Explaining intergenerational variations in English language acquisition and ethnic language attrition

Abstract

This paper aims to improve contemporary understanding of intergenerational variation in English (L2) and ethnic language (L1) proficiency. Analysis using wave 1 data from the Children of Immigrants Longitudinal Survey in Four European Countries (CILS4EU) on 1,032 fifteen-year-olds living in England across six generations shows that there is a strong effect of generation on L1 and L2 proficiency. Intergenerational variations in individual attitudes to assimilation, family contact factors and community contact factors affect oral and literal proficiency in both languages. Pro-assimilation attitudes have a strong association with increasing oral and literal proficiency in L2 across all generations, but especially for earlier generations. Anti-assimilation attitudes have neither similar associations with L1 proficiency, nor negative associations with L2 proficiency. Frequent visits to parental birth countries have the strongest positive associations with L1 maintenance. Exposure to L2 in school and local neighbourhood settings does not contribute to increased L2 proficiency.

Keywords:

• language
• assimilation
• transnational ties
• youth
• migration
• generation

Introduction

Language proficiency is an important issue for students of all ethnic groups as it is an essential requirement for success in the education and labour markets. Destination-country language acquisition is seen to be a key indicator of the integration of immigrants into society (Gordon 1964; Espenshade and Calhoun 1993), an important component of achievement in the labour market (e.g. Berman et al. 2002; Dustmann and Fabbri 2003) and a major factor for accomplishment in education (McLaughlin 1987; Kao and Tienda 1995; Portes and Rumbaut 2001). Similarly, the proficiency and use of origin-country languages are important indicators of the development and preservation of an ethnic social identity (Bourhis and Giles 1977; Giles and Johnson 1981). In England, the diversity of immigrant groups and languages spoken is large, allowing for considerable variation in language use and fluency across different groups and generations. This makes the issue of language proficiency extremely important and worthy of attention.

Given the theoretical connection between language acquisition and immigrant integration (Gordon 1964), it is plausible that language proficiency in both origin- and destination-country languages will vary across generations. This paper makes use of cross-sectional data from the first wave in England of the recent Children of Immigrants Longitudinal Study in Four European Countries (CILS4EU) (Kalter et al. 2013) to investigate patterns of language fluency among adolescents. The study attempts to determine how individual attitudes towards assimilation, family contact factors and community contact factors affect language proficiency in the destination-country language of English (L2) and in the host-country ethnic language (L1). The variation of these effects across generations and across ethnic groups is explored.

Previous research

Both generation and language proficiency have been linked to immigrant adaptation in traditional and contemporary theories of assimilation (Watkins and Melde 2010). Straight-line assimilation theory predicts greater L2 proficiency over generations and so posits that later-generation children have an educational advantage over earlier-generation children (Gordon 1964). Empirically, there is a clear pattern of changing L1 and L2 use over generations, with a strong tendency for a decline in L1 use in tandem with increasing L2 dominance in the latter generations compared to earlier generations (Portes and Rumbaut 1990).

There is evidence that students who are not proficient in English are at a higher risk of falling behind or underperforming in school compared to their language-proficient peers (e.g. Portes and Schauffler 1994; Rumberger and Larson 1998; Golash-Boza 2005). Such evidence supports the importance of identifying the factors that contribute to L2 acquisition.

There are various socio-linguistic theories that attempt to explain these patterns of language preference and proficiency – specifically, these theories posit explanations for L2 acquisition and L1 attrition. The term attrition¹ is used here to describe the intergenerational loss in language competence (Andersen 1982; Major 1993).

The first theory suggests that age is the critical factor in language acquisition and attrition. Proponents of this theory argue that the younger the child is when she or he moves to a new language environment, the more rapid, and intense, the language attrition (e.g. De Bode 1996; Isurin 2000; Schmid 2002). The age at which learners are first exposed to a language is a major predictor of their long-term success in the language (Long 2004). A particularly important distinction arising from the empirical literature is that made between pre-puberty and post-puberty migration, with youth in the latter group being better able to maintain L1 after migration than those in the former group (e.g. Olshtain 1986; Cohen 1989; Kaufmann 2000).

A second theory is that attitudes, motivations and other affective factors influence language acquisition and attrition (Gardner and Lambert 1972). A pro-assimilation attitude may result in an integrative motivation to learn L2 in order to be part of the L2 community and develop an L2 identity (Gardner 1985). Similarly, an aspiration to achieve upward social mobility may result in an instrumental motivation to learn L2 so as to be competent in a key skill needed to access educational and occupational opportunities (Gardner 1985). It follows that a positive attitude towards segregation or separation (an anti-assimilation attitude) may result in a reduced desire to learn L2 and an increased desire to maintain L1 in order to conserve an L1 identity and avoid assimilation into the L2 community. The empirical evidence for this theory of attitudes and motivations is mixed. Gardner and Lambert find that individual attitudes do make a difference to L2 acquisition. However, Waas (1996) finds that attitudes to maintaining L1 only influence self-perceptions of proficiency rather than actual proficiency.

The third set of theories focus on how sources of language contact and opportunities to use a language determine the rate of attrition. Opportunities for both L1 and L2 language contact within the social context of the individual language learner should not be overlooked (Köpke and Schmid 2004). Researchers speculate that language attrition results when languages come into contact with each other (e.g. Sharwood Smith and van Buren 1991; Seliger 1996). There is some evidence of this, although linguistic contact effects do not appear to be linear. For example, Hulsen, De Bot and Weltens (2001) found that the relationship between number of L1 speakers in the neighbourhood and L1 attrition was not linear in that larger numbers of L1 contacts did not necessarily correspond to higher L1 test scores. In a different study on language contact, De Bot, Gommans and Rossing (1991) found that length of time in the L2 environment had a linear effect on L1 attrition only in the instance where there was very little contact with L1 speakers – in other situations, time and contact interacted to produce a non-linear effect on L1 attrition.

In this study, the factors influencing language proficiency will be explored. The overall patterns of L2 acquisition and L1 attrition over generations and across ethnic groups will be examined. Following this, individual attitudes and motivations towards language acquisition and attrition will be investigated. Family and community factors that provide opportunities for linguistic contact will be tested to see how these affect language acquisition and attrition. Finally, these factors will be interacted with generation to see if their effects vary across generations.

Research questions

1. How do L1 and L2 proficiency vary across ethnic groups and across generations?

2. Are the rates of L1 attrition and L2 acquisition over generations the same for all ethnic groups?

3. Do individual attitudes towards assimilation, family contact factors and community contact factors explain variations in language proficiency?

4. Are these effects the same for:

   1. oral and literal proficiency?

   2. L1 and L2 proficiency?

   3. all generations?

Data and methods

Data from wave 1 of CILS4EU (collected 2010–11) are used for this study. These comprise initial data on 4,427 fifteen-year-old students attending 107 schools in England. The school sample was chosen using a stratified random sampling method, with stratification taking place according to the ethnic-minority density (proportion of non-white British students) of schools. School ethnic-density strata can be seen in Table 1.

Table 1. Definition of school ethnic-minority density strata

Ethnic-minority density strata	Percentage of ethnic-minority students in school
1	0–10
2	10–30
3	30–60
4	60–100

Two mixed-ability² classes were interviewed in most³ schools with a mean of 40.1 students interviewed per school.⁴ Students missing information on gender (nine cases) or generation (131 cases), wave 1 study dropouts (112 cases), and those who attended independent schools⁵ (348 cases) were removed from the analysis. This left 3,827 students, who were divided into two groups depending on whether or not they spoke (or potentially had spoken in the past) an ethnic language. Only students in the ethnic language group were kept for this analysis as it does not make sense to model ethnic language proficiency for students who originate from countries⁶ where English is typically used and so for whom there is no real scope for language assimilation. Limiting the L2 sample in the same ways allows inferences to be made about the parallel nature of L1 attrition and L2 acquisition.

Generational categorization

The relevance and importance of ‘decimal’ generation distinctions were emphasized by Rumbaut (1997) and are now well established in the literature. Following Rumbaut, this study makes full use of the information available to accurately classify students’ generation. The criteria for each generation can be seen in Table 2.

Table 2. Definition of student's generation

Generation	Student's place of birth	Student's age of migration to destination country	Parents' place of birth
1.25	Abroad	10 years or older	Both born abroad
1.5	Abroad	6–10 years	Both born abroad
1.75	Abroad	5 years or younger	Both born abroad
Second	UK	N/A	Both born abroad
2.5	UK	N/A	One parent born abroad; one parent born in UK
Third	UK	N/A	Both born in UK

This definition of generation better acknowledges that there is significant variation within the first and second⁷ generations due to differences in age of arrival and duration of residence. The logs of generation⁸ values are taken to better fit the non-linear trend in language assimilation and attrition over generations.

Language group categorization

Students were included in the ethnic language group if they themselves spoke an ethnic language (837 cases), or if they or their parents were born in a country where an ethnic language would typically be spoken, even if they did not currently report using an ethnic language (195 cases).⁹ These countries where an ethnic language would typically be spoken were determined using the country of birth and language spoken distribution of the 1.25 generation. This distribution guided the assignments of students to the ethnic language group if they originated from any of the countries¹⁰ where the majority of students of the 1.25 generation reported using an ethnic language. This procedure was necessary to identify those students from later generations who did not report using an ethnic language potentially because they had already integrated into mainstream society to the extent that their ethnic language use had ceased prior to the time of the survey. Without including these cases, the rates of language acquisition and language attrition will be underestimated, as those who have experienced the most extreme L1 attrition and corresponding L2 acquisition would be excluded from the analysis.¹¹

A further 452 cases were excluded due to missing one or more of the explanatory variables of interest. This was done to keep the sample size constant, which was necessary in order to compare model fit criteria across models. This left 1,032 cases upon which to run the analysis.

Table 3 shows the distribution of these 1,032 students across ethnic group, generation and gender. Descriptive statistics are weighted to account for the oversampling of students at strata 3 and 4 schools.

Table 3. Distribution of generation by ethnic origin and gender (row percentages)

	India			Pakistani			Other Asian			African			European			Other
			Unweighted			Unweighted			Unweighted			Unweighted			Unweighted			Unweighted	Total
Generation	M	F	n	M	F	n	M	F	n	M	F	n	M	F	n	M	F	n	row%	N
1.25	3.6	1.5	2	5.0	1.8	6	2.4	13.4	14	8.4	8.9	14	16.4	11.5	21	14.3	12.9	17	100.0	74
1.5	1.6	4.4	8	1.9	2.0	6	8.2	2.9	16	17.0	11.1	24	12.0	25.4	35	9.2	4.2	24	100.0	113
1.75	4.0	4.5	11	6.1	5.4	14	6.5	4.9	16	10.8	11.3	23	9.5	10.9	24	13.1	13.1	30	100.0	118
Second	8.0	11.2	62	13.5	16.8	97	11.7	7.6	54	13.3	10.3	82	2.3	0.4	10	1.2	3.8	18	100.0	323
2.5	12.3	7.9	63	14.0	9.1	71	4.0	3.3	19	4.9	14.6	40	6.4	10.6	32	8.6	4.5	30	100.0	255
Third +	16.7	4.1	45	4.2	2.9	12	0.8	2.8	5	11.0	5.7	35	9.7	19.6	28	7.6	15.0	24	100.0	149
N	113	78	191	114	92	206	67	57	124	107	111	218	73	77	150	71	72	143	100.0	1,032

Note: weighted percentages.

It was not possible to determine whether all generations had equal access to the same form of L1 as that used by the earliest (1.25) generation. Thus it is difficult to accurately assess L1 attrition of these generations when compared to that of the 1.25 generation because of potential differences in the initial level and form of L1 acquisition (Yağmur 1997; Montrul 2004). It is also not necessarily the case that age at migration is equivalent to age of first exposure to L2 (Long 2004).

Independent variables

The independent variables in the analysis included in all models are: age; gender; father has a university degree; mother has a university degree; father currently employed; mother currently employed; school ethnic minority density strata;¹² logged generation; ethnic origin; and ethnic origin interacted with logged generation.

Ethnic origin was constructed using a four-stage process. First, students born abroad were allocated to ethnic origin groups based on their region of birth. Second, the remaining second- and 2.5-generation students were allocated based on the region of birth of their parents. If either parent was born abroad, the student was allocated to the relevant ethnic origin group. Where there were differences in the regions for the mother and the father, the student was allocated to the mother's region. Third, the third + -generation were allocated to ethnic origin groups using their response to an ethnic-belonging question.¹³ Fourth, any remaining students were allocated to ethnic groups according to the ethnic language that they reported speaking at home. Ethnic groups with small numbers were combined into one group. This resulted in six broad ethnic-origin groups: Indian; Pakistani; other Asian; African; European; and other. This categorization correlated significantly with the ethnic-belonging measure. The European group was chosen as the reference category because of its reasonable size and its relative similarity with the white British majority group, which is typically used as a reference group in ethnicity research.

Individual, family and community factors

Students were asked how much they agreed or disagreed that ‘ethnic minority groups should adapt to British society’ (pro-assimilation) and that ‘ethnic minority groups should do all they can to keep their customs and traditions’ (anti-assimilation). A dummy variable indicating agreement was created for each of these two questions in order to capture each student's attitudes towards assimilation. Three dummy variables on family contact factors were constructed based on whether any of the student's grandparents live with them; whether the student visits relatives at least once a week; and whether the student visits their mother's and/or father's country of birth at least once a year. Two dummy variables on community contact factors were constructed based on whether a lot or all of the student's friends were from a different ethnic background to their own;¹⁴ and whether a lot or all of the people living in the student's neighbourhood were of a white British background.¹⁵ Each of these three sets of predictor variables were tested in separate models and then interacted with logged generation to see if the effects of the predictors varied by generation. All models also included interactions between ethnicity dummies and logged generation.

Dependent variables

The dependent variables in the analysis were the self-reported oral and literal proficiencies¹⁶ in L1 and L2. Recent studies suggest that self-assessments of language proficiency correlate with linguistic test scores (Yağmur 1997; Hulsen 2000). Alongside the models presented in this paper, the analysis was repeated using English language test score¹⁷ as the dependent variable.¹⁸

This study follows suggestions in the literature (e.g. Olshtain 1986; Köpke 2004) that it is important to distinguish between oral skills (understanding and speaking) and literal skills (reading and writing). The oral proficiency score for each language is the sum of students’ self-reported proficiency in the skill domains of understanding and speaking,¹⁹ which could range from 1 (‘not at all well’ in one skill and no response given for the other skill) to 10 (‘excellently’ in both skills). The literal proficiency score for each language is the sum of students’ self-reported proficiency in reading and writing,²⁰ which could also range from 1 to 10.

It is important to be upfront about the problems with endogeneity in the analysis of associations between language use and proficiency, making the direction of causality difficult to determine, especially with cross-sectional data. As such, variables specifically concerning language use are not employed in this analysis, with the focus instead remaining on access to L1 and L2 contact in terms of family and community sources. Similarly, it is plausible that it is L1 or L2 proficiency itself that causes the attitudes, family and community contact factors identified in this research, rather than the other way around, so causal interpretations must be made very cautiously.

Results

Before modelling the data, it is useful to consider some relevant descriptive statistics. Table 4 shows intergenerational variations in L2 language test scores and in average self-reported L1 and L2 proficiency scores in oral and literal skills.

Table 4. Language proficiency over generations

	L1 self-reported proficiency				L2 self-reported proficiency					L2 test score
	Oral		Literal		Oral		Literal
Generation	M	SD	M	SD	M	SD	M	SD	Unweighted N	M	SD	Unweighted N
1.25	7.3	3.8	6.8	3.8	8.1	1.7	8.1	1.9	74	14.4	4.7	72
1.5	6.3	3.4	4.5	3.2	8.7	1.7	8.8	1.6	113	16.1	3.4	110
1.75	4.9	4.0	3.3	3.3	9.2	1.3	9.0	1.4	118	17.2	3.4	117
Second	6.7	2.7	3.8	2.5	9.6	1.1	9.5	1.3	323	17.2	3.1	322
2.5	4.7	3.5	3.1	2.8	9.5	1.7	9.3	1.8	255	17.5	2.9	249
Third +	3.3	3.2	2.5	2.6	9.5	1.0	9.2	1.5	149	18.2	3.2	140
Total	5.2	3.6	3.5	3.0	9.3	1.4	9.2	1.6	1,032	17.2	3.4	1,010

Note: Weighted means and standard deviations. Max L2 test score = 25.

There are clear trends in improving L2 oral and literal proficiency²¹ in tandem with decreasing L1 oral proficiency, and especially literal proficiency, over generations. This trend is also evident in the increase in L2 language test score over generations. The objective L2 test measure correlates well with the subjective L2 proficiency measures.

Importantly, the rate of L2 acquisition (according to subjective proficiency) does not appear to be linear over generations. Rather, there is a rapid increase for the earlier generations, which flattens out by the second generation and then slightly declines for the later generations. Similarly, the rate of L1 attrition is most rapid for the earlier generations and then reduces somewhat. There is still a general decline in both L1 proficiencies over generations, but especially so for literal proficiency. There is an exception in this trend for the second generation, who report higher L1 oral and literal proficiencies than the 1.75 and 2.5 generations. This is perhaps due to more heterogeneity in the second-generation sample.

Socio-demographic and generational effects

Table 5 shows models predicting oral and literal language proficiency for L1 and L2. The same model specification is used to predict L2 test scores and provides evidence robustness of the results for L2 oral and literal proficiency.²² The coefficients for L2 test score and L2 self-reported oral and literal proficiencies are very similar indeed, which is reassuring for this analysis. The socio-demographic controls and interactions between ethnic-origin dummies and logged generation are presented in Table 5 to show the general trends of these variables.

Table 5. Language proficiency models with socio-demographic and generation variables

	L1 self-reported proficiency				L2 self-reported proficiency				L2 test score
	Oral skills		Literal skills		Oral skills		Literal skills
	B	SE	B	SE	B	SE	B	SE	B	SE
Constant	1.74	2.80	−0.06	3.07	8.71***	1.30	9.02***	1.46	6.35***	1.23
Female	0.09	0.19	0.02	0.16	0.13	0.10	0.26*	0.13	−1.36	0.12
Age	0.33	0.17	0.41*	0.19	−0.08	0.09	−0.11	0.10	0.23*	0.09
Father has a degree	−0.17	0.23	0.08	0.21	0.10	0.08	0.11	0.10	0.30*	0.10
Mother has a degree	−0.09	0.23	0.12	0.20	0.10	0.10	0.21	0.11	0.75*	0.12
Father has a job	0.25	0.15	0.18	0.14	0.00	0.10	0.08	0.11	−0.13	0.10
Mother has a job	−0.68***	0.18	−0.23	0.19	0.14	0.12	0.10	0.14	0.12	0.08
School ethnic-minority density strata	0.76***	0.14	0.49**	0.18	0.02	0.06	−0.01	0.07	−0.28	0.07
Ethnic origin (ref.=European)
Indian	0.28	0.85	−3.08**	0.89	1.23*	0.58	1.51*	0.57	1.22*	0.60
Pakistani	0.48	0.90	−1.73	1.00	0.63	0.63	1.29*	0.64	0.44	0.60
Other Asian	1.26	0.81	0.12	0.97	−0.61	0.58	−0.16	0.60	0.08	0.60
Black African	0.98	0.79	−1.06	0.81	1.26**	0.41	1.33*	0.53	0.89	0.52
Other	−0.23	0.89	−0.52	0.91	0.18	0.39	0.30	0.57	0.22	0.61
Logged generation status	−3.93***	0.81	−4.07***	0.80	1.79***	0.53	1.58***	0.44	1.76**	0.51
Logged gen*Indian	0.75	1.01	2.78*	1.07	−1.03	0.62	−1.08	0.67	−1.06	0.67
Logged gen*Pakistani	0.31	1.32	1.00	1.29	−0.41	0.77	−0.98	0.79	−0.70	0.74
Logged gen*Other Asian	−2.31	1.40	−2.77*	1.35	1.22	0.79	0.71	0.82	0.06	0.72
Logged gen*Black African	−4.12***	1.12	−1.70	1.06	−1.26**	0.46	−1.28*	0.68	−1.06*	0.62
Logged gen*Other	−0.42	1.21	−0.25	1.16	−0.45	0.72	−0.58	0.86	−0.52	0.80
N	1,032		1,032		1,032		1,032		1,010
R^2	0.32		0.24		0.10		0.07		0.12
F (18, 94)	41.30***		27.89		7.04		5.74		7.95

Note: Unweighted. Clustered standard errors. L2 test score has been rescaled to make coefficients comparable across models.

* p<.05, ** p<.01, *** p<.001

Overall, associations with control variables are as expected and consistent across latter models. As such, coefficients for these covariates are controlled for but not displayed in the later models shown in Table 6. Females are well known to outperform males in English, so the positive gender coefficient for L2 literal proficiency is not surprising. The negative effect of mother's employment status on oral skills in L1 implies that stay-at-home mothers may be more likely to use L1 with their children, thus fostering development of L1. This could be because stay-at-home mothers from ethnic minorities are those that typically have lower English-language skills and so predominantly use L1 in any case.²³ School ethnic-minority density has a positive effect on L1 oral skills, which could be because of peer concentration effects, which make ethnic languages more commonly used in more ethnically dense schools, thus increasing oral proficiency.

Table 6. Models predicting self-reported language proficiency

The most important finding concerns associations between generation and proficiency. It is clear that logged generation has a persistent significant negative effect across models on L1 oral and literal skills, but a significant positive effect on L2 oral and literal skills. Logged generation implies that the decline in L1 proficiency and increase in L2 proficiency is rapid for the earlier generations but then slows down for the later generations – that is, the effect of generation is non-linear. This corresponds to the patterns evident in Table 4. This is to be expected as acculturation takes place over generations. Logged generation is the variable that explains most of the variance in the dependent variable across all models.

The interactions between ethnicity and logged generation suggest that compared to the L1 attrition rates of children of European origin, L1 oral skill attrition is more rapid over generations for children with an African origin, while L1 literal skill attrition is less rapid over generations for children with an Indian origin. The negative interaction between logged generation and African origin for both oral and literal L2 proficiency suggests that the rate of increase in L2 proficiency is significantly slower for children of African origin compared to children of European origin. This could be because children of African origin generally start off with a higher level of English than children of European origin²⁴ (although it is important to consider the high degree of within-group heterogeneity in both of these broad ethnic-origin groups). There are no other significant interactions between ethnicity and logged generation, suggesting that generation is generally associated with language proficiency in a similar way across ethnic groups.

Predicting L1 and L2 proficiency

There are three models for each proficiency type in both L1 and L2: model 1 tests individual assimilation attitude factors; model 2 tests family contact factors; and model 3 tests community contact factors. Each set of factors was tested in a separate model alongside all of the socio-demographic and generation interaction covariates included in Table 5, but for space reasons covariate coefficients are not shown in Table 6. Following Raftery (1995), Bayesian Information Criterion (BIC)²⁵ statistics were used to determine model selection.

Table 6 shows the relevant coefficients for each set of factors for each proficiency type. Looking first at L1 proficiency, the size and direction of the coefficients in the chosen family contact models are similar for both oral and literal proficiency. The main effect of interest in these models is that of visiting the mother's/father's country of birth at least once a year. This is logical as experiences in the country of origin might well boost L1 skills as these may be needed to communicate with extended family members and the community more generally. The effect of visiting mother's/father's country of birth is the same for all generations for both proficiency types. Despite not being the preferred model, the factors included in model 1 are worthy of note because of the absence of a significant effect of a student's anti-assimilation beliefs on L1 oral and literal proficiency. This is an important finding as it indicates that a student's belief in the maintenance of ethnic customs does not necessarily result in an action to support this belief, at least with regards to ethnic-language maintenance. However, parents’ beliefs on the maintenance of ethnic customs may well be relevant but are not directly captured.

Overall, it appears that the same family contact factors contribute to both oral and literal L1 skill maintenance. Individual attitudes towards assimilation do not seem to matter significantly, and associations tend not to vary over generations.

Looking now at L2 proficiency, again there is a similar pattern for both oral and literal skills. The preferred models are those with individual assimilation attitude factors (model 1 for both proficiency types). There is a positive significant effect on L2 oral proficiency of a pro-assimilation belief. This effect varies by generations with the significant negative interaction effect suggesting that pro-assimilation beliefs have the strongest effect on L2 oral proficiency for the 1.25 generation and subsequently decline rapidly and then flatten out for the later generations. The same main and generational interaction effects of pro-assimilation beliefs are insignificant for L2 literal proficiency.

This finding makes sense, as those students who agree that ethnic minorities should adapt will be those who have an integrative orientation and so are most ready and willing to adapt and integrate themselves. The weakening of the effect over generations could be because those in the earlier generation who are eager to adapt will be the ones keenest to acquire English as a signal of adaptation, whereas in the later generations, most students would have reached a decent level of proficiency in English and so will have to adapt other customs to assimilate further.

There are no significant effects of the community contact factors under investigation on L1 and L2 oral and literal proficiencies (model 3 in Table 6). The lack of a white British neighbourhood effect is possibly because the measure of neighbourhood contact does not capture the actual interactions that may affect proficiency. Having friends of a different ethnic background to one's own does have negative effects on L1 proficiency but these effects are not large enough to be significant.

The significant positive interaction between visiting parental-origin country and generation on L2 oral proficiency may well be a false positive as the main effect is negative and insignificant.

Conclusion

This paper looked at trends in L1 and L2 proficiency across ethnic groups and over generations, and investigated individual and social factors that might influence proficiency levels. L2 proficiency in both oral and literal skills varies somewhat across ethnic groups but tends to increase strongly over generations. L1 proficiency varies much more than L2 proficiency across ethnic groups and between skill types, with faster attrition in literal skills compared to oral skills. Nevertheless, there is a clear intergenerational trend in L1 attrition, with the highest levels of proficiency being reported by the newest arrivals and a rapid non-linear decrease over generations.

Individual attitudes towards assimilation seem to influence L2 proficiency in the way expected, but not L1 proficiency. An anti-assimilation belief does not reduce the rate of L1 attrition. However, a pro-assimilation belief does have a positive effect on L2 acquisition. Thus it is plausible that interethnic differences in education may arise because of a lack of a pro-assimilation belief among some ethnic groups.

Family contact factors in the form of visits to mother's/father's country of birth have unsurprising positive associations with L1 proficiency. There are no contextual effects of school ethnic-minority density on L2 proficiency. This is unsurprising as individual ethnic-minority status itself is not significantly negatively associated with L2 proficiency. The relevant school contextual effect for future investigation should be first-generation density.

Effects do not typically vary significantly across generations with one exception: a decline in the positive effect of a pro-assimilation attitude on L2 oral proficiency over generations. It is yet to be determined as to whether L2 acquisition causes L1 attrition, although the results suggest that these two developments appear to happen in tandem.

The first main implication of this research is that L2 acquisition and L1 attrition are associated not just with individual attitudes but also with the wider social context.

The second point of significance is the asymmetry in associations with covariates for L1 and L2 proficiencies. While a pro-assimilation attitude positively contributes to L2 proficiency, there is no similar negative contribution of an anti-assimilation attitude. Likewise, the positive associations between transnational ties and L1 proficiency are not to the detriment of L2 proficiency – maintaining a connection to the country of origin does not appear to hold back L2 acquisition. This suggests that fears of acculturation being limited because of the maintenance of ethnic customs and transnational ties are unfounded.

The final important conclusion is that the effects of generation on L2 acquisition and L1 attrition are difficult to neutralize. The increased intergenerational integration of ethnic minorities into the majority society appears to encourage the dominance of L2, but at the expense of L1 survival outside of the country of origin.

Additional information

Notes on contributors

Meenakshi Parameshwaran

MEENAKSHI PARAMESHWARAN is a Research Associate at the Centre on Dynamics of Ethnicity at the University of Manchester

Notes

1. Language attrition is frequently accompanied by language shift, which describes an intra-generational movement from the use of L1 to L2.

2. Mixed-ability classes were requested to ensure that the sampled students were representative of the ability distribution in that school.

3. Only one class was interviewed in a handful of schools.

4. Standard deviation 9.17 (2.d.p.). Max = 60; min = 12.

5. Independent schools were excluded because of lack of information on the ethnic-minority density of these schools.

6. For this analysis, English-using countries are deemed to be the UK, Australia, South Africa, the USA, New Zealand, Canada and the English-speaking Caribbean countries.

7. The definition of the 2.5 generation does not distinguish between UK-born parents from the majority group and UK-born parents from the minority group.

8. Generation is treated as a continuous measure in this analysis.

9. Students were asked if a language other than English was spoken in their home.

10. The countries/regions were India, Pakistan, other Asia (Bangladesh, China and some others), Africa, Europe and other (Turkey, some Middle Eastern countries and various others)

11. The analysis was repeated without these 195 cases, with results suggesting that such an exclusion of cases would have underestimated attrition and acquisition effects for the black African and Pakistani groups.

12. This control is included to adjust for the oversampling of students attending strata 3 and 4 schools.

13. Students were asked which group(s) they felt that they belonged to (multiple-choice with an open-response box). This question was not used as the main determinant of ethnic origin due to greater difficulties in interpreting students’ responses compared to the clearer measures of student's country of birth and parents’ countries of birth.

14. Four questions were posed asking students what proportion of their friends were from each of the following backgrounds: white or white British; black or black British; Asian or Asian British; other background. Student's own ethnic background was matched to one of these broader groupings.

15. This is based on the student's own perception of the ethnic composition of their neighbourhood rather than an objective measure of composition.

16. Students were asked how well they thought they could speak, understand, read and write English. Response options were ‘not at all’ = 1, ‘not well’ = 2, ‘well’ = 3, ‘very well’ = 4, ‘excellently’ = 5. The same question was asked of proficiency in an ethnic language.

17. The English language test was a multiple-choice synonym-identifier test administered at the time of the survey.

18. The results with English language test score as the dependent variable mostly follow the patterns demonstrated for the models predicting oral and literal English proficiency. These results are available upon request.

19. Correlations between understanding and speaking proficiencies were 0.76 for the ethnic language and 0.78 for the English language.

20. Correlations between reading and writing proficiencies were 0.90 for the ethnic language and 0.79 for the English language.

21. For comparison, the mean weighted self-reported L2 oral proficiencies for the non-ethnic language group were 9.7, 8.9 and 9.8 for the 1.25, 1.5 and 1.75 generations, respectively. For literal proficiency the means were 10.0, 9.0 and 9.8. Mean L2 tests scores were 16.6, 15.5 and 20.8. This indicates that students who originate from countries where English is typically used achieve faster language assimilation compared to their similar-generation peers from countries where an ethnic language is typically used.

22. There are significant positive effects of father's and mother's degree-level education on L2 test score but not on self-reported L2 oral or literal proficiency.

23. Unfortunately it is not possible to test this explicitly due to the low response rate to the parental questionnaire.

24. This assertion is supported by the data, which shows that African 1.25 generation students reported a mean L2 oral proficiency score of 8.7 (8.4 for literal proficiency), compared to 7.7 (7.3) for European 1.25 generation students.

25. BIC aids in model selection by providing a statistic of the likelihood of the model parameters that includes a penalty for the number of parameters in the model. The model with the lowest BIC is preferred.