The development of non-essentialist concepts of ethnicity among children in a multicultural London community.

Ethnic constancy, the belief that a person cannot change ethnicity, is an important component of ethnic essentialism, the conviction that members of ethnic groups share an immutable underlying essence. Most children in previous research viewed ethnicity as increasingly immutable with age. However, some evidence suggests that children growing up in communities, which define ethnicity primarily in terms of changeable features (e.g., lifestyle) rather than fixed features (e.g., ancestry), may not follow this trajectory. This study examined ethnic constancy development in a community which defined ethnicity primarily in terms of changeable features. It was hypothesized that older children would view ethnicity as more changeable than younger children, but that because of personal investment which increases with age, children would view their own ethnicity as more stable than a peer's ethnicity, entailing a significant interaction between age and self-other. Ninety-two children in three age groups (mean ages 7, 9, and 11 years) from a multicultural school in London were interviewed individually. Their ethnicities were 45% Indian, 16% English, 7% Pakistani, 7% Somali, 2% unknown, and 25% other. Children's explanations were analysed thematically. All hypotheses were supported. Children's conceptions of others' ethnicity as changeable were supported by definitions focusing on religion, and by the concept of freedom of choice. This suggests that in a community in which ethnicity is primarily defined in terms of attributes which are seen as mutable (in this case, religion), children may not essentialize ethnicity. Still, ethnic change may rarely occur in practice due to an emotional commitment to one's own ethnic group. Statement of contribution What is already known on this subject? Most research finds that children develop concept of ethnicity as fixed and essential; But limited evidence of non-essentialist developmental pathways for ethnicity For gender, children assert constancy more for themselves than others What does this study add? Evidence of developmental pathway to non-essentializing, mutable concept of ethnicity Evidence that definitions of ethnicity centred on religion may encourage ethnic non-essentialism Children are more willing to embrace ethnic mutability for others than for themselves.

Background

Psychological essentialism is the view that certain categories are characterized by an enduring, underlying essence that determines many of the properties and behaviours exhibited by category members, and renders category distinctions objective and natural (Gelman, 2003, 2009). One such category is ethnicity. While this is often defined very broadly, incorporating ideas about language, religion, cultural practice, and race (Stone, 1996), its essentialized form usually focuses on biological ancestry (Aboud, 1984; Eriksen, 2002; Gil‐White, 1999, 2001; Nesdale, 2004; Waters, 1990; although see Brown, Spatzier, & Tobin, 2010; Haines, 2007).

The current study focuses on one strand of essentialist thinking, the conviction that category membership is constant, permanent, and immutable. This belief has implications for the politics of ethnicity in everyday life (Gil‐White, 1999). For example, strategies people employ to cope with membership of a low status group vary according to whether they see the boundary between their own and a higher status group as permeable (Hogg & Abrams, 1998). Moreover, in some circumstances, the more a person sees group membership as fixed, the more prejudiced against group members the person is likely to be (Haslam, Bastian, Bain, & Kashima, 2006; Morton, Hornsey, & Postmes, 2009), and the more likely they are to stereotype members (Pauker, Ambady, & Apfelbaum, 2010). The import of these effects can hardly be overstated given that in many societies, ethnicity is a key dimension along which people are categorized, affecting social mobility, access to resources (Baumann, 1996; Eriksen, 2002), interpersonal relationships (Baerveldt, Van Duijn, Vermeij, & Van Hemert, 2004; Nukaga, 2008; Schneider, Dixon, & Udvari, 2007; Schofield, 1981), and intergroup conflict (Stone, 1996).

Ethnic essentialism: Universal development or culture specific?

Some social scientists have hypothesized that the emergence of ethnic essentialism (including constancy) during childhood is innately specified, in the form of a tendency to process endogamous descent‐based social groups as if they were species (Gil‐White, 2001), or a drive to seek out the ‘human kinds’ of significance in the child's culture (Hirschfeld, 1996, 1997). In support of this hypothesis, there is evidence that children in a range of cultures increasingly essentialize ethnicity as they get older (Diesendruck & haLevi, 2006; Gelman, 2003). Further evidence for this developmental trajectory comes from the ethnic constancy tradition, based on Kohlberg's (1966) seminal work on gender constancy. Studies in this tradition have consistently found that as they get older, children in the USA, Europe, and Canada increasingly view ethnicity (Aboud, 1984; Aboud & Skerry, 1983; Bernal, Knight, Garza, Ocampo, & Cota, 1990; Ocampo, Knight, & Bernal, 1997; Ruble et al., 2004) and race (Aboud & Doyle, 1995; Bales & Sera, 1995; Giménez & Harris, 2002; Pauker et al., 2010; Rutland, Cameron, Bennett, & Ferrell, 2005; Semaj, 1980) as immutable properties of the person, and they continue to do so as adults (Haslam, Rothschild, & Ernst, 2004).

More recently, however, a few studies have also found evidence for cultural diversity in ethnic essentialism development. Birnbaum, Deeb, Segall, Ben‐Eliyahu, and Diesendruck (2010) found that religious Jewish Israeli children attribute higher levels of inductive potential to ethnicity than do secular Israeli or Arab Israeli children. Diesendruck, Goldfein‐Elbaz, Gelman, Rhodes, and Neumark (2013) found that the rate at which ethnic essentialism emerges was different for American versus Jewish Israeli children. In Rhodes and Gelman's (2009) study of children growing up in urban and rural communities in Michigan, USA, 5‐ and 7‐year‐olds in both communities, and 10‐ and 17‐year‐olds in the urban community saw ethnicity as a relatively conventional category. The older children in the rural community, however, saw ethnicity as natural and objective (i.e., essential), suggesting that culturally variable input (such as the use of generic language terms to refer to ethnic group members) was required for the development of essentialist beliefs about ethnicity (Gelman, 2009; Rhodes, Leslie, & Tworek, 2012).

With respect to the constancy aspect of ethnic essentialism, ethnographic research reveals that many adults in certain societies espouse views consistent with a mutable, non‐essentialized definition of ethnicity. Specifically, 47% of Torguud Mongolian nomadic pastoralists (Gil‐White, 1999) and over 75% of Vezos of Madagascar (Astuti, 2001; Astuti, Solomon, & Carey, 2004) assert that a child who is adopted at birth by members of a different ethnicity will acquire the ethnicity of the adoptive parents. This view is partially held (with respect to only some ethnic groups) by most Vezo children aged between 6 and 13 years too (Astuti et al., 2004). For the Vezo, this concept of ethnic mutability seems to be based on their oft‐spoken performative theory of ethnicity, whereby one becomes Vezo by virtue of where one lives and (therefore) what one does, rather than of what or where one is born. For instance, Vezo people live on the coast and hence, learn to fish well. This theory is also put into practice, with frequent interchange between the Vezo and the members of another local ethnic group (Astuti, 1995).

The current study

The Madagascar research suggests that children growing up in communities, which define ethnicity in terms of cultural or behavioural attributes they consider changeable, may be less likely to construct a concept of ethnicity as fixed and essentialized compared with people whose definitions focus around attributes they consider fixed (such as ancestry). However, as yet, no study has demonstrated that in such a community, children's concepts of ethnicity follow a trajectory towards mutability. The current study explores this possibility among 6‐ to 11‐year‐old children growing up in a multicultural community in London, selected because prior ethnographic research found that children and adults in that community often defined ethnicity in terms of apparently changeable attributes: food, films, clothing, religion, and language (Woods, 2005; R. Woods, unpublished data).

The study's methodology is based on Slaby and Frey's (1975) adaptation of Kohlberg's (1966) gender constancy methodology, which includes three components. The first, ethnic identity (ability to correctly label self and others), was included as a check that children recognized the terms for ethnicity employed in the study and generally applied these terms in ways consistent with official school records of ethnic identity. The second component was ethnic stability, which assessed the belief that ethnicity is unchanging over time (past and future). This component provided the main, most generic measure of ethnic constancy. The third component, ethnic consistency, assessed children's beliefs that ethnicity is unchanging over a series of specific transformations, thus providing additional insights into children's concepts.

The three main hypotheses all concerned the ethnic stability component. Based on ethnographic data indicating that at school, children mainly experienced ethnicity in terms of changeable features (Woods, 2005; R. Woods, unpublished data), which may support the development of a performative theory of ethnicity (Astuti, 1995), it was hypothesized that older children in this community would view ethnicity as less stable than younger children (H1).

Previous studies of ethnic constancy (EC) asked children about either themselves (e.g., Aboud & Skerry, 1983) or an unknown other (e.g., Aboud, 1984). The current study is the first to include ethnic stability questions for both self and other. Children assert gender constancy for themselves earlier than for other people (Aboud & Ruble, 1987; Marcus & Overton, 1978), which may be because they feel more threatened when their own gender identity is questioned (Szkrybalo & Ruble, 1999). Ethnic identity is positively correlated with self‐esteem among minority children (Phinney, 1992), and British schools encourage children to take pride in their ethnic heritage (Baumann & Sunier, 2004); therefore, it was predicted that children would view their own ethnicity as more stable than a peer's ethnicity (H2).

These predictions are complicated by evidence that ethnic identity tends to become more important to children as they get older (Phinney, 1989), which is likely to mean that with age, children will become increasingly unwilling to consider changing their own ethnicity. This tendency may counteract the age‐based decrease in stability (predicted in H1) for those items concerning children's own ethnicity. Therefore, a significant interaction between the main effects of age and subject of question is also predicted (H3). Specifically, it was anticipated that while ethnic stability scores for an unknown peer would decline with age (as children construct a performative theory of ethnicity from their school experiences), ethnic stability scores for the self would remain more constant (as children's developing performative theory of ethnicity is countered by their increasing emotional investment in their own ethnic identity with age). Consequently, the gap between self and other ethnic stability scores was expected to be greater for the older than the younger children.

While the stability component of EC was used to test hypotheses regarding children's overall commitment to ethnic constancy, the final, consistency component addressed the question of how the children actually defined ethnicity, and specifically, whether they based their concepts on mutable attributes, as seemed to be the case among the Vezo (Astuti, 1995). Previous ethnic constancy studies have included remarkably few attributes in the consistency component, usually only clothing (Aboud, 1984; Aboud & Skerry, 1983) and physical appearance such as hair colour (e.g., Bernal et al., 1990). These studies found that with increasing age, children invariably came to see these changeable attributes as irrelevant to ethnic identity. Because studies have neglected other potentially relevant changeable attributes, such as religion, food, and language, we do not know whether children ever see changes to these attributes as defining of, and therefore efficacious in transforming ethnicity. This study is the first to include a broad range of six transformations: clothing, food, and language were included to assess the centrality of cultural attributes to children's definitions of ethnicity; skin colour enabled an assessment of whether children saw ethnicity primarily as a racial issue; religion was included in recognition of the important role it can play in people's understanding of ethnicity (Eriksen, 2002), and marriage to provide insight into whether children considered ethnicity to change via relationships. This broad range increases the possibility of uncovering a transformation that is seen as efficacious, thus providing more insight than previous studies into children's ethnic concepts.

Method

Participants

Consent was sought from the parents of 270 children in years 2, 4, and 6 at a large primary school in west London in the United Kingdom, situated in a socially and economically deprived area. The 92 children whose parents gave consent were interviewed individually: 14 boys and 16 girls in year 2 (M age = 7 years, 4 months; SD = 2.76 months), nine boys and 20 girls in year 4 (M age = 9 years, 4 months, SD = 3.50 months), and 14 boys and 19 girls in year 6 (M age = 11 years, 4 months, SD = 3.75 months). Throughout this article, terms for ethnicities are those used in school records. Ethnicities of the children were 45% Indian, 16% English, 7% Pakistani, 7% Somali, 2% unknown by the school, and 25% a range of other ethnic groups.1 Proportions were similar across year groups in the sample: 50% Indian, 20% English, 3% Pakistani, 7% Somali, 20% other for year 2; 48% Indian, 10% English, 10% Pakistani, 3% Somali, 27.6% other for year 4; 36% Indian, 15% English, 9% Pakistani, 9% Somali, 30% other for year 6. Thus, the Indian population was somewhat smaller for year 6 than the other year groups. Proportions in the school overall were described by the head teacher as approximately 30% Indian, 14% English, 14% Somali, 7% Pakistani, and the remainder drawn from around thirty other ethnic groups. Thus, the sample as a whole did not entirely reflect the school population overall, with an over‐representation of Indian children and under‐representation of Somali children.

Procedure

The study was approved by Canterbury Christ Church University's research ethics committee. The researcher (of English ethnicity) interviewed participants individually in a quiet room. All interviews were recorded and transcribed in full. At the outset, children were invited to participate and advised that they could withdraw at any time. At the end, younger children were offered a sticker as thanks. Interviews focused on the ethnicities Indian, Somali, and English, because they were the most common ethnicities at the school, and earlier participant observation at the school revealed that children often used these terms themselves. The order of the three components (identity, stability, consistency) was counterbalanced as far as possible across four versions (limited by the need for the identity component to precede the stability component, which referred to children's own ethnicity).

Measures

Ethnic identity

The child was shown a list of the names of all children in their year. As a warm up, they were asked to name peers with short and long hair. The researcher then said, ‘Some of the children in your year are English, some of them are Indian, some of them are Somali, and some of them are something different again. Can you think of someone in your year who is English?’ Having chosen a name, the child was asked how they knew that child was English. The same questions were asked for peers who were Indian and Somali, with the order counterbalanced across versions of the interview schedule. Then, the child was asked, ‘Are you English, Indian, Somali or something else?’ If the child answered with a term that was different from school records, the researcher asked whether they were the ethnicity on school records.

Children received one point for each correct identification of a peer's or their own ethnicity, based on school records, yielding an identity score out of four. Children who only identified with their own ethnicity after prompting received half a point. Identity scores could not be calculated for two children whose ethnicity was unknown on school records, nor for two others who named a peer whose ethnicity was unknown.

Ethnic stability

The ethnic stability component was made up of four sets of four questions: self‐past, self‐future, other‐past, and other‐future. The order in which these were presented was counterbalanced across four versions of the interview schedule.

The self‐past questions were as follows: (1) ‘When you were a baby, were you X or something else?’ where X was the child's self‐reported ethnicity (stated in the Identity section). (2) ‘Were you ever Y do you think?’ (3) ‘Were you ever Z do you think?’ where Y and Z were two of English, Somali, and Indian (but not the child's own ethnicity). (4) ‘Why do you think that? How do you know?’ The self‐future questions were as follows: (1) ‘When you grow up, will you be X or something else?’ (2) ‘Could you change into a Y person if you wanted to?’ (3) ‘Could you change into a Z person if you wanted to?’ (4) ‘Why do you think that? How do you know?’ Question (4) was asked only once for each block. The other‐past and other‐future blocks were identical except that they asked about an unknown local peer of the same gender and ethnicity as the child.

For questions 1 to 3 in each set, one point was awarded for the claim that the child concerned would remain the same ethnicity. The self‐past and future scores were combined to give a self‐stability score out of 6, and the same was done for other‐past and future questions.

Stability scores and could not be calculated for 16 children who did not identify with their ethnicity on school records in the identity component, and for two additional children whose ethnicity was unknown. One participant was not asked stability questions because of experimenter error, and two more answered incomplete versions because they seemed distressed by the questions. Therefore, complete stability data were available for 71 children: 12 boys, 13 girls in year 2; 8 boys, 14 girls in year 4; 11 boys, 13 girls in year 6. Those children with ‘other’ ethnicities were disproportionately excluded (because they most often did not identify with their ethnicity as specified on school records), so that the ethnic composition of each year group for the stability sample was 50% Indian, 20% English, 7% Somali, 3% Pakistani, 20% other (for year 2); 55% Indian, 14% English, 0% Somali, 9% Pakistani, 23% other (for year 4); and 50% Indian, 21% English, 9% Somali, 13% Pakistani, 9% other (for year 6).

Ethnic consistency

There were two sets of questions, on Indian‐English and Indian‐Somali transformations. The order of the two sets, the direction of the transformation (e.g., Indian to English vs. English to Indian), and the order of the transformations were counterbalanced across four interview schedules.

The researcher stated, ‘Now I'm going to tell you about some children who live around here, tell you about their ideas, and ask you what you think about their ideas’. For the English‐Indian set she continued, ‘These questions are gonna be about some X girls/boys [depending on gender of interviewee] who live round here. Each of them is thinking about something different’, where X was either Indian or English, depending on the version. Children were then told six transformations applied to six different people: ‘One X girl/boy is thinking of…’ followed by ‘putting on some Y clothes’, ‘learning to speak a Y language’, ‘painting her/his skin a different colour’; ‘eating some Y food’, ‘changing her/his religion’, and ‘marrying someone who is Y when (s)he grows up’. For each item, the child was asked, ‘If (s)he did that, would (s)he still be X or would (s)he become Y?’ where Y was English or Indian, depending on the version, and ‘Can you tell me why you think that?’ The Indian‐Somali block was identical with the words ‘Indian’ and ‘Somali’ substituted for X and Y.

Children gained one point every time they asserted that the protagonist's ethnicity would remain the same. This yielded a score out of two for each transformation (combining the English‐Indian and Indian‐Somali blocks).

Qualitative analyses

The reasons that children gave for their answers (both spontaneously and in response to the ‘why’ questions) for the stability and consistency components were coded by the author. Explanations were excluded if the child said that they did not know the answer to the question, or answered that the protagonist would be both ethnicities. The coding categories, or themes, were generated using thematic analysis, following Braun and Clarke's (2006) procedure. All themes representing <3% of all coding for each component were removed by either combining them with other categories or putting them into a catch‐all ‘other’ category. Some explanations incorporated more than one theme; all were coded. The codes are listed in Tables A1 and A2 in the Appendix, with examples.

Table A1

Coding scheme for children's explanations of answers to stability questions

Code	Description	Examples
Reference to religion	Reference to religion in general, or to specific religion(s)	‘Because like God makes you and He gives you a religion, any religion He likes that then He, when he gave you that religion, you're a baby for that, you're an adult for that, and you're a toddler for that. You can never change anything’. (Indian boy, year 2, explaining why an Indian child was also Indian as a baby)
Reference to language	Reference to language in general, or to specific language(s)	‘As she was growing, growing she started, she might start speaking Indian but not English or Somalian because she wouldn't know’. (Indian girl, year 2, explaining why an Indian child was also Indian as a baby)
Reference to skin colour	Reference to skin colour in general, or to specific skin colour(s)	‘If she was um Indian or Somali she would um, she would have a different colour of her skin’. (English girl, year 2, explaining why an English child was also English as a baby)
Where or what protagonist was at birth/as baby	Reference to where or what protagonist was at birth/as a baby	‘Because he was born in England’. (English boy, year 6, explaining why an English child was also English as a baby)
Reference to protagonist's family/marriage	Reference to protagonist's family in general or specific family member(s) or protagonist's marriage	‘I wouldn't want to make my family upset, and be the odd one out of my whole family’. (Somali girl, year 6, explaining why she could not become Indian or English in the future)
Evidence (e.g., photographs, personal observation)	Citation of evidence to support statement: photographs, videos, testimony, and/or witnessing another person	‘I have videos of when I was a kid’. (Indian boy, year 4, explaining why he was Indian as a baby)
Protagonist prefers one ethnicity	Reference to which ethnicity the protagonist prefers/wants to be	‘Yes but I wouldn't never want to’. (Indian girl, year 4, in response to question, ‘Could you change into an English person if you wanted to?’)
Protagonist free to choose own ethnicity	Claims that protagonist is free to choose/to be what they want/to make their own decision regarding their ethnicity	‘If he wants to, it's his choice.’ (Pakistani boy, year 4, explaining why a Pakistani boy can change to be Indian)
Child cannot explain answer	Child says that they do not know why they gave the answer they did
Explanation fits none of the above	None of the above codes apply to any part of the explanation. This code was only applied if the whole explanation was codable. If only part of it was codable, that part was ignored	‘He can change his passport, he can move to places and then get like a visa is it, so then he'll become a citizen of there.’ (English boy, year 6, explaining why an English child could become Indian or Somali in the future)

Table A2

Coding scheme for children's explanations of answers to consistency questions

Code	Description	Examples
Change in question is irrelevant to ethnicity	Claims that attribute in question does not define ethnicity OR changing that attribute does not change ethnicity OR attribute is not exclusive to specific ethnic groups OR changing the attribute leads to superficial changes that do not alter ethnicity OR the attribute in the question is a necessary but not sufficient definer of ethnicity	‘Even she's got those clothes on, she's still she's still Indian because that's just like dressing up’. (Indian girl, year 2, on why an Indian child wearing English clothes would still be Indian)
Change in question is relevant to ethnicity	Claims that attribute in question defines or is synonymous with ethnicity OR changing that attribute changes ethnicity	‘He changed his religion and um I'm Christian, if I changed my religion to like Muslim now I'll have to be like Somalian’. (Black African boy, year 6, on why an English boy who changed his religion would become Indian)
Protagonist's origin is relevant to ethnicity	Claims that where or what one was at birth/as a baby OR where one is from OR one's blood is defining of or synonymous with ethnicity	‘That's how he was born’. (English boy, year 2, on why an Indian boy would remain Indian even if he changed his religion)
Religion is relevant to ethnicity	Claims that religion is defining of or synonymous with ethnicity. Only coded when religion not mentioned in question	‘Doesn't matter about what clothes you wear, you'll still follow the same religion’. (Indian boy, year 6, on why an Indian child wearing English clothes would remain Indian. This explanation was also coded as ‘irrelevant attribute’.)
Language is relevant to ethnicity	Claims that language is defining of or synonymous with ethnicity	‘Because if she still remembers the English words and she might know almost um a zillion words and then after she might know all the words and that might show her that she really wants to be a English’. (Somali girl, year 2, on why an English girl who changed her religion would remain English)
Various other attributes are relevant to ethnicity	Skin colour/clothing/food/place of residence and/or family is defining of or synonymous with ethnicity. Only coded when the attribute is not mentioned in question	‘Because of the colour of his skin, and if he's still dressed up as an English person he'll still be that colour.’ (English boy, year 2, on why an Indian child wearing English clothes would remain Indian)
Impossible to change attribute and/or ethnicity	Claims that it is not possible/desirable to change the attribute as described in the question OR to change one's ethnicity	‘You can't just change for a different reason cos you want to be, you have to believe in yourself, you have to be happy with which religion you have’. (girl with mixed ethnicity, year 2, on why an English child who changed religion would remain English)
Freedom and personal preference	Claims that it is the protagonist's choice to change their ethnicity OR protagonist can change ethnicity if s/he wants OR protagonist wants/doesn't want to change their ethnicity OR prefers own/other ethnicity	‘It's up to him, if he wants to’. (Arab boy, year 6, on why an Indian boy who changed his religion would become Indian)
Evidence	Reference to evidence, including child's own experiences, observations and testimony	‘Cos I only put some Chinese clothes and on the Roman field trip, it's on Monday we went on the Roman field trip, we had to go in this room, and we had to wear Romans, Roman clothes, but still I'm no Roman, I'm not Roman’. (Nepali boy, year 4, on why an Indian boy who wore Somali clothes would remain Indian)
Child cannot explain answer	Child says that they do not know why they gave the answer they did
Explanation fits none of the above	None of the above codes apply to any part of the explanation. This code was only applied if the whole explanation was codable. If only part of it was codable, that part was ignored	‘Cos it might be itchy and not, and uncomfortable’. (Indian girl, year 4, on why an English girl who wore Indian clothes would remain English)

For the stability component, the analysis included the reasons given by children who gave constancy answers to all three questions in a set with those who gave mutability answers to two or three of the questions in each set (because very few gave mutability answers to all three questions in any set). For children who gave only two mutability answers, their reason for the other answer was not coded.

For the consistency component, explanations for the questions about clothing and religion were coded. Explanations for the other questions were not analysed because, as will be seen later, children's views of constancy on these questions were very similar to the clothing question. Only explanations for the English‐Indian question were used. The Indian‐Somali question was not used because there was no significant difference in children's answers to these two questions, and because children often gave the same explanations for their answers to both English‐Indian and Indian‐Somali questions.

Inter‐rater reliability analyses using the kappa statistic were performed on a random selection of explanations. If only one of the two coders coded a particular explanation with more than one code, the code for the other coder was recorded as ‘missing,’ which was given its own code. Hence, such cases were recorded as mismatches in the usual manner.

For stability, 69 explanations (30% of total explanations provided) were second coded, yielding 126 codes. Fifty‐four explanations for consistency questions (32% of total explanations provided for clothing and religion questions) were randomly selected, yielding 76 codes. There was substantial inter‐rater agreement; for stability, κ = .693, p < .001; for consistency, κ = .669, p < .001.

Results

Analyses of children's scores on the identity measure are presented first, as a check that the children applied ethnic terms in a similar way to school records. These are followed by analyses of the stability data, pertaining to the three main hypotheses regarding age differences in self‐ and other‐stability scores. Next, children's consistency scores for six different attributes were analysed by age, to assess whether the children were incorporating any of the attributes into their definition of ethnicity as they got older. Finally, a multiple regression model for children's views on the stability of an unknown peer's ethnicity in the future is reported.

Mauchly's and Levene's tests demonstrated that the assumption of homogeneity of variance was not violated in any of the analyses, unless indicated otherwise below. For the multiple regression, assumptions of independence of residuals (Durbin–Watson statistic = 2.30), normality (assessed by Q–Q plot), linearity, and homoscedasticity (assessed by partial residual plots and a plot of studentized residuals against predicted values) were met. There were no Cook's distance values above 1, and no outliers (defined as standardized residuals exceeding ± 3 SDs). There were three data points with slightly high leverage (ranging from .221 to .322). Regressions were conducted both with and without these data points. Minimal differences were found, so the full data set was used.

Initial independent‐samples t‐tests found no significant gender differences in identity, stability, and consistency scores. Preliminary one‐way ANOVAs found no significant differences between ethnic groups (categorized as Indian, English, other; there were insufficient numbers of any other ethnic group to be entered into the analysis) for the identity and stability components, nor for any transformation in the consistency component with one exception, reported in more detail below.

Identity

To test whether children of all ages were competent in identifying their own and their peers’ ethnic identities, a one‐way between‐subjects ANOVA was conducted. This found that the difference between age groups was not significant, F (2, 85) = 2.356, p = .101, η2 = .053, power = .465. Table 1 reveals that children of all ages were knowledgeable about ethnicity, with mean scores above 75% for all year groups. Children's mean success rate in identifying a classmate of English ethnicity was particularly high at 86%; the corresponding figures for Indian, Somali, and self‐identification were 78%, 78%, and 76%, respectively. Children's explanations for how they knew the ethnicities of their peers usually referred to religion, language, and/or skin colour.

Table 1

Children's ethnic identity, stability, and consistency percentage scores by school year group

Component	Year 2		Year 4		Year 6
	M	SE	M	SE	M	SE
Identity (N = 28, 29, 31)	76.3	4.33	76.7	4.29	86.3	2.41
Stability (N = 25, 22, 24)
Self	84.7	3.84	77.3	3.56	79.2	3.78
Other	76.7	5.44	62.1	6.02	51.4	6.72
Consistency (N = 30, 29, 33)
Clothing	78.3	7.06	91.4	5.01	98.5	1.52
Skin colour	83.3	6.49	82.8	6.22	92.4	4.42
Food	78.3	6.65	93.1	4.10	98.5	1.52
Language	78.3	7.06	82.8	6.22	95.5	3.35
Religion	43.3	8.21	17.2	6.69	31.8	7.48
Marriage	66.7	8.42	41.4	8.26	72.7	7.57

Stability

To test the hypotheses that (1) older children would view ethnicity as more unstable than younger children; (2) children would view their own ethnicity as more stable than a peer's ethnicity; and (3) these two main effects would interact, and a 3 (age group) × 2 (subject of stability questions) mixed ANOVA was conducted, with repeated measures on the second factor.

All three hypotheses were supported. There was a significant main effect of age (H1), F (2, 68) = 3.351, p = .041, η2 = .090, power = .615. Bonferroni post‐hoc tests found that year 6 saw ethnicity as less stable than year 4 (p = .043). There was also a significant main effect of subject of question (H2), with children asserting the stability of their own ethnicity significantly more often than that of an unknown peer, F (1, 68) = 35.327, p < .001, η2 = .342, power = 1.000. This difference was clearest among children in year 4, F (1, 21) = 11.290, p = .003, η2 = .350, power = .893, and year 6, F (1, 23) = 21.495, p < .001, η2 = .483, power = .993. It approached significance for year 2, F (1, 24) = 3.815, p = .063, η2 = .137, power = .466.

Finally, there was a significant interaction between age and subject, F (2, 68) = 4.231, p = .019, η2 = .111, power = .723. As predicted, stability scores declined with age only when children were asked about a peer, F (2, 69) = 4.512, p = .014, η2 = .117, power = .752; there were no significant age differences in children's stability ratings for themselves, F (2, 68) = 1.061, p = .352, η2 = .030, power = .228. Mean self‐stability scores were high (above 75%) for all age groups, while other‐stability scores steadily declined with age, from 76.7% in year 2 to 51.4% in year 6. See Table 1 and Figure 1 for means and standard errors.

Figure 1

Percentage ethnic stability scores for self and other broken down by year group.

To investigate this effect further, children's scores for ‘other future’ and ‘other past’ components were analysed separately. A one‐way ANOVA found significant differences between year groups’ ‘other future’ scores, F (2, 69) = 7.982, p = .001, η2 = .188, power = .948. Bonferroni post‐hoc tests established that year 2 children scored significantly higher than year 4 and year 6 (both p = .003). Mean scores out of three were 2.24 for year 2, 1.22 for year 4, and 1.21 for year 6 (SDs = .93, 1.09 and 1.10, respectively). There were no significant age group differences in ‘other past’ scores, F (2, 69) = 2.111, p = .129, η2 = .058, power = .419.

Children's explanations, summarized in Table 2, help us to understand why children asserted constancy for themselves more than for others. Almost half of children justified their claim that they were the same ethnicity as a baby with reference to Evidence (e.g., videos, photographs); no child mentioned this for an unknown peer. Thus, there was a sense of certainty about their own past which they did not hold about an unknown other. Among children asserting that they would have the same ethnicity as an adult, 18.2% mentioned their Preference for their own ethnicity, compared to only 8.7% mentioning this for an unknown peer. Furthermore, more than a quarter of those children who said that their ethnicity could change in the future nevertheless asserted that they would not want it to because they preferred their own ethnicity. Children mentioned Freedom of choice more often when talking about an unknown peer's future ethnicity than when talking about their own (mentioned by 46% and 19%, respectively, of children claiming that a peer's or their own ethnicity could change). Freedom of choice was mentioned by three children in year 2, six children in year 4, and 10 children in year 6. These explanations reveal children's commitment to their own ethnic group. This commitment was often related to family belonging and control; for instance, ‘I think that some parents are strict with their children and I think they might not wanna talk to them or not really get in contact with him if he changed his religion [sic] because I don't think the whole family would change with him’.

Table 2

Children's explanations for constancy and majority mutability answers to stability questions

Explanation	% of children who gave this explanation
	Self‐past		Self‐future		Other past		Other future
	Constancy (N = 61)	Mutability (N = 3)	Constancy (N = 33)	Mutability (N = 21)	Constancy (N = 42)	Mutability (N = 13)	Constancy (N = 37)	Mutability (N = 23)
Reference to religion	21.3	0	30.3	61.9	35.7	38.5	47.8	29.7
Reference to language	18.0	33.3	18.2	4.8	9.5	15.4	17.4	13.5
Reference to skin colour	11.5	0	15.2	9.5	9.5	0	13.0	0
Where/what protagonist was at birth/as baby	4.9	0	12.1	4.8	14.3	7.7	26.1	0
Reference to protagonist's family/marriage	62.3	66.6	15.2	23.8	19.0	38.5	26.1	37.8
Evidence	47.5	0	6.1	0	4.8	0	0	2.7
Protagonist prefers one ethnicity	3.3	0	18.2	28.6	0	0	8.7	8.1
Protagonist free to choose own ethnicity	0	0	0	19.0	0	7.7	0	45.9
Child cannot explain answer	4.9	0	6.1	9.5	16.7	0	4.3	0
Explanation fits none of the above	3.3	33.3	6.1	0	16.7	23.1	4.3	10.8

Consistency

The consistency analyses explored whether any of six different attributes were incorporated into children's definitions of ethnicity with age, such that transforming that attribute would change one's ethnicity. A paired‐sample t‐test found no significant difference between children's consistency scores for the English‐Indian and the Indian‐Somali transformation questions, so these were combined to yield a score out of two for each transformation. In almost all cases, variance differed significantly between age groups and between ethnic groups, so nonparametric tests were used.

Kruskal–Wallis tests with two‐tailed p values were conducted with age as the independent variable and consistency scores for each of the six transformations as dependent variables, to explore children's perceptions of ethnic consistency in the face of various transformations. Significant results were followed up with pairwise Tamhane's T2 post‐hoc tests. Means and standard deviations are presented in Table 1. Ethnic consistency increased significantly with age for clothing, χ2 = 8.110, df = 2, p = .017; food, χ2 = 10.068, df = 2, p = .007 (in both cases, year 6 scored significantly higher than year 2, p = .026 and .017, respectively); and marriage, χ2 = 8.450, df = 2, p = .015 (year 6 scored significantly higher than year 4, p = .021). There were no significant differences between age groups for skin colour, χ2 = 2.428, df = 2, p = .297 and language, χ2 = 5.298, df = 2, p = .071. Year 6 children's consistency scores were close to ceiling for clothing, food, language, and skin colour.

Children's views on religious transformations followed a different trajectory. There was a significant age difference, χ2 = 6.138, df = 2, p = .046, with year 2 children gaining significantly higher scores than year 4, p = .050. Mean scores were relatively low for all age groups: 43%, 17%, and 32% for years 2, 4, and 6, respectively. Many children (40% of year 2, 62% of year 4, 48% of year 6) asserted for both English‐Indian and Indian‐Somali transformations that if the protagonist changed their religion, their ethnicity would also change.

Kruskal–Wallis tests compared the views of children of differing ethnic groups. There were reasonable numbers of English (six, four, and five in years 2, 4, and 6 respectively) and Indian (15, 13, and 12 in years 2, 4, and 6, respectively) children in the sample, but not of any other ethnic group, so the remaining pupils were all placed in a third category, ‘other’ (9, 12, and 16 in years 2, 4, and 6, respectively). There were no significant differences between ethnic groups for clothes, χ2 = 2.555, df = 2, p = .279; skin colour, χ2 = 5.538, df = 2, p = .063; food, χ2 = 1.865, df = 2, p = .393; language, χ2 = 2.205, df = 2, p = .332 or marriage, χ2 = 5.299, df = 2, p = .071, but there was for religious transformations, χ2 = 12.816, df = 2, p = .002. Tamhane's T2 post‐hoc tests found that Indian children asserted that religious change would lead to ethnic change significantly more often than did English children (p = .008) and children of other ethnicities (p = .039). Indian children's mean consistency score for the religious items was just 14.6% (SD = 30.1) compared with 60.0% for English children (SD = 47.1) and 38.2% for children of various other ethnicities (SD = 46.2).

To understand why so many children believed that religious change would lead to ethnic change, the reasons they gave for their answers to the English‐Indian clothing and religion questions were analysed; see Table 3. Fifteen children, which was around one‐fifth of all those who gave constancy answers for the questions about clothing, stated that it was religion (rather than clothing) that would need to change in order to alter ethnicity. For instance, one child commented, ‘Even if you're English you can still wear Indian clothes but you'll still be English. Doesn't matter about what clothes you wear, you'll still follow the same religion’ (coded as both ‘change in question is irrelevant’ and ‘religion is relevant’ themes). Around half (47%) of the children giving ‘religion is relevant’ type answers for this question were of Indian ethnicity; none were English.

Table 3

Children's explanations for constancy and mutability answers to consistency questions

Explanation	% of children who gave this explanation
	Clothing question		Religion question
	Constancy (N = 79)	Mutability (N = 9)	Constancy (N = 33)	Mutability (N = 49)
Change in question is irrelevant to ethnicity	68.4	0	21.2	8.2
Change in question is relevant to ethnicity	0	66.7	15.2	77.6
Protagonist's origin is relevant to ethnicity	7.6	0	9.1	0
Religion is relevant to ethnicity (only coded when religion not mentioned in question)	19.0	0	0	0
Language is relevant to ethnicity	5.1	0	15.2	8.2
Various other attributes are relevant to ethnicity	5.1	0	9.1	4.1
Impossible to change attribute and/or ethnicity	2.5	0	24.2	0
Freedom and personal preference	0	0	3.0	4.1
Evidence	22.8	11.1	0	0
Child cannot explain answer	2.5	11.1	12.1	0
Explanation fits none of the above	5.1	11.1	15.2	6.1

For the religion questions, more than three‐quarters of children who gave mutability answers justified these by stating that changing one's religion would lead to a change of ethnicity. In fact, children often answered in a way that suggested that they saw religious and ethnic terms as equivalent or inseparable, such that the question involved a truism. For instance, explanations for why an English child who changed religion would become Indian included ‘You just told me that somebody wants to change religion, so if they want to change their religion they'll be a different religion’ and ‘I think she'll become Indian because if she changed her whole religion she won't become English anymore, she'll become Indian’. Note also that children often mentioned religion when explaining their answers to the stability questions, which did not ask about religion or indeed any mechanism of transformation (see Table 2)—further evidence of the centrality of religion to their concepts of ethnicity.

Regression model

The section on stability above established that as they got older, children claimed increasingly that an unknown peer could change their ethnicity when they grew up. To throw light on the reasoning behind this viewpoint, a multiple regression was conducted using the enter method, with ‘other future stability’ as the outcome variable. Predictor variables were age and consistency scores for religion, marriage, clothes, language, and skin colour. Food consistency was highly correlated with clothes (.758) and language (.658), so was excluded.

The model predicted 29% of the variance in children's other future stability scores, F (6, 65) = 5.752, p < .001, adjusted R 2 = .286. Age and consistency scores for religion, marriage, and clothes added significantly to the prediction (p = .008, .027, .013, .021, respectively, but when high leverage data points were excluded, clothing consistency's contribution only approached significance; p = .056). Regression coefficients and standard errors are reported in Table 4. ‘Other future stability’ scores were significantly positively associated with religion and marriage consistency scores. Thus, children's assertions that ethnic change was possible appeared to be partly based on their conviction that religious conversion and/or interethnic marriage would alter a person's ethnicity. The significant negative relationship with clothing consistency scores indicates that children were not basing their beliefs in ethnic mutability on clothing‐based transformations. There was also a significant negative relationship between age and other future stability score, indicating that age changes in transformation scores did not entirely explain the age‐based increase in ethnic instability.

Table 4

Summary of multiple regression model for ‘other future stability’ scores

Variable	B	SE B	β
Intercept	4.159	.709
Age (months)	−0.017	.006	−.297a
Clothes consistency score	−0.643	.271	−.300a
Marriage consistency score	0.353	.139	.287a
Religion consistency score	0.338	.149	.253a
Skin colour consistency score	−0.258	.217	−.148
Language consistency score	0.166	.208	.097

Note. B = unstandardized regression coefficient; SE  = standard error of the coefficient; β = standardized coefficient.

p < .05.

Discussion

The current study is the first to find evidence that children in specific communities follow a developmental trajectory towards a concept of ethnicity as mutable, at least during middle childhood. With respect to an unknown peer, children's ethnic stability scores decreased significantly with age, with most 10‐ and 11‐year‐olds asserting the possibility of ethnic change at least once.

Previous research suggested that a concept of ethnicity as mutable might be arrived at in communities which operationalized ethnicity primarily in terms of changeable cultural attributes (Astuti, 1995; Astuti et al., 2004). In the current study, the attribute which was most central to children's definitions of ethnicity was religion. Most children asserted that if a person changed their religion, they would also change their ethnicity, a view which strengthened with age. This does not in itself indicate non‐essentialist thinking, in that religion itself can be essentialized and (therefore) considered fixed (Toosi & Ambady, 2011). However, a multiple regression showed that children's mutability answers with respect to religious transformations significantly predicted their beliefs in the possibility of ethnic change in an unknown peer, suggesting that their concepts of ethnic mutability were based partly on the perceived possibility of religious conversion.

Children's explanations suggested that many conceived ethnicity and religion as inseparable and mutually implicating. This connection was also apparent in an ethnographic study at the same school and persisted in spite of teachers’ efforts to teach children to separate the two (Woods, 2005, 2013; under review). Religious practice was a major forum in which many children at the school experienced ethnic identity (R. Woods, unpublished data). This centrality of religion in definitions of ethnicity is not unusual; adults in some other cultural groups also define ethnicity in terms of religion, at least in certain circumstances (Haines, 2007; Rong, 2001), and Gil‐White (2001) reports that when faced with a scenario in which a child of Mongol birth parents is adopted by a Kazakh couple, Kazakhs asserted that the child would be Kazakh only if s/he adopted Islam.

Most children in all year groups did not define ethnicity in terms of language, clothing, or food. Ethnographic research found all these attributes to be widely associated with ethnicity in practice at the school (Woods, 2005, 2007), raising the question of why children did not define ethnicity in these terms. Furthermore, in the identity component of the current study, children frequently referred to language and skin colour in their explanations for how they knew peers’ ethnic identities, yet did not think that changing them would alter ethnicity. For the cultural aspects, this may be because children regularly witnessed border‐crossing, whereby children and adults at school of one ethnicity participated temporarily in cultural activities associated with another ethnicity (Woods, 2005, 2007, 2013). In contrast, this rarely occurred for religion, and indeed, school policy forbade teachers from engaging children in other‐religion practice (Woods, 2013). Therefore, children's experiences may have led them to view ethnicity to be connected much more fundamentally to religion than to cultural attributes.

The multiple regression found that age significantly contributed to ethnic instability beliefs for an unknown other, even when children's consistency scores were taken into account, suggesting that children's emerging concepts of ethnic mutability are not based only on a definition of ethnicity in terms of religion. Explanations for mutability answers to stability questions often referred to the concept of freedom of choice, and this was especially common for older children. This is an intriguing finding in that teachers at the school did not espouse the view that ethnicity can be changed. However, teachers did tell children that one is free to choose one's religion (Woods, 2005). Given that many of the children closely connected ethnicity and religion, they may have learned to apply the concept of free choice to ethnicity as well. If so, then the children were not receiving a definition of ethnicity as mutable from older generations, but constructing it anew in a way that the adults around them may not have intended.

The preceding discussion suggests that there is more than one way in which to construct ethnicity as a non‐essential concept. While it remains possible that humans have an innate tendency to conceive of ethnicity in essentialist terms (Gil‐White, 2001; Hirschfeld, 1996, 1997), nevertheless the conceptualizations of ethnicity that humans are capable of constructing clearly vary substantially across communities, such that the developmental pathway to essentialism cannot be taken for granted (Astuti, 2001; Astuti et al., 2004; Birnbaum et al., 2010; Diesendruck et al., 2013; Rhodes & Gelman, 2009).

The current study is the first to demonstrate that as with gender constancy (Aboud & Ruble, 1987; Marcus & Overton, 1978), children are more open to ethnic change for an unknown peer than for themselves. Their explanations suggested that this was partly because they had evidence (in the form of photographs, videos, and testimony) about their own history that they did not have for an unknown peer, and partly because they were personally invested in their own ethnic identity, with its implications for family belonging and unity. Unlike their other‐stability scores, children's self‐stability scores did not change with age, which may be because their strengthening commitment to their ethnic group (Phinney, 1989) was offset by their increasing view of ethnicity as mutable. The implication is that even in communities in which performative theories of ethnicity prevail, emotional and social dimensions of ethnic identity may mean that these beliefs are rarely translated into practice.

In one sense, the findings of the consistency component of this study cohere with earlier constancy studies in that most children did not consider ‘superficial’ changeable attributes (clothing, language, food) to define ethnicity (Aboud, 1984; Aboud & Skerry, 1983; Bernal et al., 1990; Ocampo et al., 1997; Ruble et al., 2004. However, most (particularly Indian) children did define ethnicity in terms of religion, such that religious change necessitated ethnic change. This finding, of very different responses to different attributes, demonstrates the usefulness of including a range of different attributes to assess ethnic consistency.

An alternative interpretation of the current findings is that children understand the terms ‘English’, ‘Indian’, and ‘Somali’ to refer not to ethnicities, but to nationalities. This is implausible, because (1) most children at the school are British, yet few participants incorrectly identified a British non‐English child as English in the identity component; (2) in their explanations, children rarely referred to migration or legal processes involved in attaining a specific nationality.

How generalizable are these findings? Children's beliefs in ethnic mutability were grounded in the concept of freedom and a close connection between ethnicity and religion. Similar findings might be expected in any community that these two ideas prevail. There is evidence that religion is often central to people's definitions of ethnicity (Gil‐White, 2001; Haines, 2007; Rong, 2001), and this may be particularly likely in communities where ethnic groups tend to differ along religious lines. As already noted, definitions of ethnicity in terms of religion only entail mutability if religion is not itself essentialized (Toosi & Ambady, 2011). Teachers at the school where the current research took place strongly discouraged religious essentialism by promoting the view that religions are freely chosen (a view which was probably contradicted by many parents at home). Therefore, the version of ethnic mutability uncovered in this study seems most likely to occur in communities where ethnic and religious differences largely coincide, and where non‐essentializing views of religion hold sway. However, it is eminently plausible that ethnic mutability could be constructed out of a different set of circumstances, as appeared to be the case for the Vezo of Madagascar, for instance (Astuti et al., 2004).

This study had several limitations. Its design was cross‐sectional; ideally, age differences should be replicated with a longitudinal study. It was carried out in only one school, raising questions about generalizability discussed above. It asked children about just one transformation at a time. Most children did not see isolated changes to food, clothing, skin colour, and language as impacting ethnicity, but might have considered them more potent in combination. The research cannot tell us whether these children's concepts of ethnicity change further in adolescence.

A final limitation concerns the relationship between ethnic constancy and ethnic essentialism. Constancy is considered to be a key component of essentialist thinking (Gelman, 2003, 2009), and several studies have used ethnic or racial constancy as a proxy for essentialist thinking about ethnicity and race (see, e.g., Kinzler & Dautel, 2012; Pauker et al., 2010). However, it is logically possible that children might view a particular category (such as eye colour) as immutable, without also subscribing to other essentializing properties, such as a high inductive potential. Further research is needed to assess the legitimacy of inferring ethnic essentialism from ethnic constancy.

To conclude, this is the first study to provide convincing evidence of children constructing a developmental trajectory towards a non‐essentialized view of ethnicity as mutable. The evidence suggests that for the children in this particular community, this trajectory was based on the definition of ethnicity in terms of religion, accompanied by a non‐essentializing concept of religion as changeable. In other words, as with the Vezo of Madagascar (Astuti et al., 2004), these children were learning to define ethnicity in terms of attributes they deemed changeable. However, the developmental trajectory towards ethnic mutability was apparent only when children contemplated an unknown peer. Children's emotional, familial, and social investment in their own ethnicity means that their ethnic mutability concepts will likely rarely translate into practice, raising important questions about how and when ethnic mutability concepts influence people's intergroup behaviour.