The Threshold Hypothesis revisited: Bilingual lexical knowledge and non-verbal IQ development

Bilingualism and cognition

The relation between bilingualism and cognition has been under investigation from as early as the first half of the 20th century, where research found a bilingual disadvantage and negative correlations between bilingualism and general cognition. Saer (1923) reported negative effects of bilingualism for general intelligence scores of children in Wales. It has been argued that this study and other studies of that time were methodologically weak for a number of reasons, including lack of control for socio-economic-status (SES), schools attended and a lack of appropriate statistical procedures (for a detailed critique see Baker (2011)). These studies were also probably politically biased, because bilingualism was seen at the time as a ‘psychological and educational problem’ (Darcey, 1946, p. 21). A comprehensive overview of earlier research can be found in Hakuta (1989). Peal and Lambert (1962) were among the first researchers who reported positive effects of bilingualism on intelligence. They report that bilingual children (mean age 10) outperformed monolingual peers both in verbal and non-verbal intelligence tests. Their findings are not undisputed as there was a possible bias in the selection of the participants (see Hakuta & Diaz, 1985, p. 322/323). However, the discussion needs to go beyond the identification of possible methodological flaws to develop a more in-depth understanding of the relation between bilingualism and cognition. We therefore need to discuss first relevant theoretical frameworks in the next section.

One of the most influential theoretical frameworks on the relation between bilingualism and cognition is Cummins’ ‘threshold hypothesis’ (Cummins, 1976, 1979, 1980, 1981, 1991). This hypothesis assumes that bilingualism has negative cognitive effects below a certain threshold of proficiency in both languages. Above this level, there is no negative effect, and if the proficiency rises above the second threshold level positive effects can be found. In other words, only high proficiency in both languages leads to positive effects on cognition. One aspect of cognition that is investigated in several studies on bilingualism is non-verbal spatial tasks. This means that a link is made between a linguistic characteristic (bilingualism) and non-linguistic cognition. In a summary of previous research, Diaz and Klinger (1991, p. 167) conclude, ‘children’s bilingualism is positively related to concept formation, classification, creativity, analogical reasoning, and visual-spatial skills’ (italics added). In the context of the present study, the last part of this statement is especially important as it includes a non-verbal aspect of a bilingual advantage (see also Hakuta & Diaz, 1985). Diaz (1985) investigates bilingual non-verbal cognition with 100 first-grade Spanish–English bilinguals using Raven’s Coloured Progressive Matrices (RCPM). He finds significant correlations between measures of verbal and non-verbal intelligence (r = .39, p < .05) but concludes that contrary to Cummin’s ‘two-threshold hypothesis’ not only children with high proficiency in both languages show cognitive advantages. Jarvis, Danks and Merriman (1995) try to replicate Diaz’s findings with 50 Spanish–English bilingual children at an average age of 9.7 from middle-class backgrounds. They could not identify any significant correlation between the degree of bilingualism and cognitive advantages.

Bialystok and Majumder (1998) investigate the effects of bilingualism on non-verbal problem-solving abilities of children (n = 71, average age 8:8). They compared a group of ‘balanced’ French–English bilingual children, a group of ‘partial’ ¹ Bengali–English bilinguals for whom English was the dominant language and a group of monolingual English-speaking children, all from middle-class backgrounds. The study shows that the French–English bilingual children have a significant advantage in a non-verbal task, the Block Design Task. This task is part of the Wechsler Intelligence Scale for Children-Revised (Wechsler, 1974), where children are required to use coloured blocks to duplicate patterns that are presented in pictures. Bialystok and Majumder argue that the metalinguistic advantages of bilingual children ‘extend into non-linguistic problem solving’ (1998. p. 81) even when the tasks are spatial in nature. They give a tentative explanation for this advantage by explaining that bilinguals ‘must be attentive to non-salient features of the input, such as the language in which messages are spoken’, and that this will lead to the ‘ability to focus attention selectively on required aspects of a problem’ (Bialystok and Majumder, 1998, p. 83). In other words, the bilingual experience requires a constant attention to the language of the input, which in turn improves the ability of the children to focus their attention on problem solving in general. This explanation is tentative, but may be a first step towards explaining why a linguistic capacity (bilingualism) may lead to advantages for non-linguistic tasks. This is supported by Bialystok and Martin (2004), who report on three studies where bilingual children (average age 5 years) outperformed monolingual children in non-verbal cognitive tasks (a dimensional change and card sorting task and a colour–shape task) but not on semantic tasks. Their interpretation is that bilinguals constantly have to inhibit the non-relevant language in a given context and that ‘the inhibition of the non-relevant language is controlled by the same cortical centres used to solve tasks with misleading information’ (Bialystok & Martin, 2004, p. 338). This leads to non-verbal cognitive advantages. After a discussion of several studies, Kroll and Bialystok (2013, p. 504) conclude that the influence of bilingualism on non-verbal cognitive processing is ‘unique to this research and unexpected’. These findings support the threshold hypothesis by explaining why bilingualism may lead to cognitive advantages. However, there are also some conflicting results in research on bilingual (dis)advantages. In a meta-analysis Costa, Hernandez, Costa-Faidella and Sebastián Gallés (2009) show that exactly 18 of the 36 studies show a bilingual advantage and 18 do not. Paap (2015) and Paap, Johnson and Sawi (2015) point to the scarcity of large-scale studies which identify a bilingual advantage. Valian (2015) gives an overview of current research comparing advantages of executive functioning of bilinguals over monolinguals and concludes that the outcomes of these studies are inconsistent. The reason for these inconsistencies is probably the fact that there are too many independent variables, and that therefore different studies cannot be compared. In a similar vein, Thordardottir (2011) argues that there are many different bilingual populations and that many factors influence the outcomes of studies on bilinguals, such as time of onset, amount of exposure, status of the two languages and SES of the parents. In this context, it is important to note that Luk and Bialystok (2013) argue that bilingualism is not a categorical variable, and that at least two dimensions have to be included in studies on bilingualism, that is, ‘language proficiency and usage’, and we follow this approach in the present study. Luk (2015) argues that inconsistent results for studies of executive function advantages which compare bilinguals with monolinguals may be explained by different monolingual and bilingual ‘experience’ (2015, p. 35), that is, differences in language usage, language acquisition settings, language proficiency and socio-economic background, which are confounding factors in these studies. Overall, there is no agreement in the literature whether there is a bilingual cognitive advantage, nor whether there is a threshold or a certain degree of bilingualism that would ensures a cognitive advantage.

Vocabulary and cognition

An intriguing aspect of vocabulary knowledge is its close relation with cognitive ability measured either with standardised IQ scores or with other ability tests in an experimental setting. One of the earliest studies on the relation between vocabulary and intelligence is Terman, Kohs, Chamberlain, Anderson, and Bess (1918). They reported a correlation of .91 (Pearson) between mental age and the vocabulary sub scale of the Stanford Revision of the Binet–Simon IQ test (Binet & Simon, 1905/1916). The authors claim that vocabulary tests can be used as a short measure for intelligence tests in general (see Terman et al., 1918, p. 454). Anderson and Freebody (1979) give an overview of studies where vocabulary subtests are correlated with other subtests of IQ tests. These correlations range from .71 to .98, and the authors conclude ‘the strong relationship between vocabulary and general intelligence is one of the most robust findings in the history of intelligence testing’ (1979, p. 2; see also Hakuta, 1987). In a similar vein, Sternberg (1987, p. 90) concludes, ‘Vocabulary is probably the best single indicator of a person’s overall level of intelligence’. One has to bear in mind that many IQ tests, such as the Binet scale that Terman used, rely also at least partially on language, and there is the potential of a circular argument when vocabulary size and IQ test scores are correlated (for a discussion see Kaplan & Saccuzzo, 2012). However, this circularity is not given when non-verbal IQ scores are used that are entirely non-verbal, such as the RCPM.

In this context, it is important to discuss the so-called ‘bilingual gap’ where a deficit in vocabulary knowledge is attested for bilinguals when they are compared with monolingual control groups. This can be used as an argument for a bilingual cognitive disadvantage. This ‘bilingual gap’ is identified in many studies (Bialystok, Craik, Green, & Gollan, 2009; Bialystok & Feng, 2010; Bialystok, Luk, Peets, & Yang, 2010; Bialystok & Martin, 2004; Bialystok & Viswanathan, 2009; Eilers, Pearson, & Cobo-Lewis, 2006; Oller & Eilers, 2002; Oller, Pearson, & Cobo-Lewis, 2007; Pearson, Fernández, & Oller, 1993; for a detailed overview see Thordardottir 2011). These studies typically measure only one language of the bilinguals (generally English) because standardised vocabulary tests in other languages are not always available (see Bialystok & Martin, 2004). However, even when both languages are investigated this gap is found when bilinguals are compared with their monolingual peers, especially with regard to productive vocabulary (Junker & Stockmann, 2002; Marchman, Fernald, & Hurtado, 2008; Oller & Eilers, 2002; Pearson, Fernandez & Oller, 1993, 1995; Petitto & Kovelman, 2003). These findings are based on separate comparisons for each language of the bilinguals with monolingual control groups. These separate comparisons seem to indicate a bilingual disadvantage for vocabulary size. However, this does not account for the fact that bilinguals use their two or more languages in different domains, for example school and home language, and the development of their proficiency in their language follows the ‘complementary principle’ (see Grosjean, 1982, 2001, 2015). Therefore, it is natural that bilinguals develop smaller vocabularies in certain domains and larger vocabularies in others in each language. A comparison of the vocabularies of bilinguals with monolinguals needs to take both languages of a bilingual into account and the vocabularies need to be studied together, not separately. In line with this approach, the notion of total conceptual vocabulary (TCV) was developed (see Pearson et al., 1993; Swain, 1972) where both languages of a bilingual child are taken together and the knowledge of lexicalised meanings is counted regardless in which language these meanings can be understood or expressed by the child. ‘Bilingual TCV, then, abstracts away from the number of languages a particular meaning is known’ (De Houwer, Bornstein, & Putnick, 2014, p.1192). The child gets credit for the knowledge of concepts rather than for knowing the word for it in both languages.

One specific aspect of TCV is the unit of counting for translation equivalents (TE), that is, words that have the same meaning in both languages, for example ‘cat’ in English and ‘kedi’ in Turkish. Within the framework of TCV a child is only credited once for the knowledge of TEs even if s/he knows the word in both languages. TCV is therefore smaller than the sum of the words in both languages, but normally larger than the vocabulary in each language of a bilingual. For monolingual control groups TCV is identical with their vocabulary size. The TCV of bilinguals reaches or exceeds that of demographically matched monolinguals in many studies (De Hower et al., 2014; Hoff et al., 2012; Oller et al., 2007; Pearson et al., 1993; Vermeer, 1992), although there are studies that confirm this only for receptive vocabulary (Gross, Buac, & Kaushankayav, 2014). Overall, there seems to be no bilingual disadvantage when TCV is taken into account and a certain exposure to both languages is provided. Thordardottir (2011) investigates the vocabulary development of 84 children (age range 4;6 to 5;0) in a Canadian bilingual context; 49 children were French–English bilinguals, 19 French monolinguals and 16 English monolinguals. She draws the conclusion that bilinguals have in their two languages together ‘roughly the same number of things and concepts for which monolingual children of the same age have a single label’ (Thordardottir, 2011, p. 443), provided there is a minimal critical level of exposure and a supportive environment. Poulin-Dubois, Bialystok, Blaye, Polonia, and Yott (2013) carried out a study with 43 monolingual and bilingual aged 24 months. The vocabulary size of the children was estimated with a parental questionnaire. In line with the expectations, the vocabulary size of the bilinguals was significantly smaller than of the monolinguals when only one language was compared. However, when taking both vocabularies of the bilinguals together there were no significant differences between the groups in vocabulary size. A bilingual disadvantage in the area of vocabulary, or a ‘vocabulary gap’, might be simply an artefact of the research methodology and might be non-existent if the TCV is taken into account.

Cognitive development and support for L1

A further theoretical framework that is important for an analysis of the cognitive development of bilingual children is the Common Underlying Proficiency (CUP) hypotheses (Cummins 1976, 1979, 1980, 1991), which states that support for one language of a bilingual is also beneficial for the other language (see Cummins & Swain, 1986, p. 87). Cummins (1991) reports on a series of studies that show a close relation between proficiency in L1 and L2 in bilinguals. These studies show that reading skills, writing skills, and vocabulary knowledge in the two languages of a bilingual are related. One reason for this might be that conceptual information acquired in L1 transfers to L2, and ‘minority-language children learn a second language best when their first language is maintained and developed’ (McLaughlin, 1986, p. 35; see also Umbel, Pearson, Fernández, & Oller, 1992). According to this framework, the development of vocabulary in L1 and the development of concepts of bilingual children are linked, and this knowledge is beneficial for the acquisition of L2 vocabulary. For this reason Cummins (1976 1976, 1979, 1980, 1981, 1991) argues that support for the minority language in an immigrant setting is crucial for the development of both languages of a bilingual child. This is supported by Collier’s (1989) meta-analysis of research on bilingualism. She concludes

Preschool children who begin second language acquisition any time between ages 3 and 5 (sequential bilingualism) are not at any disadvantage as long as they continue to develop their first language at the same time that they are acquiring the second language. (1989, p. 511).

These findings clearly focus on the important role of support for the minority language and its continuous development alongside the majority language. There is, however, a gap in our knowledge about the importance of parental support for L1 and its influence on the linguistic and cognitive development of bilingual children.

Participants

The participants in this study were 100 Turkish–English sequential bilingual children living in the UK, as well as 25 English monolingual children in the UK and 25 Turkish monolingual children living in Turkey. There were 43 male and 57 female children in the bilingual group (mean age = 9;4 years; range 7;1–11;9), 12 male and 13 female children in the Turkish monolingual group (mean age = 9.3 years; range 7;1–11;10) and 12 male and 13 female children in the English monolingual group (mean age = 9.4 years; range 7;1–11;6). All children come from a middle-class background and at least one of the parents has a university or college degree. The parents of the bilingual children all emigrated from Turkey and the children were all born in the UK. The bilingual children were first exposed to Turkish at home and then gradually had more contact with English speakers, especially after entering pre-school around the age of 4. In addition to the data from the children, we have questionnaire data from the parents. In contrast with many other studies, the bilingual children in the present study have only occasional contact with Turkish speakers outside the family as they live in different areas and do not have Turkish neighbours.

As mentioned earlier, bilinguals are tested in many cases only in one language, mainly English. This is because standardised tests are not available for many immigrant languages or because the participants have a wide range of other languages, and it is not possible to analyse the proficiency of all informants in these languages. The present study controls for L1 background by including only bilinguals that have the same ‘other language’, namely Turkish.

Measures and procedures

All tasks for the bilingual group were administered in the homes of the children in the UK. For organisational reasons, the monolinguals carried out the tasks in schools in the UK and in Turkey. All data were collected by the second author of this study. The language of test instruction was Turkish for the Turkish tests and English for the English tests. The following tests were administered:

The receptive vocabulary of the bilinguals was measured with ‘X-lex’ in English and Turkish (for the test format see Meara & Milton, 2003). X-lex is based on the yes–no format where participants have to indicate whether they know a word or not. It was originally designed for EFL learners at college level but has been used in with children aged 7 years (Milton, 2006). Each test includes 100 words ordered according to frequency bands (20 words from K1, K2, K3, K4 and K5). For English we used the existing test (Meara & Milton, 2003); for Turkish we created a comparable test based on the frequency data based on a 3.3 million word corpus provided by sketch engine (Ambati, Reddy, & Kilgarriff, 2012; Kilgariff, Rychly, Smrž, & Tugwell, 2004). Every 50th word from the first 5000 words in this frequency list was selected and included in the Turkish X-lex. The order of the words in the tests was based on a random number generator (https://www.random.org/). Both tests also include 20 pseudo words that are phonologically plausible in each language but do not exist. These words are included to correct for possible guessing. In line with Meara and Milton (2003), the final score for each language was then computed by giving 50 points for each accepted real word and −250 points for each accepted pseudo word. By this, a maximum score of 5000 (all real words but no pseudo word accepted) and a minimum score of zero (all real words and all pseudo words accepted) could be obtained. The tests were administered orally to avoid potential problems with unfamiliar spelling. The test instructions were given in the same language as the tests.

The productive vocabulary was measured with a verbal fluency test, which is widely used for psychological and neuropsychological assessments (Deutsch, 1995; Lezak, 2010). This test format has also been used for the measurement of verbal abilities and vocabulary knowledge and lexical access (Cohen, Morgan, Vaughn, Riccio, & Hall, 1999; Federmeier, McLennan, Ochoa, & Kutas, 2002; Milton & Roghani, 2015). Participants are either asked to produce as many words in 60 seconds that start with a certain letter (phonemic fluency task) or words that belong to a certain semantic category (category fluency task), such as animals fruits or colours. According to Friesen, Luo, Luk, and Bialystok (2015), the lexical fluency task demands more resources from executive control than the semantic task, and in addition, you need to be literate in both languages.

For our sample, we therefore decided to use the semantic fluency task and to give the participants two minutes rather than one because of the age range of our participants. The categories in our study were names for food, body parts, clothing and colours. The test was administered in both languages with a time gap of 2 weeks between the recordings to avoid priming effects. The language of instruction was English for the English task and Turkish for the Turkish task, and the participants did not attempt to use the other language during the test. The answers were tape recorded and then transcribed for the analysis.

As a test of non-verbal intelligence, we used RCPM (Raven, 1962; Raven, Raven, & Court, 2004). Test-takers are asked to complete a set of abstract graphical patterns with a matching pattern from a multiple-choice set of possible answer patterns (maximum score = 36). According to Raven et al. (2004, p. 1) the test ‘is designed to assess … mental development up to intellectual maturity’. It is described as ‘non-verbal estimate of fluid intelligence’ (Bilker et al., 2012, p. 354). For our study, it is important to stress that this test is non-verbal.

The Bilingual Dominance Scale (Dunn & Fox-Tree, 2009) was used as a questionnaire for the parents of the bilingual participants. This questionnaire consists of 12 questions about the language dominance of the participants. There are different weightings for each question, and in total, a maximum score of 31 was possible for dominance in each language. We computed a dominance score (scores for Turkish minus scores or English) which leads to a scale from −31 (only English preferences) to +31 (only Turkish preferences). Mothers and fathers were interviewed separately and did not know the score of the other partner. The average score of the parents was then counted as parental dominance score.

In addition, we administered a questionnaire about language use at home. This questionnaire includes 12 questions and is part of the language and social background questionnaire (Luk and Bialystok, 2013). The questions were on the proportion of English and Turkish usage in daily life at home.

Receptive vocabulary of the bilinguals

The results of the receptive vocabulary tests (X-lex format) for the bilinguals are given in Table 1 for English and Turkish.

OPEN IN VIEWER

Table 1.

Receptive vocabulary scores in Turkish and English.

	n	min	Max	mean	St. Dev.
Receptive Turkish	100	3900	4650	4337.0	162.0
Receptive English	100	3900	4750	4328.0	208.9

The correlation between the scores in both languages is strong and highly significant (r = .611, p < .001). Figure 1 shows how the scores are related to the age of the participants.

OPEN IN VIEWER

Figure 1.

Receptive vocabulary (X-lex scores) in Turkish and English by age.

For the youngest children the scores are higher in Turkish than in English, but the English scores increase rapidly and equal the Turkish scores at the age of 9–10. Then the English scores increase at a higher rate than the Turkish scores but both scores are still increasing steadily. This pattern is most likely the effect of school attendance (all participants attend English-speaking schools). The most important finding in the given context is that the scores go up in both languages. The correlation between age and receptive vocabulary is significant for Turkish (r = .477, p < .001) and English (r = .77, p < .001).

Productive vocabulary of the bilinguals

In order to establish the reliability of our fluency tests we computed Cronbach’s alpha for the four productive subtests (food, body parts, clothing and colour terms) in each language. Cronbach’s alpha for Turkish is .829 and for English .841, which is a high value for tests which consist of four items only (see Nunnally, 1978). This is a clear indication that the test as a whole is uni-dimensional and that it measures only one trait, namely productive vocabulary in each language. The results of the productive vocabulary tests for the bilinguals are given in Table 2 for English and Turkish.

OPEN IN VIEWER

Table 2.

Productive vocabulary scores in Turkish and English.

	n	min	Max	mean	St. Dev.
Productive Turkish	100	19	65	37.41	12.20
Productive English	100	16	87	42.25	14.01

Receptive and productive vocabulary correlate highly for each language individually (Turkish: n = 99, r = .782, p < .001; English: n = 100, r = .437, p < .001), which is in line with expectations. What is more surprising is that both receptive and productive vocabulary correlate significantly between the languages (receptive Turkish/English: n = 100, r = .611, p < .001; productive Turkish/ English: n = 99, r = .732, p < .001). Productive and receptive vocabularies are clearly related between both languages.

Comparison of productive and receptive vocabulary with the control groups

A comparison of the bilinguals (n = 100) and the monolingual control group (n = 25) for Turkish is shown in Figure 2 (receptive vocabulary) and in Figure 3 (productive vocabulary)

OPEN IN VIEWER

Figure 2.

Bilingual and monolingual scores for receptive vocabulary in Turkish (X-lex scores).

OPEN IN VIEWER

Figure 3.

Bilingual and monolingual scores for productive vocabulary in Turkish (verbal fluency scores).

The bilinguals have a significantly lower vocabulary than the monolinguals for receptive (t = 12.033, df = 123, p < .001, η² = .541, r² = .495) and for productive vocabulary in Turkish (t = 9.22, df = 32.670, p < .001; equal variance not assumed, η² = .409, r² = .409). Figures 4 and 5 show the comparison between the bilinguals and the English monolingual control group (n = 25) for receptive and productive vocabulary in English.

OPEN IN VIEWER

Figure 4.

Bilingual and monolingual scores for receptive vocabulary in English (X-lex).

OPEN IN VIEWER

Figure 5.

Bilingual and monolingual scores for productive vocabulary in English (verbal fluency scores).

The bilinguals also have significantly lower vocabulary sizes for the receptive (t = 4.054, df = 123, p < .001, η² = .118, r² = .118) and the productive vocabulary in English (t = 6.484, df = 122, p < .001, η² = .256, r² = .259). The effect sizes for the differences in Turkish are much larger than for the differences in English, both for productive and receptive vocabulary.

The question is whether the discrepancy between the vocabularies of bilinguals and monolinguals manifests itself also in the TCV. However, a comparison of the total conceptual receptive vocabulary is not possible with the receptive vocabulary size tests used in this study because the yes–no tests in both languages contain partly different test items. There is no bilingual vocabulary test which measures the same concepts in both languages whilst being representative for different frequency layers in the lexica of each language and ensuring that item difficulty in each language is matched. For the productive vocabulary, however, this analysis is less difficult as a conceptual match of the items (food, body parts, clothing and colour terms) is more obvious. We analysed the productive conceptual vocabulary of the bilinguals by adding up the number of items that were named in the fluency test in both languages minus the number of TE, for example counting a certain body part item only once even if it was given in both languages. The results are given in Figure 6. For the control groups their conceptual vocabulary is identical to their total vocabulary.

OPEN IN VIEWER

Figure 6.

Conceptual productive vocabulary of bilinguals and control groups.

There is more variance in the bilingual group, which indicates that there are individual cases below or above the monolingual controls but, overall, the differences between the groups are not significant, either for English or for Turkish. This means that the vocabulary gap that could be identified in each language separately does not exist when the TCV is taken into account. One has to bear in mind that the two languages involved are structurally different, and that there are not many cognates between these languages (see also Meara, 1993).

Vocabulary knowledge and non-verbal IQ scores

We administered RCPM to measure the non-verbal IQ and correlated the IQ scores of our bilingual participants with their vocabulary scores in both languages controlling for age through partial correlation. Table 3 shows the partial correlations of the IQ scores with the vocabulary scores.

OPEN IN VIEWER

Table 3.

Non-verbal IQ scores and vocabulary scores of the bilinguals (n = 96, controlling for age).*

	Receptive Turk	Productive Turk	Receptive Eng	Productive Eng
Non-verbal IQ	r = .266	r = .186	r = .209	r = .188
	p = .008	p = .067	p = .038	p = .064

*

A full set of data was only available for 96 participants due to various organisational reasons (see also Table 4).

The receptive vocabularies in both languages correlate significantly with the IQ scores, whereas the correlations for productive vocabularies and IQ scores approach significance.

Non-verbal IQ scores for bilinguals and monolinguals

The non-verbal IQ scores of the bilinguals and the two monolingual groups are almost identical (mean for bilinguals = 34.25, for monolingual Turkish speakers = 34.2 and for monolingual English speakers = 34.24, and the small differences are far from being statistically significant (one-way ANOVA, F(2, 147) = .023, p = .977). We therefore conclude that in our study there is no general bilingual advantage with respect for non-verbal IQ scores.

However, the picture changes when the parental language dominance for L1 is taken into account (language dominance questionnaire, Dunn and Fox-Tree, 2009). When we split the bilingual group at the median of the parental dominance scores into two subgroups, one with strong L1 dominant parents (DomHigh) and one with less strong L1 dominant parents (DomLow), the group with strong L1 dominant parents outperforms all other groups, including the two monolingual groups, as shown in Figure 7.

OPEN IN VIEWER

Figure 7.

Bilingual non-verbal IQ scores according to parental dominance preferences and monolingual control groups.

An omnibus ANOVA shows that there is an overall difference between the four groups (one-way ANOVA, F(3, 146) = 12.487, p < .001; η² = .217), but a multiple comparison (post hoc Tukey) reveals that the only significant difference between groups is between the bilingual group with high parental dominance for Turkish and the other three groups. The development of the non-verbal IQ scores according to the age of the bilingual children is shown in Figure 8.

OPEN IN VIEWER

Figure 8.

Development of non-verbal IQ scores bilinguals according to their parental language dominance.

Obviously the IQ scores of the children are higher for older children, but the children of parents with a lower dominance preference for L1 start at a lower level and do not seem to catch up with the children of parents with a higher parental dominance preference for L1.

In addition to the dominance questionnaire we administered a questionnaire about language use at home (Luk & Bialystok, 2013) to the parents. A partial correlation (controlling for the age of the children) between the two parental reports, the vocabulary measures and the IQ scores of the children was carried out. The results are shown in Table 4.

OPEN IN VIEWER

Table 4.

Parental language dominance and language use at home and children test scores (n = 96).

Parental preferences	Children’s scores
	Non-verbal IQ	Productive Turkish	Productive English	Receptive Turkish	Receptive English	Conceptual Vocabulary
Dominance	.433**	.405**	.422**	.716**	.303*	.389**
Language use at home	.398**	.367**	.437**	.361**	.368**	.381**

*

p < .01; ** p < .001.

Both questionnaires of the parents correlate significantly with the vocabulary and the non-verbal IQ scores of the children in the same direction. A higher dominance score for Turkish and a higher score for the use of L1 at home goes together with higher receptive and productive vocabulary scores of the children in both languages and with a higher score for the non-verbal IQ test. We also split the group of bilinguals at the median for the language use questionnaire (Luk and Bialystok, 2013). The group with higher use of Turkish at home have significantly higher IQ scores (t = 6.3, df = 98, p < .001). Table 5 shows the differences between the two groups.

OPEN IN VIEWER

Table 5.

IQ scores of children and reported language use by parents.

Group	N	Mean	St. Dev.
More Turkish use at home	50	34.84	0.955
More English use at home	50	33.66	0.917

Although the actual difference in the IQ scores (mean difference 1.18 out of a possible maximum score of 36) seems to be small at first sight, there is a large effect size of Cohen’s d = 1.26. Values for Cohen’s d above 1.0 are rare, but since this statistic is based on the ratio between the numerator (mean difference) and the denominator (Sq. R. of pooled St. Dev.) a low pooled standard deviation can lead to such a high value for the effect size. A multiple regression with ‘Dominance’ and ‘Language use at home’ as independent variables and IQ scores as dependent variables is not possible in the present study because the two independent variables correlate strongly with each other (r = .901, p < .001) and there would be problems with multicollinearity. Parents with dominance preferences for L1 also report using L1 more at home.