Stability and change in ethnic groups in England and Wales

Po p u l a t i o n Tren d s 1 2 1

Autumn 2005

Stability and change in ethnic groups in England and Wales Lucinda Platt University of Essex Ludi Simpson Cathie Marsh Centre for Census and Survey Research, University of Manchester Bola Akinwale Office for National Statistics

The Office for National Statistics (ONS) Longitudinal Study (LS) is an exceptional resource for exploring dynamic processes in individuals’ lives for a representative sample of the population of England and Wales and across a thirty year period, including how those processes vary by ethnic group. However, analyses tend to assume a certain stability in the meaning of the ethnic group being studied: the insights into ethnic group differentiation are premised on the fact that the group has the same meaning over time.

INTRODUCTION The ONS Longitudinal Study (LS) is well recognised as an exceptional resource for exploring dynamic processes in individualsʼ lives for a representative sample of the population of England and Wales across a thirty year period. Issues such as the relationship between social class and mortality, inter- and intra-generational occupational mobility, changing patterns of fertility and household formation, geographical mobility and the changing characteristics of the areas in which individuals live are a few examples of such dynamic processes that the LS can shed light on. Since the inclusion of an ethnic group question in the 1991 Census, with a second question asked in 2001, it has also been possible to examine how such processes vary by ethnic group, enhancing our understanding of differences between ethnic groups and of groupspecific processes. Indeed, the sample sizes in the LS compared to those in other longitudinal resources covering England and Wales, such as the British Household Panel Survey or the cohort surveys, render it a unique resource for examining variation in dynamic processes within a sample of the minority ethnic population.

Here we show how the LS allows us to challenge such notions of group stability. This has practical implications for the ways we measure and conceive of Britain’s minority ethnic groups. We illustrate this point with two examples: one exploring the change in ethnic group identification by the same individuals between 1991 and 2001, and the second exploring how loss to follow up is differentially experienced according to ethnic group.1 We provide some suggestions on the implications of this ethnic group instability for other research.

However, such studies tend to assume a certain stability in the meaning of the ethnic group being studied: the insights into ethnic group differentiation are premised on the fact that the group has the same meaning over time, and that differences in timing of first birth, geographic mobility, mortality and so on can tell us something about that group. Here, using two main examples, we suggest that the LS shows us ways to challenge such notions of group stability. These have practical implications for the ways we measure and conceive of Britainʼs minority ethnic groups and shed light on, or provide cautions to, the results of investigations charting minority group processes over time. They also

35

N a t i o n a l S t a t i s t i cs

Po pul ati on Tren ds 1 2 1

Autumn 2005

offer interesting avenues of research in their own right. By looking at dynamics within the composition, formation and identification of ethnic minority groups themselves, we illustrate the potential of the LS to allow investigation of new research questions relating to ethnicity and to shape and modify existing ones. First, we explore change in ethnic group identification by the same individuals between 1991 and 2001. Second, we investigate how loss to follow-up within the LS varies by ethnic group, with the result that observed ethnic groups within the LS post-1991 are of a different size and have different characteristics than we might anticipate from their distributions in 1971. To begin with, in the next section we describe the LS and discuss the 1991 and 2001 ethnic group questions in more detail.

METHODS The ONS Longitudinal Study The ONS Longitudinal Study (LS) is a 1 per cent sample of the population of England and Wales that is followed over time. The LS was started in 1974 and was initially obtained by taking a sample of the 1971 Census, based on those born on one of four birth dates (day and month). Information from samples taken at each subsequent census has been added to the study. Members are also added to the study by linking information on births and immigrations using the same selection criteria. No more information is linked where study members have records indicating that they have died or left England and Wales, though linking recommences for returning emigrants. The sample approximates to a one per cent sample of the resident population of England and Wales, or about 500,000 individuals, at each census.

The ethnic group questions in the 1991 and 2001 Censuses of England and Wales The first Census ethnic group question was asked in 1991 when 7 per cent of the LS sample described themselves as belonging to a minority ethnic group.2 Given the size, the longitudinal design of the LS and its core variables, this introduced the potential for extensive new analyses of dynamic processes across ethnic groups, linking the ethnic group information of LS members present in 1991 to earlier and subsequent records. The census measurement of ethnic group recognises that ethnicity is not necessarily a stable attribute, and that responses could be affected by the nature and wording of questions.3 In 2001, the ethnic group question was revised to reflect evolution in the terminology used to describe ethnic groups in England and Wales, and at this point, 9 per cent of LS members were recorded as identifying with a minority ethnic group.4 The 2001 question differed from the 1991 question both in its wording and range of categories. These different categories enabled new options to self-identify and the opportunity for individuals to change their choice of group. Respondents were presented with the new options of identifying themselves as being Irish or as having a ‘Mixedʼ ethnic origin, using pre-coded tick boxes in 2001. The inclusion of an Irish category recognised evidence for the poor social conditions, on average, not only for Irish-born but their families born in Britain, and enabled separate analysis of this group within the census.5 The inclusion of mixed categories enabled analyses of the special issues faced by children of parents from different ethnic groups and the growing number of mixedorigin residents, estimated before the census at more than ten per cent of all residents of non-White origin.6 The census questionnaires of 1991 and 2001 gave different guidance notes to aid respondents in understanding what is meant by their ‘ethnic groupʼ. In 2001 respondents were asked to tick or write in their ‘cultural backgroundʼ, while in 1991 the note used the terms ‘descendedʼ and ‘ancestryʼ, giving more emphasis to family rather than cultural origins. The two questions are illustrated in Box 1.

Nati on a l S t a t ist ic s

36

In 2001, tick boxes were grouped into five sections with space for a write-in answer within each one, while in 1991 there were just two write-in spaces. These changes to the question mean that, while some groups had the same label, people could select different categories on the two occasions simply because of the different options available. In addition, the new category options and wording changes had the potential to cause shifts in the meaning of the categories that had the same label in both 1991 and 2001, and mean that the composition of groups with the same label at both censuses may not be directly comparable.

IMPUTATION OF THE ETHNIC GROUP ITEM IN THE 2001 CENSUS At the 2001 Census, 2.9 per cent of responses to the ethnic group question were imputed, falling to 2.1 per cent among LS members linked between 1991 and 2001. Imputation rates varied by ethnic group, with higher proportions of imputed values in the Mixed, Black and Asian groups. Table 1 shows imputation rates by ethnic group within the LS. Overall, almost 91 per cent of 2001 ethnic group imputations were consistent with the 1991 category recorded for an LS member, but this includes the white majority group. Among those from 1991 minority ethnic groups with an imputed ethnicity in 2001, less than half were imputed to the same ethnicity as they responded with in 1991. Just under 50 per cent of those who responded as Pakistani or Indian in 1991 and who had an imputed ethnicity in 2001 were imputed to these ethnic groups; and only 1.4 per cent of those who responded as Black Other in 1991 and were imputed to an ethnic group in 2001 were imputed to the Black Other group. The other minority groups fall in between, with 37 per cent of Bangladeshis, under 30 per cent of Black Caribbeans and Black Africans and 10 per cent of Chinese subject to imputation of ethnic group at 2001 being imputed to the same ethnic group. This compares with 97.5 per cent of those recorded as White in 1991 whose ethnicity was imputed in 2001 being imputed to the same group. Rates of imputation based on all records for an ethnic group show that non-response for the ethnic group question was low varying from 1.8 per cent of Chinese to 6.0 per cent of ‘Black Othersʼ. Consequently the overall effect of imputation is quite small with just 0.2 per cent of all records imputed to a different ethnic group in 2001. However, the results of census imputation of ethnic group can be seen to introduce inconsistency in the recording of the item, when examined longitudinally, for individuals from minority ethnic groups. This is unsurprising as item imputation is designed to provide accurate cross-sectional, aggregate variable distributions and, as such, imputation is much more accurate for the majority ethnic group, White. Studies of individual level LS data with 2001 ethnic group as an independent variable should therefore consider omitting those cases where ethnic group has been imputed, or, at the very least, compare the distribution of outcomes of interest among ethnic groups separately for those with imputed values to check for possible bias.

INSTABILITY OF ETHNIC IDENTIFICATION BETWEEN 1991 AND 2001 Researchers are generally aware that ethnicity is an aspect of identity that is subjective, multi-faceted and contingent upon biographical and broader socio-economic and political events.7, 8 However, analysis by ethnic group usually assumes ethnicity is a stable characteristic for individuals throughout the life course. Using the LS, it is possible to quantify the stability of responses to the ethnic group question between 1991 and 2001. For this analysis, we include all records linked between the two censuses, including those where ethnic group was imputed in 2001 because we aim to explore the reliability and meaning of ethnic


Autumn 2005

Box one The 1991 and 2001 England and Wales Census ethnic group questions 1991 Ethnic Group

White Black-Caribbean

Please tick the appropriate box

Black-African Black-Other

® ® ® ®

0

® ® ® ® ®

3

1 2

please describe

Indian Pakistani Bangladeshi

If the person is descended from more than one ethnic or racial group, please tick the group to which the person considers he/she belongs, or tick the ‘Any other ethnic group’ box and describe the person’s ancestry in the space provided

Chinese Any other ethnic group

4 5 6

please describe

2001

What is your ethnic group?

C

Choose ONE section from A to E, then tick the appropriate box to indicate your cultural background. A

®

Indian

®

Bangladeshi

®

Any other Asian background please write in

®

Pakistani

®

African

White

®

British

®

Any other White background please write in

®®®®®® ®®®®®® B

Asian or Asian British

®

Irish

®®®®®® ®®®®®® D

Mixed

®

White and Black Caribbean

®

White and Black African

®

White and Asian

®

Any other Mixed background please write in

®®®®®® ®®®®®®

Black or Black British

®

Caribbean

®

Any other Black background please write in

®®®®®® ®®®®®® E

Chinese or other ethnic group

® ®

Chinese Any other, please write in

®®®®®® ®®®®®®

37



Table 1

Autumn 2005

Imputation of ethnic group 2001 for each 1991 ethnic group Percentages Rates of imputation Based on all records

Ethnic group in 1991

White Black Caribbean Black African Black Other Indian Pakistani Bangladeshi Chinese Other groups – Asian Other groups – Other All groups

Based on records with ethnic group imputed in 2001

Per cent imputed

Per cent imputed to a different group

Per cent imputed to the same group

Per cent imputed to a different group

2.1 4.9 3.7 6.0 2.8 4.3 4.3 1.8 3.3 3.9 2.1

0.1 3.4 2.7 5.9 1.4 2.2 2.7 1.6 2.8 3.9 0.2

97.5 29.9 28.3 1.4 49.2 49.5 37.5 10.0 15.9 0.0 90.9

2.5 70.1 71.7 98.6 50.8 50.5 62.5 90.0 84.1 100.0 9.1

Source: ONS Longitudinal Study, authors’ analysis

group as published in census output (exclusion of records with imputed 2001 ethnic group does not materially affect the results). Table 2 provides a transition matrix of LS membersʼ 1991 and 2001 ethnic group identification. It shows that, while most LS members responded with the same ethnic group label at both censuses, some categories had higher proportions of individuals responding with the same ethnic group at both time points than others. People who responded as White in the 1991 Census gave the most consistent responses to the 2001 question, with over 99 per cent choosing one of three White response categories in 2001. The new 2001 White Irish group was comprised almost entirely of people who identified as White in 1991 (99.3 per cent). The 2001 Other White group included a proportion of people who were identified as Other in 1991 (3.4 per cent) in addition to those who were grouped as White in 1991 (94.6 per cent). There were also high levels of consistency in recorded ethnic group between 1991 and 2001 among people who identified with one of the South Asian groups or as Chinese in 1991. Over 91 per cent of those who identified as Indian, Pakistani, Bangladeshi or Chinese in 1991 also chose the same group in 2001. In the same way, among people recorded in a South Asian group or as Chinese in 2001, the vast majority were recorded in the same group in 1991, ranging from 88.6 per cent among Chinese to 94.8 per cent among Bangladeshis. The consistency of responses to the 2001 ethnic group question was lower among people categorised as Black African or Black Caribbean in 1991. Around 77 per cent of people who were categorised to one of these groups retained the same label in 2001. Almost 8 per cent of people categorised as Black Caribbean in 1991 shifted to the Other Black group in 2001, though Black Africans were less likely than Caribbeans to opt for the Other Black label (3.4 per cent). Around 6 per cent of 1991 Black Caribbeans opted for the new mixed White & Black Caribbean label, and slightly over half that share of 1991 Black Africans opted for their corresponding mixed label, White & Black African (3.3 per cent) in 2001. This difference between 1991 Black Africans and Black Caribbeans in the likelihood of individuals changing their reported ethnicity to a mixed White & Black category may be connected to the greater likelihood of Black Caribbeans being in inter-ethnic partnerships,9,10 and, by implication, higher rates of mixed ethnicity among the next generation. In addition, 7 per cent of 1991 Black Caribbeans and 8 per cent of Black Africans were recorded in 2001 with one of the White categories.


38

Unsurprisingly, the 1991 Other groups (Black Other, Other Asian and Other) were much less consistent in their responses. Membership of the Asian Other group was the most consistent of the three between the 1991 and 2001 Censuses. While among 1991 Other Asians, substantial proportions opted for the residual Other category (29.1 per cent) in 2001, and notable proportions also identified as Indian (14.7 per cent) or Pakistani (4.9 per cent), a third (33.6 per cent) of 1991 Asian Others retained that label in 2001. This was four times as many as for the 1991 Black Other group and was despite the fact that the Other Asian category was constructed in 1991 from responses to the Other option, whereas it was a distinct tick-box option in 2001. Table 2 shows that the 1991 Black Other group were distributed among a diverse group of categories in 2001 and just 8.3 per cent identified with the 2001 Black Other category. Over a fifth identified themselves as Black Caribbean at 2001 (20.6 per cent). The revised headings to the 2001 question may play a role in this change in identification, since these 2001 headings explicitly labelled the Black groups, ‘Black and Black Britishʼ, and a large number of those who had defined themselves as Black Other in 1991 had written in a description of Black British.11 Black Other has often been combined with the Black Caribbeans in analysis to increase sample sizes; however, these results show that under a third of 1991 Black Others identified with either Black Other or Black Caribbean at 2001, which would tend to cast doubt on the appropriateness of such an approach. For the residual 1991 ‘Otherʼ group only 7 per cent fell under the same label in 2001. Of the 2001 Other group, 52 per cent were drawn from the 1991 Other Asian group, which is consistent with the fact that 1991 Other Asians had in fact identified as ‘Any other ethnic groupʼ, in 1991 and just 19.1 per cent of the 2001 Other group had been categorised as Other in 1991. For the three 1991 Other groups part of the inconsistency appears to represent a lack of clear identification with a single category: more than 15 per cent of 1991 Black Others chose one of the 2001 White groups and there was a large shift to the Black Caribbean category (20.6 per cent). Among 1991 Other Asians, substantial proportions identified as Indian (14.7 per cent) and Pakistani (4.9 per cent) in 2001. For the residual Other group, a large share moved to White British in 2001 (15.2 per cent). However, a substantial proportion of the change in these Other groups (as well as in the Black Caribbean and Black African categories) can be associated with the increased options available in 2001. The introduction in 2001 of a set of mixed categories appears to have given many people

Table 2

Transition in ethnic group between 1991 and 2001 Numbers and percentages Ethnic group in 2001

Ethnic group in 1991

Other White

Black Caribbean

380,096 205 84 148 156 96 40 44 24 313 381,206

4,704 10 4 4 10

5,660 22 11 38 34 10

Per cent of 1991 group White Black Caribbean Black African Black Other Indian Pakistani Bangladeshi Other groups – Asian Chinese Other groups – Other Total Per cent of 2001 group White Black Caribbean Black African Black Other Indian Pakistani Bangladeshi Other groups – Asian Chinese Other groups – Other Total

Numbers White Black Caribbean Black African Black Other Indian Pakistani Bangladeshi Other groups – Asian Chinese Other groups – Other Total

39 N a t i o n a l S t a t i s t i cs

Black African

Other Black

Indian

Pakistani

165 2,617 30 255 16

68 22 956 38 22 7

51 264 42 103 4

8

9

18

16

99 5 4 20 48 3,960 14 66

5 9 5 11 1,543 12

113 12 7 21 381 134 25 451

5 4,737

201 5,984

68 3,160

21 1,152

35 515

126 8 24 24 7,825 49 17 197 3 81 8,354

68 4,284

13 1,627

96.8 6.0 6.8 12.0 1.8 2.2 2.4 3.3 2.1 15.2 91.3

1.2 0.3 0.3 0.3 0.1 0.0 0.0 0.0 0.0 0.2 1.1

1.4 0.6 0.9 3.1 0.4 0.2 0.0 0.6 0.0 9.8 1.4

0.0 77.2 2.4 20.6 0.2 0.0 0.0 0.7 0.0 3.3 0.8

0.0 0.6 77.4 3.1 0.3 0.2 0.0 1.3 0.0 1.0 0.3

0.0 7.8 3.4 8.3 0.1 0.0 0.0 1.2 0.0 1.7 0.1

0.0 0.2 1.9 1.9 91.0 1.1 1.0 14.7 0.3 3.9 2.0

0.0 0.1 0.3 1.6 0.6 91.9 0.8 4.9 0.0 3.3 1.0

99.7 0.1 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.1 100.0

99.3 0.2 0.1 0.1 0.2 0.0 0.0 0.0 0.0 0.1 100.0

94.6 0.4 0.2 0.6 0.6 0.2 0.0 0.1 0.0 3.4 100.0

5.2 82.8 1.0 8.1 0.5 0.0 0.0 0.3 0.0 2.2 100.0

5.9 1.9 83.0 3.3 1.9 0.6 0.0 1.6 0.0 1.8 100.0

9.9 51.3 8.2 20.0 0.8 0.0 0.0 3.1 0.0 6.8 100.0

1.5 0.1 0.3 0.3 93.7 0.6 0.2 2.4 0.0 1.0 100.0

2.3 0.1 0.1 0.5 1.1 92.4 0.3 1.5 0.0 1.6 100.0

Note: Includes cases where 2001 ethnic group is imputed Source: ONS Longitudinal Study; authors’ analysis

Bangladeshi

29

Other Asian

Chinese

67

Other Ethnic Group

127 5 14 21 3

White and White and Black Black Caribbean African

367 194 9 396

98 7 41 79 5

White and Asian

Other Mixed

Total

487

325 26 13 66 18 5 4 58 20 204 739

392,582 3,392 1,235 1,235 8,596 4,310 1,652 1,341 1,118 2,053 417,514

141 1,285

38 1,017 26 1,148

390 48 144 752

300 1,269

79 309

20 51 35 9 31 6 354 993

0.0 0.0 0.4 0.7 0.1 0.3 93.4 0.9 0.0 0.6 0.4

0.0 0.4 0.6 1.7 4.4 3.1 1.5 33.6 0.0 6.9 0.3

0.0 0.0 0.0 0.0 0.0 0.0 0.0 2.8 91.0 1.3 0.3

0.0 0.0 0.4 1.1 0.2 0.1 0.0 29.1 4.3 7.0 0.2

0.1 5.7 0.7 32.1 0.0 0.0 0.0 0.2 0.0 14.6 0.3

0.0 0.2 3.3 6.4 0.1 0.0 0.0 0.0 0.0 3.8 0.1

0.1 0.0 0.0 1.6 0.6 0.8 0.5 2.3 0.5 17.2 0.2

0.1 0.8 1.1 5.3 0.2 0.1 0.2 4.3 1.8 9.9 0.2

100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0

1.8 0.0 0.3 0.6 0.3 0.7 94.8 0.7 0.0 0.8 100.0

8.8 0.9 0.5 1.6 29.7 10.4 1.9 35.1 0.0 11.0 100.0

5.8 0.0 0.0 0.0 0.0 0.0 0.0 3.3 88.6 2.3 100.0

16.9 0.0 0.7 1.9 2.8 0.4 0.0 51.9 6.4 19.1 100.0

28.9 15.3 0.7 31.2 0.0 0.0 0.0 0.2 0.0 23.6 100.0

31.7 2.3 13.3 25.6 1.6 0.0 0.0 0.0 0.0 25.6 100.0

49.0 0.0 0.0 2.0 5.1 3.5 0.9 3.1 0.6 35.7 100.0

44.0 3.5 1.8 8.9 2.4 0.7 0.5 7.8 2.7 27.6 100.0

94.0 0.8 0.3 0.3 2.1 1.0 0.4 0.3 0.3 0.5 100.0

3

Autumn 2005

White Irish


White British


Autumn 2005

an opportunity to identify with a preferred label, possibly indicating that 2001 ethnic group identification is more precise than that for 1991. In 1991, there were no tick boxes for mixed categories, and people could only identify as mixed with a write-in answer. Those who did write in a mixed identity were coded as mixed before being allocated to one of the residual categories used in standard tabulations from the census. Table 3 shows that less than half of those identifying with one of the 2001 mixed groups used the write-in opportunity to identify themselves as mixed in 1991. Of the 2001 mixed groups, people who identified as White & Black Caribbean in 2001 were most likely to have written in a mixed ethnicity response in 1991, while just 26 per cent of those in the Other Mixed group wrote in a mixed response in 1991. Exploiting the possibility for identification with a new mixed category, the largest share of the 1991 Other Black group went to White & Black Caribbean (32.1 per cent). Additionally, over 31 per cent of White & Black Caribbeans, and 25 per cent of White & Black Africans had been in the 1991 Black Other group. Some 1991 Other Asians also moved to the 2001 mixed categories, Other Mixed (4.3 per cent) and White & Asian (2.3 per cent). The largest shift for the residual 1991 ‘Otherʼ group was to the mixed White & Asian category (17.2 per cent). Moves to mixed categories among the White group emphasise the heterogeneity of this majority category. Although they form a small proportion of the 1991 White group as a whole, individuals identifying as White in 1991 represent 29 per cent of 2001 White & Black Caribbeans, 32 per cent of White & Black Africans, 49 per cent of the White & Asian group and 44 per cent of the Other Mixed group. Change to a mixed category was much less likely for people included in one of the Table 3

Mixed ethnic groups in 1991 and 2001

Table 4

Reliability of ethnic group in the 1991 Census: per cent choosing each category in Census Validation Survey

Ethnic group reported at quality check 4–6 weeks after the 1991 Census 1991 Census ethnic group

White

Black

South Asian

Other

All =100 per cent

White Black South Asian Other

99.6 1.5 0.2 9.6

0.1 88.0 0.1 11.3

0.0 1.6 98.7 1.0

0.3 8.8 1.0 78.1

12,017 264 669 130

1991 Census Validation Survey, reproduced from OPCS (1994).12 Each percentage based on weighted sample numbers; cell counts not given in the report.

main South Asian groups, with less than one per cent of 1991 Indians, Bangladeshis or Pakistanis choosing a mixed category in 2001. Over two per cent of people who identified as Chinese in 1991 identified as White & Asian or Other Mixed in 2001. Inconsistency in individualsʼ ethnic identification is an inevitable result of the change in ethnic group question between the 1991 and 2001 Censuses. However, unreliability of ethnic identification also arises from ambiguity that the categories present to some respondents, leading to different responses when asked the same question on different occasions. The 1991 Census Validation Survey identified those who changed their broad ethnic group between the 1991 Census and an interview conducted four to six weeks later, thus measuring reliability directly. The results (Table 4) showed much less consistency for the Black (88.0 per cent) and Other (78.1 per cent) groups than for South Asians (98.7 per cent). The White group had the highest level of consistency at 99.6 per cent.

Percentages 1991 ethnic group 2001 ethnic group White and Black Caribbean White and Black African White and Asian Other Mixed Not Mixed Total

Mixed

Not Mixed

Total

49.7 40.3 33.7 26.1 0.1 0.4

50.3 59.7 66.3 73.9 99.9 99.6

100.0 100.0 100.0 100.0 100.0 100.0

Notes: Includes cases where 2001 ethnic group is imputed; Write-in responses were used to identify respondents as ‘Mixed’ or ‘Not Mixed’ in 1991 Source: ONS Longitudinal Study, authors’ analysis

Table 5

While the ethnic group question will continue to develop in order to reduce the numbers allocating themselves to residual ‘Otherʼ categories, there is evidence that inconsistency in recorded ethnic group is also associated with characteristics that will grow in the population, including being UK-born and in a minority ethnic group. Some countries or regions of birth are more strongly associated with the ethnic group labels than others. For example, Pakistan and ‘Pakistaniʼ ethnicity, China and ‘Chineseʼ ethnicity, and African countries and ‘Black Africanʼ ethnicity. Table 5 shows the percentage of people changing from a 1991 minority ethnic group to another ethnic group other than Mixed in 2001, by whether their country of birth is associated with their 1991 ethnicity. People changing to a Mixed ethnic group are not counted

Per cent changing 1991 ethnic group in 2001 by country of birth

Country of birth in 1991 and ethnic group in 2001 Born in country or region related to 1991 ethnic group 1991 ethnic group

Black Caribbean Black African Indian Pakistani Bangladeshi Chinese

Percentage in same ethnic group or mixed in 2001

Percentage changed group

94.7 94.8 95.5 96.1 97.0 99.3

5.3 5.2 4.5 3.9 3.0 0.7

Notes: Excludes records with ethnic group imputed in 2001 Source: ONS Longitudinal Study


40

Born in UK

Percentage in same ethnic group or mixed in 2001 79.6 73.2 92.1 94.1 94.9 95.0

Born elsewhere


Percentage in same ethnic group or mixed in 2001


20.4 26.8 7.9 5.9 5.1 5.0

54.1 34.8 89.1 81.7 75.0 86.0

45.9 65.2 10.9 18.3 25.0 14.0


as having changed ethnic group because their responses are related to new ethnic categories in 2001. It is clear that people born in a country associated with an ethnic group label have least difficulty identifying with the corresponding ethnic category at a subsequent census. Less than 6 per cent of people from ethnic minorities who were born in countries or regions associated with their 1991 ethnic group changed their ethnic group in 2001. Those born in the UK are generally less consistent in reporting their ethnic group. For example, 20.4 per cent of 1991 Black Caribbeans born in the UK changed their group in 2001, compared with 5.3 per cent of those born in the Caribbean. Those born elsewhere in the world are least consistent of all in their responses to the ethnic group question. Among 1991 Black Caribbeans, 45.9 of those not born in the UK or Caribbean changed their ethnic group in 2001. Thus the consistency of identification with ethnic categories depends on individualsʼ history.

DIFFERENTIAL ATTRITION RATES BY ETHNIC GROUP The updating of the LS from the 2001 Census reveals the differential loss to follow-up by ethnic group, based on those who answered the ethnic group question in 1991 but were not observed in 2001. This can give us some understanding of the coverage of different ethnic groups in the LS. Tracking forward from the 1991 ethnic group categories shows that particular ethnic groups were affected by relatively high rates of loss to follow-up. LS members are said to be ‘lost to follow-upʼ if their records are not found at census, yet there is no record of a prior embarkation or death for them. Loss to follow-up can thus result from census nonresponse, emigration without embarkation being recorded and failure to link death or embarkation records to an LS member. Other losses occur because of respondent or processing error. Between the 1991 and 2001 Censuses, 41 per cent of Black Africans were lost to follow-up. Among those enumerated in the 1991 Census and not known to have ‘exitedʼ through death or emigration in the meantime, people from ethnic minorities comprised 6 per cent of the LS sample found at the later census, compared with 15.9 per cent of those not found.13 In the 1991 Census, Black and Asian ethnic groups were less likely to be enumerated because of their concentration in low coverage areas.14 Studies which use ethnic group measured at either 1991 or 2001, linked to prior records from the 1971 or 1981 Censuses are useful for answering longitudinal questions (assuming stability of ethnic group); but such linkage cannot, by definition, directly provide analysis by ethnic group of those who were lost to follow-up before the ethnic group questions were asked. Nevertheless, loss to follow-up by ethnic group and its implications over the history of the LS can be studied using proxy ethnicity based on country of birth of the LS member or their (co-resident) parents at 1971 or 1981 for cases where ethnic group is not known; and this proxy ethnicity can, additionally, be validated by ethnic group information where it is known, to look at differences in characteristics between those who are lost and those who are not. This type of analysis is presented in the following example, analysing attrition across the life of the LS. The rate of loss to follow-up by 2001 for two cohorts of LS members, those who were children aged 4–15 at either the 1971 or 1981 Census was examined. This research is complementary to a study of ethnic group social mobility,15 and it aimed specifically to examine the potential role of loss to follow-up in shaping observed patterns of intergenerational social mobility across ethnic groups.16 Observation of intergenerational social mobility required that the full thirty years of the LS be used in order to observe people in childhood when co-resident with their parent(s) and then in adulthood. In standard cross-sectional surveys, including those that collect data retrospectively, by definition, we know nothing about those who are not included but who might have been had they not failed to respond or died. In the LS, however, we know quite a lot

Autumn 2005

about those LS members who were observed in 1971 or 1981 but were not observed in 2001. For the children studied here, we know about their parentsʼ country of birth, occupation and age, and the housing tenure and car ownership of the household in which they were living in 1971 or 1981 from data collected on household co-residents of all LS members. Moreover, for those who were still observed at 1991 but not observed in 2001 we know about their 1991 ethnic group. In addition, for some of those not observed at 2001 we know why they were not observed through information on embarkations and deaths linked to the LS. The following analysis uses all these linked data for people observed at several points in the LS history to try and give a more detailed picture of those LS members, observed as children in 1971 or 1981 who were not observed as adults aged 24 to 45 in 2001. This includes both those who were not expected in 2001 because they died or embarked in the interim and those who are genuinely lost to follow-up. Rather than simply focusing on loss to follow-up between 1991 and 2001 where the loss can be related to 1991 ethnic group, this analysis examines loss to follow-up by 2001 across the full life of the study. This is pertinent to longitudinal analysis over an extended period to which the LS is well-suited, and of which the substantive origin of this research – exploring patterns of social mobility – is a clear example. Therefore, for the purposes of this study a proxy ethnicity was created based on parentsʼ country of birth for those cases found in 1971 or 1981 who were not observed in either 1991 or 2001.17 This proxy ethnicity was checked and validated against both an existing 1971 LS proxy ethnicity variable for those observed in 1971, and against ethnic group for those recorded in 1991. Given the age range of the sample (24–45 in 2001), the number of recorded deaths is small (967 by 2001, or 0.5 per cent of the original sample of LS members aged 4–15 in 1971 or 1981 who were living with at least one parent). The number of known emigrations is more substantial (4,258 members by 2001, or 2.4 per cent of the original sample and 11.7 per cent of all those not observed in 2001). Based on notifications from the National Health Service Central Register (NHSCR), which may record expressed intention to emigrate, it is estimated that the LS only captures about half of actual emigration,18, 19 with the other half of emigrants not conclusively identified as such. Instead, they fall into the group for whom disappearance from the observed sample is unexplained (31,099 by 2001 or 17.5 per cent of the original sample), though as well as non-recorded emigrations it will include underenumeration at census.20, 21 Because of the low mortality rates for this sample, there was no systematic exploration of characteristics associated with death by 2001. Instead, focusing on emigrants and those lost to follow-up, the characteristics of the two groups were compared. As some of the unexplained losses could be assumed to be unrecorded emigrations (including moves to Scotland), it is interesting to see if the two groups look similar, especially as emigration could be anticipated to be more likely for those of immigrant parentage; and we already know that minority ethnic groups have higher rates of census non-response.20 In addition, the propensity to emigrate or for unexplained loss to follow-up was modelled, investigating ethnic group effects and controlling for other relevant characteristics, to explore the independent effect of ethnic group. Table 6 compares unexplained attrition and emigration across selected characteristics by contrasting the percentage of people with an attribute in the original sample, with the percentage among those who emigrated, and those who were lost to follow-up. It shows that men are more likely to both emigrate and to be lost to follow-up without explanation; those from higher social class (Service class, in the terminology of the classification system) backgrounds are also more likely to emigrate as they are over represented among emigrants (36.3 per cent) compared with the original sample (26.3 per cent). However, overall there was little

41



Table 6

Autumn 2005

Characteristics of LS members aged 4 to 15 in 1971 or 1981; emigrants and unexplained losses compared Column per cent Not present at 2001 Census Percentage of cohort with this characteristic in 1971 or 1981

Emigrated by 2001

Unexplained losses by 2001

Sex Male Female

51.0 49.0

57.3 42.7

61.2 38.8

Parental class ‘Service’ (professional and managerial) Intermediate Working Other

26.3 18.1 49.8 5.8

36.3 16.7 42.5 4.5

25.2 17.4 49.0 8.4

Ethnic group* (proxy) White non-migrant Caribbean Indian Pakistani White migrant

86.5 1.6 1.4 0.7 3.2

74.8 1.7 1.7 1.2 8.5

77.9 3.8 2.1 1.4 5.5

Characteristic in 1971 or 1981

* Ethnic group columns do not sum to 100 as not all proxy ethnic groups have been included Source: ONS Longitudinal Study, author’s analysis

difference in unexplained loss between social classes, with the exception of a larger than expected number of people from ‘otherʼ social class backgrounds represented among those lost to follow-up. Both emigration and unexplained loss to follow-up are greater for minority ethnic groups than for the white non-migrant group.22 However, there are differences between minority ethnic groups: those from Caribbean backgrounds are over-represented among those lost to follow-up, making up 3.8 per cent of this group compared with 1.6 per cent of the original sample; while those from Indian backgrounds make up 2.1 per cent of those lost to follow-up compared with 1.4 per cent of the original sample. Those from white migrant backgrounds are heavily over-represented among emigrants: they constitute 8.5 per cent of known emigrants compared with 3.2 per cent of the original sample. The next question is to what extent these characteristics overlap: does the over-representation of those from minority ethnic groups among those lost to follow-up and emigrants stem from other characteristics that are also associated with these outcomes? In addition, are there similarities between emigrants and those lost to follow-up? To explore this, logistic regression models were estimated for the probability of emigrating or of unexplained loss to follow-up controlling for a range of characteristics. Alongside proxy ethinicity, these characteristics included, grouped age of the respondent: 4–7, 8–11 and 12–15 when first observed (in 1971 or 1981); whether the LS member was first observed in 1971 or 1981 (cohort); parentsʼ social class; the ethnic minority group composition of the area where the LS member lived in 1971 or 1981; whether or not they themselves were born abroad; whether their father or mother had higher qualifications, and the housing tenure and car ownership of the household in which they were living in 1971 or 1981.23 Table 7 shows the results of these two models. The ethnic group coefficients in Table 7 show that even after controlling for the other characteristics, all the minority groups have a significantly greater probability of both emigration and loss to follow-up than their white non-migrant counterparts, a finding which is consistent with other research. The Indians/ East Africans, however, provide an exception here. This tells us that minority ethnic group members growing up in Britain in the 1960s–1980s have a greater propensity not to be observed in 2001 as a result of both emigration and non-response and that this propensity is independent of other characteristics that might lead to such an outcome. The size of the effects is also substantial for some groups.


42

Translating the coefficients into odds ratios we find that Black Africans have three times the chances of emigrating as their white non-migrant peers and that they have nearly three and a half times the chances of loss to follow-up. Caribbeans have high chances of loss to follow-up taking account of other factors, though their increased chances of emigration are not as substantial as for the other groups (apart from the Indians). After controlling for ethnicity, the independent effect of other characteristics on probabilities of emigration or loss to follow-up at 2001 were not similar across the two outcomes. These findings suggest that while it may well be the case from these findings that many of the unexplained are unobserved emigrants, particularly given the strength of the ethnic group effects, this is not completely clear, and there is no way of of confirming this theory in the analysis.24 Indeed, it seems highly possible that while more privileged origins tend to promote emigration, those lost to follow-up are a mixture of the more advantaged (who emigrate) and the less advantaged who are ‘lostʼ (non-respondents), forming a ‘hard to reachʼ group for the purposes of Census and surveys. While the ethnic group results may not be especially surprising, they do indicate that when we are following up groups over time, observed patterns of ethnic group differences are shaped by differential patterns of loss. This may mean that we end up with a different picture of dynamic processes across ethnic groups than we would without such losses. Moreover, if those who ‘exitʼ have particular relationships to outcomes of interest that cannot be predicted by their background characteristics, then the links between ethnicity, controlled characteristics and outcomes will not be the ‘trueʼ ones. The levels of differential loss to follow-up may also themselves be of interest in relation to substantive questions, for example about geographical mobility. Thus, what difference does differential loss to follow-up make to observed measures of intergenerational social mobility? We may identify more (or less) upward mobility among certain groups than others. However, the presence of high (and differential) levels of attrition raises the question as to whether those patterns are driven by more of the least (or most) successful from that group being lost to follow-up. We went on to explore the substantive implications of these findings on differential loss to follow-up on patterns of social mobility by estimating a Heckman selection probit.25 This model attempted to take account of all those who were lost to follow-up for whatever reason and the effect of this loss on


Table 7

Autumn 2005

Logistic regression of probabilities of emigration and unexplained loss by 2001 controlling for background variables; Coefficients and (Standard Errors)

Explanatory variables

Emigration

Unexplained loss

Cohort (base is 1971)

–.826

(.035)***

.261

(.013)***

Age group (base is aged 12–15 when first observed) Aged 4–7 when first observed Aged 8–11 when first observed

–.045 (.036) –.148 (.039)***

.238 .116

(.015)*** (.016)***

Minority concentration of ward of origin (base is 0 per cent) 0–1 per cent 1–5 per cent 5–10 per cent More than 10 per cent If sample member born abroad

.234 .181 .061 –.037 .765

(.058)*** (.066)** (.092) (.090) (.065)***

.94 .156 .296 .362 .548

(.025)*** (.028)*** (.037)*** (.036)*** (.034)***

Parents’ class (base is working) ‘Service’ (professional and managerial) Intermediate Other Male

.197 (.046)*** .008 (.048) –.096 (.093) –.159 (.033)***

.058 .057 .234 .532

(.020)** (.019 )** (.031)*** (.014)***

–.221 (.138) .175 (.052)***

.384 .069

(.041)*** (.026)**

.179 (.078)* .328 (.049)***

.177 .088

(.028)*** (.024)***

Mother’s qualifications (base no higher qualifications) No co–resident mother Mother with higher qualifications Father’s qualifications (base no higher qualifications No co–resident father Father with higher qualifications Housing tenure at origin (base is owner occupation) Local authority Private rented Car ownership at origin (base is none) 1 car 2 or more cars Proxy ethnic group (base is both parents British born) Caribbean Black African Indian (and East African) Pakistani Bangladeshi Chinese and Other White migrant

–.430 .159

(.046)*** (.049)***

.182 .235

(.017)*** (.022)***

.174 .174

(.044)*** (.059)**

–.128 .039

(.017)*** (.023)

.341 (.134)* 1.138 (.249)*** .269 (.150) .699 (.161)*** .841 (.355)* 1.022 (.131)*** 1.007 (.067)***

1.104 1.239 .339 .636 .405 1.105 .678

(.046)*** (.131)*** (.056)*** (.069)*** (.152)** (.070)*** (.034)***

Constant N

–3.775 (.073)*** 176,907

–2.419 (.032)*** 170,261

Wald chi2 (df)

2,246.7

2,0997

(31)***

(32)***

Notes: *