male employees living in a home with a mortgage was 37 percent higher than for those in private rented accommodation and 65 percent higher than those living in social housing (Hills et al. 2010).Unfortunately, the census does not collect information on incomes, but it presents information on areas of employment and the roles played in those occupations -socio-economic status (see Table 2).Moreover, Hills et al. (2010) demonstrate that National Statistics Socio-economic class (NS-Sec) is a useful proxy for income -with a gradual diminution in median male full-time employee hourly wages from NS-Sec Class 1 (higher managerial, administrative and professional occupations) to Class 7 (routine occupations).For those adults who have ever worked (and excluding current full time students), Table 2 demonstrates that a higher proportion of Chinese and Jews accommodate managerial and professional positions (NS-SeC 1 and 2) than do the white British, with Bangladeshis and Sikhs including lower proportions.However, for England and Wales residents (aged 16 or over) overall, 14 percent have either never worked or are full time students.For the Chinese group the proportion is 40 percent (census table LC6206EW).Once these individuals are included, the proportion of Chinese adults falling into the NS-Sec Class 1 and 2 categories falls below the England and Wales average.Despite a relatively low proportion of managerial and professional positions in the Sikh group, that group exhibits the highest levels of home and two-car ownership of all groups.This apparent disconnect between NS-Sec and home-ownership (and its likely consequential impact on spatial distribution) is investigated here.It appears that certain factors relating specifi cally to the Sikh community, not directly discernible from the census outputs, are in play.The British Sikh community has produced an annual report about the community since 2013.The 2014 report analysed responses to a questionnaire that included relevant questions on demography, income, and property.Caution is necessary in making use of information in detail from the report as the analysis was based on a non-validated survey sample, largely self-selected via social media or word of mouth (BSR Team 2014).However, the report indicates that two-thirds of the sample live in households with four or more people and almost half of the adult respondents are living with one or more parent.Two thirds have household incomes above the average for all UK households.There is clearly a tradition of property ownership in the Sikh community; the BSR 2014 indicating that half of all families own more than one property in the UK, and half also own property in India.In addition a third of families own a business in the UK (BSR Team 2014).So it seems likely that the high home ownership level among the Sikh group arises through three factors: a higher total income per household due to a greater number of employed persons per residence (arising from household size and multi-generational households); income through business ownership, noting that the relationship of wages to NS-Sec only relates to employees; and the potential for non-employment related income through property ownership elsewhere in the UK and overseas.
Overall, therefore, we might expect larger proportions of the Jewish and Sikh groups to be located in more sought after residential areas, where home-ownership predominates, compared with the other cultural groups.
The index of multiple deprivation (IMD) provides an alternative way of considering the economic/fi nancial link with location of residence.The Department of Communities and Local Government measures each LSOA 2 in terms of its level of deprivation.The index incorporates a range of measures associated with income, employment, health and disability, education skills and training, barriers to hous-ing and other services, crime and living environment (DCLG 2011).The LSOAs are ranked using the deprivation score, and it is customary to categorise them based on the decile into which they fall (decile 1, most deprived, to decile 10, least deprived).To provide a fi xed base to monitor change in spatial distribution, the 2010 index (based on data collected between the 2001 and 2011 censuses) has been used here.The link between NS-Sec and wage income already described has been found to apply also to the deciles of the IMD (Hills et al. 2010).
Figure 2 indicates the proportion of the 2001 population of each of the small cultural groups that were resident in LSOAs falling into each decile of the index.It is important to note that the index relates to the mean status of the totality of residents of the LSOA -not just members of the groups under examination.
The graph shows that, in 2001, the Bangladeshi population was most concentrated in the most deprived deciles (particularly decile 1), with below average proportions in deciles 4 to 10.The Sikh group is over-represented in deciles 1 to 5, and under-represented in deciles 6-10 though to a much lesser degree.The Chinese group is evenly spread, with the Jewish group demonstrating a mirror-image of the Sikh pattern.
The trajectory of change in group distribution is a key issue under investigation in this paper.Therefore, rather than simply including an equivalent graph based on the 2011 population, Figure 3  Both the Bangladeshi and Sikh groups have seen a reduction in the proportion of their populations resident in the most deprived deciles.This represents a pattern of continuing socio-economic advancement, though the shape of the curve and pivot points of zero change are differently located.Conversely, the Chinese and Jewish lines show the opposite trend.The Chinese trend (which is almost a mirror-image of the Sikh line) is probably explained through the very large increase in student numbers over the decade, given that in most major cities, student residences tend to be found in less-affl uent areas.The Jewish line is less easily explained.It may suggest that the Jewish group's socio-economic progress over the last century has now plateaued and is in a period of fl uctuation; or it may simply refl ect the marked difference in age profi les and fertility levels in different parts of the community (Staetsky/Boyd 2015).
The question is whether the level of change demonstrated here is suffi cient to be refl ected in a noticeable pattern of geographic change on the ground.For example, Figure 3 shows that in 2011 a smaller proportion of the Bangladeshi population is found in LSOAs in decile 1, and a greater proportion is found in decile 3.If the affected LSOAs are close to each other the actual geographic/spatial impact of the IMD decile change would be less noticeable than if the LSOAs concerned happen to be more remote from each other.

Inter-marriage
Table 3 summarises the census information available on the cultural identity of spouses3 and other partners of partnered members of the groups under examination.Note that for the groups based on religion, relationships where the partner did not respond to the question on religion are excluded; an equivalent consideration for ethnic group does not apply as ONS imputes an ethnic group for non-respondents.
It is clear that the Bangladeshi and Sikh groups are overwhelmingly endogamous, whereas there is a material level of inter-marriage for the Chinese and Jewish groups.Gordon's (1964) perspective on this is that these higher levels could only be achieved if the relevant groups were residentially dispersed amongst the wider In reality, the degree of intermarriage may involve more than residential proximity between the groups and individual concerned, so to what extent is this refl ected in the spatial distribution of these groups?

Group Spatial Distribution
Having set out some of the overall characteristics that may have an infl uence on shaping the spatial distribution of the groups, their geographic distribution and trajectories are examined in this section.Whenever census data (particularly for small groups) is being analysed spatially, there is always a balance to be struck between data accuracy and quality, and homogeneity of areas being considered.LSOA geography (defi ned by ONS taking account of consistency of housing type) has been selected.This strikes a balance between small numbers and non-disclosure adjustments affecting OA data, and the potential for MSOA areas to mask within-area variation (see also Catney 2018).LSOAs ranked by the population of the group under examination can be plotted, and contiguous accumulations or communities can be identifi ed.The method used built on that used by Sapiro (2016) to identify Jewish accumulations using MSOA geography.For the current study it was noted that for three of the groups over 80 percent of their populations can be found in each group's most populous 10 percent of LSOAs.These LSOAs were plotted and geographically contiguous groupings of them were regarded as accumulations.Accumulations for each group that each include at least 1 percent of the group's total England and Wales population have been listed in Table 4.Where there is a strong student presence in an accumulation, a fi gure excluding students, approximating to the "permanent" group presence in the area, is also shown in the table.Note that as this approach includes all LSOAs with group populations down as far as 8 persons (Jewish) to 19 persons (Chinese), varying the criterion from the 10 percent proportion has no impact on which localities would be listed in Table 4, or materially change their rounded-tothe-nearest-thousand populations.
There is a marked similarity in the hierarchy of accumulations for the Bangladeshi, Jewish, and Sikh groups.Each has one large moderately dense accumulation in the London area, which is the home to a large proportion of the total England and Wales population (noting that the Sikh community has a second equally large congregation in the West Midlands).There are one or two more moderately sized groupings elsewhere in the London area; one or two signifi cant communities outside of London (Birmingham for Bangladeshis, North Manchester area for Jews; Leicester and Coventry for Sikhs), and then a scattering of smaller accumulations.Given the long period of establishment of the Jewish group in Britain, the level of congregation of that group is exceptionally high compared to what might be expected from traditional theory, though this type of pattern for Jewish communities (in the UK and elsewhere) has been recognised for a long time (Massarik/Chenkin 1973;Newman 1985;Waterman/Kosmin 1987;Kosmin et al. 1991;Kotler-Berkowitz et al. 2004).
The Jewish group is predominantly focussed in outer suburban areas and the small town/partly-rural hinterland beyond the suburban limit.There is no material presence in the original settlement areas of, for example, the "east end" of London and central Manchester (Williams 1990;Freedman 1992;Endelman 2002).The Sikh group follows a generally similar pattern, but with a continuing presence in inner suburbs in some of the West Midlands authorities, and in the original settlement area in Ealing (London).The Bangladeshi group is very strongly focussed in the Tower Hamlets area of east London (that is, the same settlement area occupied by Jews a century earlier), and adjoining Newham.
The pattern for the Chinese group looks quite different.There are no large area/ large population accumulations, and the student element of accumulations is very signifi cant, particularly away from London, where the focus is frequently on the

Tab. 4:
Group accumulations and population (2011) Location of accumulation (and group population in thousands including/excluding students*) Group In London area Elsewhere in England and Wales Bangladeshi Tower Hamlets, Newham, fringe of the inner areas with extensions into the outer suburbs.See Luk (2009) for a more extensive discourse on the distribution of Chinese.There is a much greater scattering of small pockets of Chinese than seen for other groups; Luk (2008) makes reference to Chinese immigrants working in the catering trade deliberately choosing to move away from traditional "Chinatown" areas in order to expand the market and avoid creating an over-supply of Chinese cuisine in a concentrated area.The pattern of change between 2001 and 2011 (as with the actual distribution) shows a remarkable degree of similarity for the Bangladeshi, Jewish and Sikh groups.In terms of the principal focus of the Bangladeshi group in east London, there is a reduced concentration in parts of the most densely populated area.An increase in congregation in a central part of the accumulation has occurred with a noticeable extension into more suburban areas that were not part of the 2001 "heartland".There is thus both a re-focussing within the main area of settlement, and a strong movement into "new" areas outwards from the centre of Londonsuburbanisation (see Farrell 2016 for an American perspective on minority group suburbanisation).In addition this is combined with a noticeable loss of population share in the smaller London communities and scattered areas elsewhere.The Jewish and Sikh groups repeat this pattern, though in those cases the extension is into semi-rural areas and might be better described as counter-urbanisation rather than suburbanisation.

to 2011 changes
There is no noticeable suburbanisation pattern for the Chinese community.Insofar as the London area is concerned there is a refocussing into two inner city areas.This is accompanied by a loss in group proportion in almost every small pocket across the rest of the London area -almost a centralisation process, potentially dis-torted through the dominance of the impact of students in some areas.Indeed, the student impact in provincial areas is much more apparent -with strong growth in university/student areas of the major cities, and (as in London) a loss in proportion in small suburban clusters.
Qualitatively therefore, some clear patterns of change can be observed, but can a closer investigation of the characteristics of shrinking and expanding areas provide a clearer explanation of these changes?That is, does heterogeneity within the groups help us to understand the patterns in a better way than a focus on distribution indices or overarching terms such as suburbanisation?
As already described, the 34,000 LSOAs have been ranked by the 2001 to 2011 change in proportion of the LSOA's population that belongs to each group under examination.This showed that material change is limited to around 400 LSOAs at each end of the spectrum.Table 5 provides an overview of the characteristics of areas that have had an increase in group proportion over the 10 year period (gaining/ growth areas), and those where there has been a reduction (contracting/shrinking areas), based on the two groups of 400 LSOAs relevant to each group.2015: 43-45) then the differences will represent the impact of births, deaths, and migration over the intercensal period.More specifi cally, the number of 0-9 year olds in 2011 will represent births over the intercensal period (subject to some migration, depending on the scale of migration for their parents' age band).Differences in profi les between the ages of 10 and 59 will represent net migration (subject to a small number of deaths); and differences over the age of 60 will be largely representative of deaths (and some older age migrations) (Ballard 2004;Simpson/Jivraj 2015).
The age band population data that ONS has made available for ethnic and religion based groups in the 2001 and 2011 censuses at LSOA level allows differences in the profi les to be split at age 10 and 60.It has been necessary to assume that the age 60+ difference all relates to deaths; however, except for the Jewish group, the proportion in that age band for the cultural groups is very small So, natural change has been determined by subtracting the over 60 difference from the number of 2011 0-9 year olds, and the age 10-59 difference has been used to represent migration.
Fortunately, for all groups, the difference between the calculated natural change and migration has been far greater than would be nullifi ed by errors caused by the simplifi cations that have had to be made.The table is thus able to indicate whether migration (denoted as "mig") or natural change ("nat") is the greater source of intercensal population change for each group.There are some common characteristics for all four groups: contracting areas have a higher proportion of people with poor health than gaining areas and (not shown in the table) this disparity has increased over the 10 year period.Given that group proportion is reducing in these areas, it is unlikely that the movement into the area of people with poor health is a signifi cant factor.The conclusion to be drawn is that people with poorer health are less likely to move home and get "left behind" in areas of group shrinkage.Indeed their health may suffer through increasing isolation from other members of their group (Smith/Easterlow 2005).Darlington et al. (2015) provide a discussion on the complexities of unravelling the relationship between health and internal migration.
For three of the groups, areas of growth have a higher proportion of 0-15 year olds and a lower proportion of those aged 75 than do contracting areas; again (not shown in the table) the disparity has increased in the intercensal period.This agerelated aspect to changes in distribution was also found in America (Winkler/Johnson 2016).As might be expected, areas with an increasing group proportion benefi t from positive natural change (excess of births over deaths), whereas the opposite situation applies to shrinking areas, though for most groups migration (whether international or internal) has a larger impact on the outcome.The Chinese fi gures are heavily infl uenced by the very high student presence in growth areas (71 percent); nevertheless the small levels of natural change are also positive for growth areas and negative for shrinking areas.
There is a marked difference in population density between the groups of LSOAs for Sikhs -refl ecting the urban locations for the shrinking areas and the suburban gaining localities.There is no differential for the (suburban) Jewish group, but the Bangladeshi fi gures refl ect the very dense urban locality of the contracting areas and the more typical urban value for growth areas.It is only the Sikh areas that show a marked difference between the average decile of deprivation in which the groups of LSOAs lie -with the growth areas on average 3 deciles less deprived than the shrinking areas.This fi gure is actually exceeded by the white British group (not shown in the table).This should be compared with the Swedish picture where there is an apparent inability of ethnic minority groups located in poorer neighbourhoods in childhood to relocate to better areas in adulthood (Gustafsson et al. 2017).
Whilst this summary provides an interesting overview, to address the infl uence of heterogeneity of individual and area characteristics, we need to examine whether these patterns apply consistently across the actual localities where reducing and increasing population shares have been found.Table 6 summarises the same characteristics, area by area.The areas have been named after the main local authority involved, though they are defi ned in terms of groups of contiguous LSOAs (annotated in the table) where there is shrinkage or expansion.For reasons of space, only the largest areas are shown.
In 2001 and 2011 the more suburban of the Bangladeshi increasing share areas had a lower proportions of 0-15 year olds, higher proportion of people over 50, degree holders, employed persons, and home owners and less over-crowding than Bangladeshis living in other growth areas.Migration has a more important role in growth areas than is the case elsewhere.Indeed, there are differences between areas where natural change has been the larger contributor to expansion.They are generally located in areas of highest deprivation, and where the proportion of 0-15 year olds exceeds 40 percent, and proportions in employment and good health are lower.
Such is the dominance of the student impact on the Chinese group that only two expanding areas (one is shown in the table) could be located that were not dominated by students, though even in these places students make up a disproportionate element.Most of the student-dominated expanding areas have seen a fi ve-to ten-fold increase in the number of Chinese residents over the decade, and also large increases in the proportion made up by students.Their differentiating characteristic is thus the level of popularity of their universities with (international) Chinese students.And as the Chinese group is relatively scattered, only one small group of 5 LSOAs that encompassed a reducing proportion could be found.
There are clear differences between some of the gaining Jewish areas.Those with the largest changes in population share in the decade are located in areas of fairly high deprivation, have extremely high proportions of 0-15 year olds, low levels (compared with the Jewish group overall) of older people, low levels of poor health, and generally low levels of degree holders and home ownership.Natural change is a more important source of population change than migration.These areas are known to be the home of Britain's strictly orthodox Jewish communities, which demonstrate a high level of fertility (Graham 2013;Staetsky/Boyd 2015).The other growth areas are located in areas of lower deprivation, and whilst they feature above average (rather than very high) proportions of 0-15 year olds and larger proportions of older people than the fi rst three areas, adults in these areas are more likely to be employed, hold a degree, and be home-owners.Unlike the other small The Sikh analysis includes a number of pairs of proximate localities between the increasing and reducing areas, primarily in the Midlands, where the contracting area has a high level of deprivation and a higher population density, and the growth area a lower or much lower level.This implies movement by those with improving socioeconomic standing from one area to the other.The growing areas also benefi t from positive natural change refl ected in higher proportions of 0-15 year olds, which may contribute to the lower level of poor health in the expanding areas.

Discussion and Conclusions
Some global characteristics of the groups, linked to traditional theories, have been explored.This has identifi ed some similarities, but also some quite distinct differences, between the groups.Have these differences been refl ected in their spatial distribution?The longevity of the Jewish group's presence in Britain is refl ected in its almost complete absence from its areas of original settlement, whereas the Bangladeshi group is still much focussed in its settlement areas and the Sikh group maintains a presence in its original areas.The broadly similar scale of clustering found in these three groups means that, contrary to traditional theory, length of presence has not led to a dispersed pattern.Moreover, the link between intermarriage levels (which vary greatly between these three groups) and level of dispersal (which does not) is weak.Socio-economic status (and more specifi cally, professional and managerial positions) has a link with location of residence, with groups with higher proportions of employment at these levels being found in areas in less deprived IMD deciles.Thus, if we review some key elements of traditional theories in the light of this study's fi ndings, we can conclude that socio-economic progress has led to relocation of groups away from their initial settlement areas towards more sought-after residential localities.However, such moves have not led to widespread dispersal (for three of the groups), nor (for the Jewish group) has intermarriage remained limited due to a lack of residential dispersal.
There are some broad similarities between the groups in areas where the proportion of a group has reduced between 2001 and 2011, and another set of broadly shared characteristics for areas that have expanded.Two key fi nding are, fi rstly, that areas where the concentration of a group has been relatively high, but has reduced, are locations of worsening health inequality.Secondly, areas where the proportional Sikh presence is above average and increasing actually demonstrate reduced socio-economic disadvantage.When an index-based approach is used, sub-areas of the study area that demonstrate a group population proportion equal to the average for the study area as a whole generally contribute a zero value to the index.Larger values are contributed by areas where the group proportion varies more from the overall average.If two time periods/censuses are being compared, then areas which have moved closer to the study area average would be making less of a contribution to the index, and this reducing value and would be considered to represent an "improving" result.Conversely, areas moving further away from the average value between the censuses would be demonstrating an increasing contribution to the index value and would be regarded as a "worsening" result.Thus in the case of the two types of area described at the start of this paragraph, an index-based approach would conclude that the fi rst type was an improving situation (and thus warranting reduced attention) and the second type a worsening scenario.That is, an indexbased approach can lead to misleading conclusions in targeting disadvantage.
When the analysis explores individual areas where the group population proportion has materially increased or reduced, not only are there many differences between the cultural groups, but there is much reduced consistency of characteristics within each group.However, there is a degree of consistency insofar as the relative importance of migration (whether international or internal) and natural change is concerned.Despite the greater presence of older people in the Jewish group compared with the other groups under examination (and its likely negative impact on natural change), positive natural change (rather than migration) is the main driving force in expanding Jewish areas.Conversely, despite the positive natural change overall refl ected in the high proportion of 0-15 year olds in the other groups (particularly the Bangladeshi group) migration is the more important element in explaining changed group proportion in increasing areas for those groups.Thus, particularly for areas where group proportion has increased, the more important element is at odds with what might be expected from the overall age profi le of the group.
The Chinese group is quite different to the other three groups that have been examined.Whilst the very high presence of international students within that group creates a distinctive pattern to that group's distribution and the large rise in student numbers over the decade overwhelms the pattern of change, it does not explain all the differences.Levels of congregation of (non-student) members of the Chinese group are very low.Peach notes that "urban concentration allows the groups to pass the threshold size at which ethnic shops and religious institutions can be maintained and the proximity to members of the groups that allows the language and norms of the groups to be maintained" (Peach 1996: 386).This is refl ected in the distribution found for three of our groups, but not the Chinese.So why might that be the case?There could well be a religion element to this.By defi nition, of course, there is a religious commonality within the Jewish and Sikh groups, though the degree of actual religiosity may vary (Graham et al. 2014), and census outputs indicate that 90 percent of Bangladeshis described themselves as Muslim (and only 1 percent had "No religion").The Chinese group, on the other hand, includes 56 percent who indicated that they had "No religion"; the next largest group (20 percent) were Christians.Neither of these groups requires residential concentration.Additionally, at one of end of the scale, the majority of Bangladeshi early immigrants originated in a relatively compact area of Bangladesh (Sylhet) and many later immigrants have family connections with the earlier arrivals (Ballard 2004).In contrast, the origin of Chinese immigrants has evolved over recent decades.Initially, the arrivals were lowly qualifi ed residents of rural Hong Kong; more recent immigrants came from a much wider section of south-east Asia.The most recent arrivals have been highly qualifi ed (Luk 2008(Luk , 2009)).There is thus a large inherent heterogeneity in the Chinese group.
The infl uence of the different elements of the benefi t of congregation clearly differ between the groups, and no one over-arching explanation for their current spatial distribution and its trajectory can be found.Indeed, the migration element of distribution appears to embody a tension between suburbanisation/counter-urbanisation "fuelled" by socio-economic progress, and a desire for group congregation.The conclusion is that not only does heterogeneity between the groups confound any attempt to produce a "model" of spatial distribution, but heterogeneity within each group means that producing any form of understanding of, or explanation for, the trajectory of distribution requires the examination of data at a detailed level, and is beyond the scope of global indices.In addition to presenting such information for previously studied ethnic groups, the paper has presented, for the fi rst time, detailed analysis for two religion-based groups.
The key message of this research is that to achieve a full understanding of minority group spatial distribution, investigation at a fi ne geographic level is required.More importantly, full account of the personal, household, and community characteristics of those involved is essential, rather than placing too much reliance on indices and other measures that underplay the issue of heterogeneity within groups.Indeed, by taking proper account of group heterogeneity in analysing minority group spatial distribution, population analysts can better assist policy makers to spatially target health, education, training, and other social programmes and, as a consequence, address issues of segregation, disadvantage, and cultural diversity.

Fig
Fig. 2:Allocation of 2001 group population to IMD deciles

Fig
Fig. 3: Percentage point change in IMD decile 2001-2011 So, what is the trajectory of distribution; where have the 2001 to 2011 changes taken place?The proportion of the population of each LSOA belonging to each of the groups in 2001 and 2011 was compared, and the LSOAs ranked by group proportion change.For each group, the 150 LSOAs (across England and Wales as a whole) where the group's share of the population had increased by the largest percentage (positive change) were identifi ed.Similarly the 150 LSOAs where the share had reduced by the largest proportion (negative change) were also determined.A mapping exercise to display spatial distribution and change has been carried out, and extracts of England and Wales maps showing the London area are shown in Figure4.To be clear, the maps show those of the 300 largest change LSOAs from England and Wales as a whole that happen to fall in the London area (which, for all groups, includes a large proportion of the 300).Hatching on the maps shows areas where the particular group made up more than 6 percent of the 2001 population.This allows the reader to identify unhatched but shaded/coloured areas where the 2001 proportion was relatively low but there has been a large intercensal change.Hatched but unshaded/uncoloured areas where the 2001 proportion was relatively high, but the change between 2001 and 2011 has not been very large, can also be observed.

Fig. 4 :
Fig. 4: Change in distribution 2001 to 2011 at LSOA level Characteristics of 400 LSOAs with highest gain and loss of group population share * "mig" and "nat" indicate whether migration or natural change (births and deaths) is the greater contributor to the overall + (positive) or -(negative) change in group population Group Group characteristics (proportion of 2011 area group population) calculations based on analysis of 2001 and 2011 census outputs

Tab. 6 :
Characteristics of areas with largest increasing and reducing group population share and "nat" indicate whether migration or natural change (births and deaths) is the greater contributor to the overall + (positive) orand "nat" indicate whether migration or natural change (births and deaths) is the greater contributor to the overall + (positive) orcalculations based on analysis of 2001 and 2011 census outputs cultural groups, for Jews, natural change is a more infl uential source of population change than migration for all the gaining and most of the contracting areas.