Change in home language environment and English literacy achievement over time: A multi-group latent growth curve modeling investigation

Abstract

In most studies investigating the educational outcomes of linguistically diverse students, variables that identify this population have been considered as static. In reality, owing to the dynamic nature of students and their families, students’ home language environments change over time. This study aims to understand how elementary school students’ home language environments change over time, and how longitudinal patterns of English literacy achievement across grades 3, 6, and 10 differ among students with various home language shift patterns in Ontario, Canada. The longitudinal cohort data of 89,609 students between grades 3 and 10 from the provincial assessments were analyzed for changes in their home language environment. A subsample of 18,000 students was used to examine different patterns of relative literacy performance over time and their associations with immigration background and early intervention programming using multi-group latent growth curve modeling. Our findings suggest a strong movement toward an English-dominant home language environment among multilingual students; yet, students whose homes remained as multilingual demonstrated the highest literacy achievement in the early grade as well as the highest improvement in relative performance over time. The paper draws implications for promoting students’ home language, instilling a positive view of multilingual competence.

Keywords

home language,language shift,immigrant students,linguistically diverse students,multilingualism,large-scale testing,literacy achievement,longitudinal analysis,latent growth curve modeling

The educational and literacy outcomes of students from diverse home language environments have been an important concern for educators and policymakers. In most studies investigating this issue, variables that identify this population have been considered as static. In other words, such labels as English language learners, English-as-a-second-language learners, or language minority students, tend to serve as a time-invariant variable, reflecting the view of students’ status as fixed. In reality, however, students’ language use and their home language environment change over time, as a result of ecological interactions among students, their families, and communities. Although it is common for students from immigrant families to change their primary language from that spoken at home (L1) to the socially dominant language (L2) (Wong-Fillmore, 1991), the extent to which they maintain their L1 in their home environments widely varies. For example, some students continue to hear and speak the L1 at home, whereas others gradually increase the use of the L2 with family members and can even lose aspects of their L1 competence.

In one study (Hammer et al., 2009) on children’s language and literacy acquisition, the authors acknowledged that students’ home language environment can change over time (e.g., the use of the L1 in their homes can increase or decrease) and that this can influence students’ L2 literacy development. However, large-scale research is scant regarding how students and their family members in multilingual homes change their language use over time. Under-researched as well is the association of such changes with students’ literacy development over time.

In the present study, we build on Hammer et al.’s (2009) findings and contribute to the current literature by recognizing home language environments as a dynamic (rather than static) variable. Whereas Hammer et al. (2009) used a small-scale dataset covering two years, we used population-level literacy performance data from Ontario, Canada over a seven-year time span in order to understand how students’ home language environments change, and how their relative achievements in English literacy differ among multiple subgroups of dynamic home language environments.

Home language environment and language shift

For many children from immigrant families, reaching the same level of language proficiency in both their L1 and L2 is often challenging. Children who were once proficient in their L1 have been reported to go through a process called language shift once they begin to receive intensive input in the L2 and become linguistically assimilated into the larger society (Flores, 2015; Wong-Fillmore, 1991). Traditionally, language shift was thought to be a process that takes generations; yet, researchers have begun to address how language shift can, in fact, take place during a single generation. Stevens and Ishizawa (2007) argued that, based on their findings that the language repertoires of siblings in the same households may differ, the inter-generation perspective of L1 shift is too simplistic. In the same vein, Pease-Alvarez (2002) acknowledged an intra-generation language shift and criticized the traditional view for characterizing bilingualism as “a temporary inter-generational bridge between monolingualisms” without taking “the complex and dynamic network of sociocultural processes” into account (p. 115).

Research on language shift in the home environment concerns families’ daily interactions, their language ideologies and beliefs, and their goals for shaping language learning and use (King & Fogle, 2017). One of the earliest studies was by Scheff (1965), who examined changes in immigrant families’ home and community language use. Scheff found that home language use was more resistant to change than community language use, but this varied by age, education, length of residence, and income. Although many researchers have focused on parents’ influence on the prevalence of languages spoken at home, children are agentive in influencing their parents’ language behavior, actively shaping their own home language environment. The process may be intuitively viewed as a negotiation between parents and their children (Fogle & King, 2013). Children have been shown to be more likely to use, prefer, and maintain their L1 if they view their parents positively and also see their family as cohesive and egalitarian (Tannenbaum & Howie, 2002). The younger the age of language-minority children when they enroll in majority-language schools, the faster the attrition of the home language, which can hold negative implications for within-family communication (Wong-Fillmore, 1991).

Although the phenomenon of language shift among multilingual children and their families itself has been extensively researched, the dynamic nature of the home language environment has not drawn much attention among educators and policymakers as a factor to consider in relation to academic achievement. In the context of schools, Slavkov (2018) examined registration forms in three Canadian provinces and found that only a small portion conceptualized home language environments as dynamic, by asking multiple language-profiling questions (e.g., first language, primary language spoken at home, other languages); the other forms operationalized home language environment as static (e.g., home language). Furthermore, in the current literature, research that examines the association between changes in the home language environment and educational outcomes over time is scarce.

Related to this intra-generation view of language shift is the idea of complex systems, or dynamic systems, which is gaining traction among many researchers in developmental psychology and education (Fusella, 2013; Larsen-Freeman & Cameron, 2008). These theories focus on changes, dynamism, and interdependencies within a complicated system (e.g., process, person, society). Larsen-Freeman and Cameron (2008) argued that focusing on both individual growth and variability as well as stability can help researchers and educators gain a better understanding of the developing system. They further encourage researchers to use appropriate analysis techniques that can capture variability at different levels and timescales, such as multivariate time-series modeling and growth curve analysis. As such, examining language shift and literacy development from a complex systems perspective can provide valuable insights into multilingual students’ learning processes and help educators construe ways to provide better support for the students.

Variation in literacy achievement by home language

The language shift within individual students and their dynamic home language environments deserves attention in research on literacy development and academic achievement in general. This line of inquiry has the potential not only to further demonstrate the benefits of bi/multilingualism but, more importantly, to inform researchers about the effects of changes in the home language environment on literacy development from a longitudinal perspective.

It is well established that language and literacy skills transfer across languages and that bilingualism and biliteracy have substantial cognitive benefits (Bialystok, 2011; Cummins, 1993). At a minimum, L1 maintenance shows no correlation with L2 literacy development (Nguyen et al., 2001), but a more common finding is that a rich home language environment and parental use of the L1 are associated with high academic achievement, as Sneddon (2000) found in a study of Gujarati-speaking families in London, and Dolson (1985) found for Spanish speakers in Los Angeles.

Similarly, in the context of Ontario, Jang et al. (2013) found that elementary school students who use multiple languages at home transcend early reading achievement gaps and have the most competent reading comprehension skill profile of all home language subgroups (including domestic students) after living in Canada for five years. Findings by Sinclair et al. (2019) also suggested students who use multiple languages at home in grade 6 are likely to have strong literacy profiles and retain those strengths in high school. However, in the latter study, students who exclusively use languages other than English at home (a group that showed high achievement in elementary school) had a tendency toward literacy skill attrition in high school. This implies that such students may benefit from intervention in middle school and early high school. Indeed, Fox and Cheng (2007) and Han and Cheng (2011) found that high school students who immigrated to Ontario and who were learning English as an L2 experienced a complex set of challenges while attempting to meet the literacy standards required for graduation.

Many factors interact to impact immigrant students’ literacy and academic achievement, including their multilingual proficiencies, national origin, educational ambition, school context, their families’ social, human, and economic capital, and the level, length, and nature of acculturation (Gunderson, 2007; Portes & Rumbaut, 2001). Relatedly, Jia et al. (2014) investigated the relationship between English reading comprehension, length of residence, acculturation to mainstream Canadian culture, and heritage enculturation (maintenance of one’s own culture) within a sample of Ontario-based Chinese immigrant adolescents. Length of residence and acculturation significantly predicted increases in reading comprehension, but heritage enculturation did not demonstrate a significant effect in the model. This suggested that acculturation positively impacts reading comprehension in L2, but retaining one’s culture has neither a positive nor a negative impact.

Both cross-sectional and longitudinal studies have examined the association between home language environment and literacy development. Some cross-sectional studies showed no statistically significant correlation between parental use of English in a multilingual home environment and English vocabulary in grade 2 (Gutiérrez-Clellen & Kreiter, 2003) or English grammatical ability in grade 5 (Duursma et al., 2007). With regard to longitudinal studies, Mancilla-Martinez and Lesaux (2011) investigated the relationship between the home language environment of Spanish-speaking language-minority students in preschool, and their vocabulary growth in Spanish and English up to the age of 12 years. Intuitively, preschool students from homes where English was used significantly more than Spanish demonstrated the highest level of English vocabulary, followed by students who used equal amounts of both languages and those who used mostly Spanish at home. However, these latter two groups’ rates of growth were significantly higher, and their rates of deceleration significantly lower, than that of the group who spoke mostly English, significantly reducing the vocabulary gap by the age of 12 years.

The longitudinal study by Hammer et al. (2009), mentioned briefly above, uniquely conceptualized home language as a dynamic, time-variant factor. These authors examined how changes in bilingual (Spanish–English) maternal language usage influenced young children’s bilingual vocabulary development. Mothers’ increased English usage did not significantly impact the children’s English vocabulary growth or emergent English literacy abilities skills, but it was associated with a decreased rate of Spanish vocabulary growth. Although this study offered insightful findings, it was limited to a relatively small sample size (n = 72), covered a short period of time (2.5 years), and had a narrow focus on language usage by mothers. With the current study, we aimed to overcome such limitations.

Latent growth curve modeling and its applications

As already noted, the focus on changes and dynamism within and between individuals necessitates the use of innovative longitudinal analytic methods. Latent growth curve modeling (LGCM) is a flexible modeling approach appropriate for tracking changes over time. Unlike analysis of variance (ANOVA), which treats within-group differences as error variance, LGCM evaluates both inter-individual and intra-individual variability (i.e., differences among individuals and changes within individuals over time) (Preacher, 2010). Specifically, LGCM can be used to examine linear and curvilinear trends of change for individuals as well as the overall sample’s change (Duncan et al., 2013). Under the structural equation modeling framework, LGCM can account for measurement error, evaluate the influence of time-invariant and time-varying covariates, and determine the variation in longitudinal patterns among multiple subpopulations (Duncan & Duncan, 2009).

Figure 1 displays a sample path diagram of an unconditional linear LGCM with three time points. Two latent factors in circles (intercept and slope) are estimated through three manifest variables in rectangles at Time 0, 1, and 2 (with the same time intervals). The model depicted in Figure 1 can be described in the following equations:

y_{0 i} = π_{0 i} + π_{1 i} * 0 + ε_{0 i}

y_{1 i} = π_{0 i} + π_{1 i} * 1 + ε_{1 i}

y_{2 i} = π_{0 i} + π_{1 i} * 2 + ε_{2 i}

where y₀ _i _, y₁ _i , and y₂ _i represent observed outcomes measured at three different time points for individual i; ε₀ _i , ε₁ _i , and ε₂ _i are composite error terms that are composed of both random measurement error and time-specific influence of individual i; π_0i is the fixed intercept factor representing the initial level of the outcome; π_1i is the fixed slope factor representing the rate of change in the outcome over time. The variances of the random effect intercept (ζ_0i) and random effect slope (ζ_1i) in Figure 1 indicate inter-individual differences in initial status and rate of growth or change, respectively. A statistically significant slope factor mean would indicate a presence of change over time on average while its statistically significant variance would indicate significant variability in their growth rate among individuals. The covariance between two latent factors (ψ_i) demonstrates the association between initial status and rate of change. For multi-group analysis, the same latent growth model is simultaneously fitted for different subpopulations, and the initial status and rate of change for each group is estimated separately. Of substantive interest to the present study is whether the parameters related to the latent factors (i.e., π₀, π₁, ζ₀, ζ₁, ψ) significantly differ across groups.

Figure 1.

Path diagram of linear unconditional LGCM.

LGCM has been implemented to evaluate the development of literacy skills, for instance, the growth in morphological awareness and receptive vocabulary (Kieffer & Lesaux, 2012), word reading and productive vocabulary (Mancilla-Martinez & Lesaux, 2010), oral reading fluency (Jimerson et al., 2013; Yeo & Park, 2014), and reading comprehension (Guglielmi, 2008; Lervåg & Aukrust, 2010). The extent to which mediating factors predict initial status and rate of growth can be evaluated by the inclusion of covariates. For instance, Kieffer and Lesaux (2012) found that phonological awareness did not statistically significantly predict the initial status or rate of growth in morphological awareness and vocabulary among Spanish-speaking language-minority students in grades 4–7.

LGCM allows the evaluation of how the initial status, growth trajectories, and influence of covariates vary across groups through multi-group LGCM or dummy-coded covariates. For example, previous studies utilizing LGCM have found that language-minority students often demonstrate lower initial oral reading fluency (Jimerson et al., 2013; Yeo & Park, 2014) and reading comprehension (Lervåg & Aukrust, 2010) than their language-majority peers, although differences in growth trajectories remain inconclusive. The inclusion of covariates in multi-group LGCM allows for the investigation of variables that predict initial status and rate of growth among groups differently. For instance, self-reported L1 proficiency has significantly predicted initial English reading achievement among Hispanic, but not Asian students in grades 8 to 12 (Guglielmi, 2008). Similarly, vocabulary knowledge has been shown to have a significantly stronger relationship with reading achievement growth among language-minority than language-majority students (Lervåg & Aukrust, 2010).

Current LGCM research on literacy exemplifies the capacity for this method to answer longitudinal research questions and to determine differences in growth trajectories or change patterns among multiple subgroups (Guglielmi, 2008; Halle et al., 2012). Few studies to date have specifically investigated longitudinal literacy achievement while incorporating dynamically measured home language environment as a grouping variable.

Current study

The purpose of this seven-year longitudinal cohort study was threefold: (1) to explore the patterns of changes in students’ home language environment over time; (2) to model longitudinal literacy achievement patterns among multiple subgroups with different home language shifts; and (3) to determine the influence of immigration status and English-as-a-second-language (ESL) program support on the relative literacy achievement pattern of each group. The last set of variables—immigration status (Canadian-born vs. first-generation immigrant) and ESL program support (whether or not the student was receiving ESL instruction at G3)—was included as they have been reported to predict literacy outcomes in the previous studies focusing on multilingual students in Ontario (Jang et al., 2013; Sinclair et al., 2019). The present study, inspired by a complex systems perspective (Larsen-Freeman & Cameron, 2008) and Hammer et al.’s (2009) conceptualization of home language as a dynamic variable, answers the following research questions:

How do students and their families change their home language use between grades 3 and 6?

How are changes in home language environment associated with students’ relative literacy achievement patterns between grades 3 and 10?

Do immigration status and ESL program enrollment in grade 3 predict the relative literacy achievement patterns of students with different home language environments?

Method

Participants

We used longitudinal cohort data of students who participated in Ontario’s provincial assessment program administered by the Education Quality and Accountability Office (EQAO). The literacy achievement of students in Ontario is measured by this assessment in the spring of grades 3, 6, and 10 (G3, G6, and G10, respectively). The data utilized in this study are from students who took the G3 assessment in 2010, G6 in 2013, and G10 in 2017. The original datasets included approximately 130,000 students in G3 and G6, and 210,000 students in G10. Students were included in the present study if they satisfied these three criteria: (1) they wrote the tests in all three years; (2) they took the English version throughout (as opposed to in French)¹; and (3) their home language information was available in both G3 and G6. This resulted in 89,609 students (49.6% female) in the final dataset.

Among these 89,609 students, 8.6% were born outside Canada, among which 60.0% had been in Canada for at least five years, and an additional 15.4% had been in Canada for at least three years at the time of the G3 assessment administration. By the G6 administration, 93.0% of non-Canadian-born students had been in Canada for five years or more. With regard to early intervention, 9.5% of all students included in the analysis were enrolled in an ESL program in G3. Among these students, 66.2% were Canadian-born, comprising the early-identified “domestic language learners” (Jang et al., 2013, p. 425). Owing to its longitudinal nature, the data did not include students who arrived in Canada or entered the Ontario school system after G3.

Measures

During the G3 and G6 EQAO literacy assessment administrations, students were asked to complete a questionnaire that included two questions about home language environment: “In which language(s) do people speak to you at home?” and “Which language(s) do you speak at home?” with options of “only English,” “mostly English,” “another language(s) as often as English,” “mostly another language(s),” or “only another language(s).” To be consistent with the classification and notations used by Jang et al. (2013) and Sinclair et al. (2019), students’ home language status was coded in this paper as language(s) heard at home followed by language(s) spoken at home; for example, EnEn (English–English) means hearing mostly or only English and speaking mostly or only English, and OtEq (Other–Equal) means hearing mostly or only another language(s) and speaking both English and the other language(s) equally. Each student was assigned to one of the nine groups (i.e., EnEn, EnEq, EnOt, EqEn, EqEq, EqOt, OtEn, OtEq, OtOt) in G3 and G6, separately.

The English version of the EQAO literacy assessment elicits the same English literacy skills at G3, G6, and G10: in reading, explicit comprehension, implicit comprehension, and making connections; in writing, topic development, and writing conventions. The G3 and G6 reading measures consist of 26 multiple-choice and 10 open-response questions using four reading passages (Cronbach’s α = .84, .88). The G3 and G6 writing measures (α = .79, .81) include eight multiple-choice and six open-response questions. The G10 literacy measure, also known as the Ontario Secondary School Language Test, integrated reading and writing measures into five sections. The reading measure incorporated 31 multiple-choice and five open-response questions (α = .81), with the writing measure including eight multiple-choice and eight open-response questions (α = .77). The EQAO reports a composite score as overall literacy skill on a scale of 0.1 to 4.9 in G3 and G6, or 200 to 400 in G10. Due to the different metrics used between G3–6 and G10, the composite scores in each grade were transformed to z-scores using all students who completed the English version of the test. With a mean of 0 and a standard deviation (SD) of 1, a student’s z-score expresses their relative standing of overall literacy skills among the test-taker population.

Analysis

In order to examine changes in home language environment over time (RQ1), each student’s home language status at G3 and G6 was determined separately. Changes in students’ home language environment between two grades were analyzed descriptively and visually.

To investigate relative literacy performance among multiple groups of home language environment (RQ2) and the predictive power of two covariates (i.e., immigration status and ESL enrollment in G3) (RQ3), we conducted multi-group LGCM. The group membership for each student was determined using the home language status at both G3 and G6. At each time point, students were identified as to whether they were living in an English-dominant (EnEn) or multilingual home (all other eight categories other than EnEn). Then, a dynamic variable that focuses on the type of changes in home language environment between G3 and G6 was created, resulting in four groups: (a) consistently English-dominant (Eng to Eng), (b) English-dominant to multilingual (Eng to Multi), (c) multilingual to English-dominant (Multi to Eng), and (d) consistently multilingual (Multi to Multi).

LGCM analyses were conducted using Mplus Version 8.2 (Muthén & Muthén, 1998–2017). All analyses used maximum likelihood estimation with robust standard errors and a Satorra-Bentler scaled test statistic, or MLM (Satorra & Bentler, 2001), which is robust to the violation of the normality assumption (Muthén & Muthén, 1998–2017). For model building, we started by testing three baseline models: intercept-only, linear, and curvilinear models (Geiser, 2012). Models were evaluated based on multiple fit indices: chi-square goodness-of-fit index, the comparative fit index (CFI, > .95), Tucker-Lewis index (TLI, > .95), root mean square error of approximation (RMSEA, < .08), and standardized root mean square residual (SRMR, < .10) (Cangur & Ercan, 2015). After identifying the best-fitting functional form, the parameters for the latent factors were allowed to differ by group (Curran et al., 2010). We then tested whether the means of the latent intercept factor (initial status in G3) and latent slope factor (rate of change) were invariant across four groups. All 12 possible pairs of the parameter estimates (six pairs for intercept, six pairs for slope) were tested using a series of Wald tests (Wang & Wang, 2012). An a priori significance level of α = .05 was applied. Finally, two covariates (immigration status and ESL enrollment in G3) were incorporated into the model to predict the latent growth factors. The path diagram of the final hypothesized model in a linear form is illustrated in Figure 2. The factor loadings (or, time scores) of the slope factor, which represent values of the time metric, were fixed at 0, 0.43, and 1 (instead of 0, 1, and 2.33) to relate the interpretation of the slope parameters to the entire seven years rather than every three years.

Figure 2.

Final hypothesized model as a linear function with two covariates.

Results

RQ1: How do students and their families change their home language use between grades 3 and 6?

We first report students’ home language use at G3 and G6, cross-sectionally, and then track changes in home language environment for individual students during this time period. Figure 3 displays the compositions of home language environment of 89,609 students in G3 and G6. For each grade, nine boxes represent different home language environments. The boxes in the top row are homes where students hear only or mostly English, whereas the boxes in the bottom row are homes where they hear only or mostly another language(s). Similarly, the boxes on the far-right side are homes where students speak only or mostly English, whereas the boxes on the far-left side are homes where they speak only or mostly another language(s). Boxes located higher and to the right represent students and their families who use more English at home.

Figure 3.

Home language use of students and their families at G3 and G6 (N = 89,609).

In G3, the majority of students (66.1%) heard and spoke only or mostly English (EnEn). The second largest group was EqEq (8.7%), where students heard and spoke English and another language(s) equally, followed by OtOt (8.2%). For these three groups (74.0%), the language(s) they spoke corresponded with the language(s) they heard. These three groups were followed by EqEn (4.8%) and OtEq (4.3%), who spoke more English than they heard. Fewer students were identified as EnEq (3.1%) or OtEn (3.0%), whereas only 1.0% and 0.9% of students were in EqOt and EnOt, respectively.

Moving on to G6, the order of the home language environment groups in terms of the group size remained largely the same. Compared to G3, more students were living in English-dominant homes (EnEn, 70.7%), followed by EqEq (6.8%) with a lower proportion. The proportion of students in EqEn (6.3%) increased by over 30%, becoming the third largest group; on the contrary, the proportion of OtOt (5.6%) decreased by 32%.

Subsequently, we tracked changes in home language environment for individual students by comparing their reported home language use between G3 and G6. Among 81 possible combinations of home language status at two time points (nine groups at G3 times nine groups at G6), the majority of students (54,613, 61.0%) reported their status as EnEn in both G3 and G6, suggesting that their home language environment remained as English-dominant throughout the time period.

In order to focus our investigation on multilingual students, we excluded these constant English-monolingual students and re-examined the data comprised of 34,996 students whose home was multilingual at one or more time points. Figure 4 illustrates the most prominent ten patterns of changes in home language environment, with the thickness of the arrows representing the relative size of groups with the corresponding pattern. Although the home language environment of some multilingual students remained unchanged (7.3% for OtOt, 4.9% for EqEq), many students tended to hear and speak more English than another language(s) at home in G6 compared to G3. For example, 6.6% of the students reported hearing and speaking another language(s) as often as English in G3, but changed to hearing and speaking mostly English in G6 (EqEq to EnEn, 6.6%). Similarly, 4.0% of the students used to hear and speak mostly another language(s) in G3, but ended up hearing and speaking English as often as another language(s) in G6 (OtOt to EqEq, 4.0%). Many students who reported hearing or speaking another language(s) as much as English in G3 found themselves in a mostly English-speaking home environment in G6 (EqEn to EnEn, 6.1%; EnEq to EnEn, 5.1%). Although some students reported their home language environment changed from most English to multilingual (EnEn to EqEn, 5.4%; EnEn to EqEq, 2.7%), students and their families, in general, tended to increase their use of English at home.

Figure 4.

Most prominent patterns of changes in home language environment of multilingual children from G3 to G6 (n = 34,996).

RQ2: How are changes in home language environment associated with students’ relative literacy achievement patterns between grades 3 and 10?

As mentioned earlier, students were classified into four groups based on the patterns of changes in home language environment: (a) Eng to Eng, (b) Eng to Multi, (c) Multi to Eng, and (d) Multi to Multi. The left side of Table 1 provides the means and standard deviations of literacy test scores (in standardized z-scores) with the sample size by group. Owing to the substantial differences in the number of students across the four groups, 4,500 students from each group were randomly selected to comprise a subsample that was used in subsequent LGCM analyses (n = 18,000). The minimal discrepancies of the descriptive statistics between the population data and the subsample (on the right side of Table 1) reflect the representativeness of the subsample. The observed literacy scores between any two time points for each group were strongly correlated (r = .62–.73).

Table 1.

Sample sizes, means, and standard deviations of literacy test performance by group.

	Population data (N = 89,609)				Subsample (n = 18,000)
Group	n	G3	G6	G10	n	G3	G6	G10
Eng to Eng	54,613(61.0%)	0.081(0.932)	0.040(0.907)	0.171(0.907)	4,500(25%)	0.084(0.937)	0.040(0.921)	0.168(0.903)
Eng to Multi	4,586(5.1%)	–0.145(0.990)	–0.099(0.961)	0.043(0.907)	4,500(25%)	–0.144(0.991)	–0.096(0.962)	0.042(0.907)
Multi to Eng	8,701(9.7%)	–0.086(0.952)	0.002(0.899)	0.113(0.874)	4,500(25%)	–0.077(0.943)	–0.006(0.900)	0.108(0.882)
Multi to Multi	21,709(24.2%)	0.060(0.994)	0.242(0.901)	0.354(0.880)	4,500(25%)	0.052(1.003)	0.226(0.901)	0.344(0.885)

Note. Standard deviations are shown in parentheses.

We used multi-group LGCM to examine differences in comparative literacy achievement patterns among students with different patterns of change in their home language environment. Literacy scores showed high kurtosis at all three points (3.14–3.88) and thus violated univariate and multivariate normality. To compensate, an estimator robust to the violation of multivariate normality was used, as mentioned in the previous section. Although homoscedasticity was also violated (p < .001), LGCM can estimate different time-point-specific variances and does not require this assumption (Preacher et al., 2008). As a preliminary step to evaluate whether there was enough variability across students, we calculated the intra-class correlation coefficient (ICC), the proportion of variance that is explained by student demographics rather than within-student longitudinal growth. The ICC was .220, which is large enough to warrant the use of LGCM (Hox, 2010). To determine the appropriate shape of the change-in-rank curve, we fitted three unconditional models with different functional forms to the data before applying multi-group analysis. Although the intercept-only model showed adequate fit, except RMSEA, χ²(4) = 1088.168 (p < .001), CFI = .954, TLI = .966, RMSEA = .123, CI₉₀ [.117, .129], SRMR = .054, the linear model demonstrated better fit, χ²(1) = 10.219 (p = .001), CFI = 1.000, TLI = .999, RMSEA = .023, CI₉₀ [012, .036], SRMR = .004. The curvilinear model with a latent quadratic factor could not be identified due to the limited number of time points in the data (Preacher, 2010).

Based on the linear model, multi-group LGCM modeling was conducted by allowing the estimated latent intercept and slope factors to vary by group. The model fit the data well, χ²(4) = 110.595 (p < .001), CFI = .995, TLI = .986, RMSEA = .077, CI₉₀ [.065, .090], SRMR = .015. This resulting model was used to investigate possible differences in relative literacy achievement patterns among multiple subgroups with various home language shift patterns.

The left side of Table 2 presents the results of the fitted multi-group model with parameter estimates for each group (see Figure 5 for the results in a diagram). The estimated mean intercepts (π₀) indicate that in G3, Multi to Multi outperformed all other three groups with their literacy performance estimated as 0.078 SD higher than the population average. Although the achievement of Eng to Eng (0.053) was close to that of Multi to Multi, the other two groups, Eng to Multi (–0.157) and Multi to Eng (–0.076), performed lower than the average. With regard to the rate of change, the estimated mean slopes (π₁) suggest that literacy achievement of Eng to Eng increased, on average, by 0.099 SD between G3 and G10. This estimate of the rate of relative performance change doubled for Eng to Multi and Multi to Eng (0.195 and 0.184, respectively) and almost tripled for Multi to Multi (0.280), suggesting that the rate of change in relative literacy achievement of Multi to Multi over time was the highest among all four groups.

Table 2.

Results from the fitted multi-group LGCM without covariates (left) and with covariates (right) (n = 18,000).

	Linear model without covariates				Linear model with covariates
	Eng to Eng	Eng to Multi	Multi to Eng	Multi to Multi	Eng to Eng	Eng to Multi	Multi to Eng	Multi to Multi
Factors
Intercept mean (π₀)	0.053***	−0.157***	−0.076***	0.078***	0.051***	−0.142***	−0.069***	0.155***
Slope mean (π₁)	0.099***	0.195***	0.184***	0.280***	0.095***	0.151***	0.163***	0.203***
Intercept variance (ζ₀)	0.591***	0.697***	0.583***	0.643***	0.591***	0.691***	0.572***	0.620***
Slope variance (ζ₁)	0.141***	0.194***	0.115**	0.025	0.140***	0.175***	0.111***	0.013***
Intercept−slope covariance (ψ)	−0.057**	−0.099***	−0.065***	−0.062**	−0.057**	−0.093***	−0.061***	−0.053***
Covariates
Intercept on Born Outside (α₀)					0.078	0.159**	0.257***	0.120***
Slope on Born Outside (α₁)					0.135	0.136**	0.066	0.090**
Intercept on ESL in G3 (β₀)					−0.025	−0.257***	−0.302***	−0.343***
Slope on ESL in G3 (β₁)					0.094	0.247***	0.182***	0.174***
R² in Intercept					0.000	0.010	0.017	0.040
R² in Slope					0.004	0.054	0.029	0.411
Effective error	0.010	0.008	0.010	0.012	0.010	0.008	0.010	0.006
Effective curve reliability	0.936	0.961	0.919	0.681	0.936	0.956	0.916	0.681

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

Note. Intercept and slope variances for the model with covariates (right) refer to residual variances after variability in their prediction by two covariates.

Figure 5.

A path diagram for the fitted multi-group LGCM without covariates, with associated fitted parameter estimates for means (π₀, π₁), variances (ζ₀, ζ₁), and covariance (ψ) for each group (n = 18,000; ***p < .001, **p < .01, *p < .05)

In order to test invariance of the means of latent intercept and slope factors across groups, a series of Wald tests was conducted. Table 3 shows that most of the p-values from Wald tests comparing all 12 possible pairs were significant, except for two: (1) the intercept factor means of Eng to Eng and Multi to Multi (p = .194), and (2) the slope factor means of Eng to Multi and Multi to Eng (p = .501). These results suggest that all the differences in mean initial status in G3 (the intercept factor) and the mean rate of change between G3 and G10 (the slope factor) across groups are statistically significant, except for the initial score in G3 between Eng to Eng (0.053) and Multi to Multi (0.078), and the rate of change for Eng to Multi (0.195) and Multi to Eng (0.184). These findings are further illustrated visually in Figure 6, which displays the estimated relative performance trajectories for each group.

Table 3.

Pairwise Wald test statistics of invariance of latent intercept means (π₀, upper diagonal) and slope means (π₁, lower diagonal) across groups (n = 18,000).

Group	Eng to Eng	Eng to Multi	Multi to Eng	Multi to Multi
Eng to Eng	–	115.03***	45.81***	1.69***
Eng to Multi	34.75***	–	17.32***	137.99***
Multi to Eng	26.52***	0.45***	–	62.53***
Multi to Multi	120.60***	27.06***	33.60***	–

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

Figure 6.

Estimated changes in comparative literacy performance by group (n = 18,000).

The parameter estimates for the variance of intercept and slope (ζ₀, ζ₁) provide information on inter-individual differences both in G3 and in their rate of change over time. Across groups, most of the variance estimates for both intercept and slope were significantly different from 0 (see Table 2 and Figure 5). These results suggest significant inter-individual variability in both the initial status and rank-change rate and, thus, justify the inclusion of covariates (e.g., immigration status and ESL program enrollment in G3 in the present study) (Byrne, 2012). One exception to the significant variance was the variance in slope for Multi to Multi, which indicates the rate of change in rank for students in the Multi to Multi group was rather homogeneous. The intercept-slope covariances for all four groups are negative, indicating that, regardless of the group, the relationship between the initial status at G3 and the rate of change is negative; that is, students with higher scores in G3 tended to have lower rate of change over time and, in turn, initial gaps in achievement within groups tended to decrease over time (Seltzer et al., 2003).

Effective error and effective curve reliability values for each group were reported at the bottom of Table 2 as measures of the precision and reliability, respectively, of the estimated rates of change (Brandmaier, 2020; Brandmaier et al., 2018). Effective error is an unstandardized index that captures how (im)precisely an individual slope estimate measures that individual’s true rate of change in a longitudinal study design, and its inverse is proportional to the precision of measurement. On the other hand, effective curve reliability considers slope variance (ζ₁) and can be interpreted as a standardized effect size. Although the reliability value of Multi to Multi (0.681) appears substantially lower than the other groups (0.919–0.961), this seemingly unsatisfying reliability should be accounted for by relatively small slope variance (0.025), rather than the precision of the instrument (Brandmaier et al., 2018; Singer & Willett, 2003). That is, despite its comparable precision (0.012), detecting inter-individual differences was more difficult for Multi to Multi than for other groups due to its distinctively small individual differences within the group.

RQ3: Do immigration status and ESL program enrollment in grade 3 predict the relative literacy achievement patterns of students with different home language environments?

Two covariates (immigration status and ESL in G3) were added to the model as predictors, resulting in improved model fit compared to the previous model, χ²(12) = 188.999 (p < .001), CFI = .993, TLI = .979, RMSEA = .057, CI₉₀ [.050, .065], SRMR = .015. With regard to the effects of covariates, the intercept and slope estimates on covariates (α₀, α₁, β₀, β₁) on the right side of Table 2 suggest that neither of the two covariates had significant effect on Eng to Eng. Intuitively, students in Eng to Eng are less likely to have immigrated or been enrolled in ESL; the low variability in the covariates in this group must have reduced their predictive power. Yet, for all the other three groups, both covariates were found to be statistically significantly associated with initial status in G3 and rate of rank change. Born Outside was positively associated with literacy achievement in G3 (α₀) across three groups, and these significant, positive associations were also found for the rate of change (α₁) for Eng to Multi and Multi to Multi. These results indicate that, on average, students from (at least at one point) multilingual homes and who were born outside Canada outperformed their Canadian-born peers in G3 and showed a higher rate of change in relative performance over time. Students who were enrolled in ESL in G3 also demonstrated a higher rate of change (β₁) across three groups, although their initial literacy achievement in G3 (β₀) was significantly lower than non-ESL students in G3.

The addition of the covariates to the model resulted in some changes to the means of intercept and slope (π₀, π₁) for these three groups, especially for Multi to Multi. Once the immigration status and ESL enrollment in G3 are controlled for, the advantage of Multi to Multi’s initial performance in G3 becomes stronger (0.078 to 0.155). Their rate of change in rank over time somewhat decreases compared to the previous model (0.280 to 0.203) along with the rates of change of Eng to Multi and Multi to Eng, but still maintains the advantage over the other groups. Yet, the decrease in variances (ζ₀, ζ₁) after adding the covariates was rather marginal across groups. Particularly for Eng to Eng, the residual variances are almost identical to those in the previous model, suggesting that, again, the two covariates explained little variance in the intercept or slope for this group.

The R² values in Table 2 indicate the explained variances in the latent intercept and slope factors by the covariates. Most of the R² values are minimal, particularly for Eng to Eng, the group for which the covariates have little to predict. Contrariwise, for Multi to Multi, the two covariates collectively explain 41.1% of the variation of the slope factor of the previous unconditional model. This implies that the immigration status and ESL enrollment in G3 have remarkably strong prediction power on the rate of change in relative literacy achievement over time for this group.

Discussion and conclusion

The present study was designed to examine changes in students’ home language environment and the association between these changes and their relative literacy achievement patterns. By investigating province-wide longitudinal data, we were able to portray the home language maps of a linguistically diverse province at different time points and delineate how home language use by individual students and their families changes over time. Between G3 and G6, the proportion of students with an English-dominant home environment increased, whereas students hearing or speaking other language(s) at least as often as English (i.e., OtOt, EqOt, OtEq, EqEq) declined. When comparing these two home language maps (one in G3 and the other in G6), linguistic diversity in the family domain appears to decrease as students become older.

Our findings on changes in home language environment from a longitudinal perspective are consistent with the literature that witnessed prevalent, and rapid, intra-generation language shift (Flores, 2015; Swidinsky & Swidinsky, 1997). Eight out of 10 of the most prominent patterns of change that we found were those moving from more use of non-English language(s) towards heavier use of English at home. Overall, nearly half of the students’ homes used English to a greater extent in G6 than in G3, whereas approximately a quarter of our sample used less English during the same period. In the homes of the remaining quarter of the students, the extent to which English was used remained the same. Some interesting patterns of changes involved the direction from English-dominant to multilingual homes led by other family member(s) (e.g., EnEn to EqEn, EnEn to EqEq, considering the former part of the notation refers to the language(s) students hear). A possible explanation of these patterns may be that once parents assumed that their children had become proficient enough in English, they might attempt to support their children’s L1 development.

Our LGCM analysis with four groups suggests that students whose home language environment shifted across time regardless of the direction (i.e., Multi to Eng, Eng to Multi) showed significantly lower literacy performance compared to their peers in the earlier grade, but a higher rate of change in relative achievement over time compared to the consistently English-dominant group (Eng to Eng) whose rate of change in rank was negative. Furthermore, students who maintained their home language (Multi to Multi) had significantly greater initial relative achievement in G3 and a higher rate of change in rank compared to the other three groups. Although their advantage (approximately 0.2 standard deviation in both the initial status and rank-change rate) could be seen as small, we would argue that this difference in population data should not be considered negligible as this interpretation should be only relative rather than absolute. Our findings suggest at a minimum that maintaining home language in the school years may not impede English literacy development of linguistically diverse students; rather, although we cannot make causal inferences based on our analyses, one possible explanation could be that they may benefit from maintaining and developing multilingual competence (Bialystok, 2011).

Examination of the roles of the covariates reveals an interesting finding that immigrant students outperformed their domestic peers in G3 across groups. Considering that 77.3% of these non-Canadian-born students were living in multilingual homes, their higher comparative outcome in English literacy may seem counter-intuitive. Yet, it is important to note that, at the time of G3 administration, the majority of immigrant students had lived in Canada at least for five years and had received formal instruction in English in school for almost five years (two years of kindergarten and three elementary school years). These considerations make it clear that our finding is consistent with that of previous research studies (Cummins, 1981; Jang et al., 2013), which report higher performance of immigrant students after five years of residence in the same geographical context as ours. Furthermore, the fact that students who were born outside Canada and whose home was multilingual a later year (i.e., Eng to Multi, Multi to Multi) demonstrated higher rates of change in rank, may imply the cognitive benefits of bi/multilingualism. Nonetheless, we acknowledge that this result may have been confounded by students’ socioeconomic status, which was not controlled for in the current study. In fact, Sinclair et al. (2019) pointed out an important sociopolitical factor, Canada’s selective immigration policy, as a plausible explanation of higher achievement among immigrant students. Because this policy is highly selective in approving immigration applications from applicants with higher language proficiency, education, and financial assets (Haque, 2017), a fair proportion of multilingual students are likely to have parents with a relatively high socioeconomic background.

Another noteworthy finding is that a relatively large proportion of students receiving ESL support in G3 was Canadian-born (66.2%). When looking by group, 9.3% of Canadian-born, Eng to Multi students were enrolled in an ESL program in G3. That these students who heard and spoke mostly English at home in G3 were identified as having limited English proficiency highlights that not all Canadian-born students from English-speaking homes are fully prepared for academic language demands at school. Early identification and intervention for these students, who have a lower chance of being identified as English learners than their non-Canadian-born peers (McGloin, 2011), might effectively support them in the long term, as evidenced in the current study by the positive relationship between ESL support in G3 and rate of change in comparative literacy achievement.

Due to the nature of Ontario’s provincial assessment system and items included in the student questionnaire, the current study is limited in that the measures were administered at three time points only; students’ home language environment was self-reported by arguably immature students and was not followed in G10; students’ socioeconomic status was not controlled for; and outcome measures were standardized via z-scores in order to place the scores in the same metric. As for the last point, the z-score metric captured students’ relative growth compared to peers at the time of testing but is not truly sufficient for measuring growth as, to measure each student’s growth independent of their peer-to-peer comparison, the underlying assessments across the years must be vertically scaled. Psychometrically, the metric scale utilized in LGCM analyses can greatly influence the results associated with growth trajectories (Goldschmidt et al., 2010). Modeling individual students’ growth trajectories in academic achievement requires vertically scaled test scores for consistent interpretation across time points (Briggs et al., 2008). Item response theory (IRT)-based vertical scaling is recommended, owing to its increased measurement accuracy (Seltzer et al., 1994); however, it requires common items between grades for either separate or concurrent calibration (Kolen & Brennan, 2004). Unless the tests assess students in adjacent grades, it is often practically challenging to use common items for achievement testing. IRT-based vertical scaling was not feasible in the current study because the tests used in the study are not vertically scaled. As such, the rate of changes in relative achievement reported here should be interpreted with caution until the tests themselves are vertically scaled and the results of this study are replicated with an IRT-based, vertically scaled metric.

The present study contributes to the literature in several meaningful ways. One of the contributions of this study is our specific focus on the complex, dynamic nature of students’ home language status, using large-scale longitudinal data. Investigating relative achievement patterns longitudinally, together with changes in home language environment, provides important evidence for identifying how such contextual factors influence educational outcomes over time. Moreover, our application of LGCM to literacy testing highlights the importance of longitudinal modeling of both intra-personal and inter-personal variations in relating individual student differences to differential learning trajectories.

Finally, this paper draws implications for promoting students’ home language, instilling a positive view of multilingual competence. Our LGCM analysis demonstrates that students whose homes maintain a multilingual environment (Multi to Multi) show the highest academic achievement in the early grades as well as the highest rate of change in relative performance over time. These students were found to have a clear, although not large, advantage over other groups, especially when they have an immigrant background and received early intervention services (ESL in G3). Yet, caution must be applied as this study was not designed as a randomized controlled trial. For example, it is possible that parents who are more supportive of education are more invested in the maintenance of the L1 and are less apt to linguistic assimilation.

Yet, it is alarming that, ironically, our findings clearly indicate a movement toward an English-dominant home environment among multilingual students and their families in Canada—a country in which linguistic diversity is supposedly celebrated (Kim et al., 2019). Although language shift from L1 to L2 among language-minority children is viewed as highly detrimental in the existing literature, especially among families in which parents’ L2 skills are limited (Wong-Fillmore, 1991), many immigrant parents have been reported to believe that maintaining the L1 would impede their children’s academic success and, thus, encourage them to maximize the use of L2 even at home (King & Fogle, 2006). Indeed, parents’ perception of, and attitude towards, maintaining the L1 has been known to be one of the most influential factors in their children’s L1 maintenance (Park & Sarkar, 2007; Swidinsky & Swidinsky, 1997). Parents who are learners of L2 (English in the context of this study) have a right to make an informed decision on their home language practices based on rigorous research. We close the paper with a call for more research on the relationship between changes in students’ home language environment and their L2 development, the latter of which is directly linked to their academic achievement. Of equal necessity is more attention from researchers to the factors affecting the shift in home language use in multilingual families, their perspectives about language and identity, and effective support strategies for publicly disseminating the positive aspects of multilingualism.

Footnotes

Declaration of conflicting interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

Note

References

Bialystok

(2011). Reshaping the mind: The benefits of bilingualism. Canadian Journal of Experimental Psychology, 65(4), 229–235. https://doi.org/10.1037/a0025406

Brandmaier

A. M.

(2020). semper: Structural Equation Modeling, power, effect size, and reliability. R package version 0.3.1. https://rdrr.io/github/brandmaier/semper/

Brandmaier

A. M.

von Oertzen

Ghisletta

Lindenberger

Hertzog

(2018). Precision, reliability, and effect size of slope variance in latent growth curve models: Implications for statistical power analysis. Frontiers in Psychology, 9, 294. https://doi.org/10.3389/fpsyg.2018.00294

Briggs

D. C.

Weeks

J. P.

Wiley

(2008, April). Vertical scaling in value-added models for student learning [Paper presentation]. The national conference for value-added modeling, Madison, WI. https://www.colorado.edu/education/sites/default/files/attached-files/BWW_VSVAM_Wisconsin_040708.pdf

Byrne

B. M.

(2012). Structural equation modeling with Mplus: Basic concepts, applications, and programming. Routledge.

Cangur

Ercan

(2015). Comparison of model fit indices used in structural equation modeling under multivariate normality. Journal of Modern Applied Statistical Methods, 14(1), 152–167. https://doi.org/10.22237/jmasm/1430453580

Cummins

(1981). Age on arrival and immigrant second language learning in Canada: A reassessment. Applied Linguistics, 11(2), 132–149. https://doi.org/10.1093/applin/II.2.132

Cummins

(1992). Contexts for second language teaching. Annual Review of Applied Linguistics, 13, 51–70. https://doi:10.1017/S0267190500002397

Cummins, J. (1993). The research basis for heritage language promotion. In M. Danesi, K. McLeod, & S. Morris (Eds.), Heritage language and education: The Canadian experience (pp. 1–21). Mosaic Press.

10.

Curran

P. J.

Obeidat

Losardo

(2010). Twelve frequently asked questions about growth curve modeling. Journal of Cognition and Development, 11(2), 121–136. https://doi.org/10.1080/15248371003699969

11.

Dolson

D. P.

(1985). The effects of Spanish home language use on the scholastic performance of Hispanic pupils. Journal of Multilingual & Multicultural Development, 6(2), 135–155. https://doi.org/10.1080/01434632.1985.9994192

12.

Duncan

T. E.

Duncan

S. C.

(2009). The ABC’s of LGM: An introductory guide to latent variable growth curve modeling. Social and Personality Psychology Compass, 3(6), 979–991. https://doi.org/10.1111/j.1751-9004.2009.00224.x

13.

Duncan

T. E.

Duncan

S. C.

Strycker

L. A.

(2013). An introduction to latent variable growth curve modeling: Concepts, issues, and application. Routledge.

14.

Duursma

Romero-Contreras

Szuber

Proctor

Snow

August

Calderón

(2007). The role of home literacy and language environment on bilinguals’ English and Spanish vocabulary development. Applied Psycholinguistics, 28(1), 171–190. https://doi.org/10.1017/S0142716406070093

15.

Flores

(2015). Losing a language in childhood: A longitudinal case study on language attrition. Journal of Child Language, 42, 562–590. https://doi.org/10.1017/S0305000914000233

16.

Fogle

L. W.

King

K. A.

(2013). Child agency and language policy in transnational families. Issues in Applied Linguistics, 19, 1–25. https://escholarship.org/uc/item/39b3j3kp

17.

Fox

Cheng

(2007). Did we take the same test? Differing accounts of the Ontario Secondary School Literacy Test by first and second language test-takers. Assessment in Education, 14(1), 9–26. https://doi.org/10.1080/09695940701272773

18.

Fusella

P. V.

(2013). Dynamic systems theory in cognitive science: Major elements, applications, and debates surrounding a revolutionary meta-theory. Dynamical Psychology. http://dynapsyc.org/2013/Fusella.pdf

19.

Geiser

(2012). Data analysis with Mplus. Guilford Press.

20.

Goldschmidt

Choi

Martinez

Novak

(2010). Using growth models to monitor school performance: comparing the effect of the metric and the assessment. School Effectiveness and School Improvement, 21(3), 337–357. https://doi.org/10.1080/09243453.2010.496597

21.

Guglielmi

R. S.

(2008). Native language proficiency, English literacy, academic achievement, and occupational attainment in limited-English-proficient students: A latent growth modeling perspective. Journal of Educational Psychology, 100(2), 322–342. https://doi.org/10.1037/0022-0663.100.2.322

22.

Gunderson

(2007). Where have all the immigrants gone? (Research Symposium Special Edition) Contact, 33(2), 118–128. http://contact.teslontario.org/wp-content/uploads/2016/09/ResearchSymposium2007.pdf#page=118

23.

Gutiérrez-Clellen

V. F.

Kreiter

(2003). Understanding child bilingual acquisition using parent and teacher reports. Applied Psycholinguistics, 24(2), 267–288. https://doi.org/10.1017/S0142716403000158

24.

Halle

Hair

Wandner

McNamara

Chien

(2012). Predictors and outcomes of early versus later English language proficiency among English language learners. Early Childhood Research Quarterly, 27(1), 1–20. https://doi.org/10.1016/j.ecresq.2011.07.004

25.

Hammer

C. S.

Davison

M. D.

Lawrence

F. R.

Miccio

A. W.

(2009). The effect of maternal language on bilingual children’s vocabulary and emergent literacy development during head start and kindergarten. Scientific Studies of Reading, 13(2), 99–121. https://doi.org/10.1080/10888430902769541

26.

Han, H., & Cheng, L. (2011). Tracking the success of English language learners within the context of the Ontario secondary school literacy test. Comparative and International Education/Éducation Comparée et Internationale, 40(1), 76–96. https://ir.lib.uwo.ca/cie-eci/vol40/iss1/6

27.

Haque

(2017). Neoliberal governmentality and Canadian migrant language training policies. Globalisation, Societies and Education, 15(1), 96–113. https://doi.org/10.1080/14767724.2014.937403

28.

Hox

J. J.

(2010). Multilevel analysis: Techniques and applications. Routledge.

29.

Jang

E. E.

Dunlop

Wagner

Kim

Y. H.

(2013). Elementary school ELLs’ reading skill profiles using cognitive diagnosis modeling: Roles of length of residence and home language environment. Language Learning, 63(3), 400–436. https://doi.org/10.1111/lang.12016

30.

Jia

Gottardo

Koh

P. W.

Chen

Pasquarella

(2014). The role of acculturation in reading a second language: Its relation to English literacy skills in immigrant Chinese adolescents. Reading Research Quarterly, 49(2), 251–261. https://doi.org/10.1002/rrq.69

31.

Jimerson

Hong

Stage

Gerber

(2013). Examining oral reading fluency trajectories among English language learners and English speaking students. Journal of New Approaches in Educational Research (NAER Journal), 2(1), 3–11. https://www.learntechlib.org/p/48026/

32.

Kieffer

M. J.

Lesaux

N. K.

(2012). Development of morphological awareness and vocabulary knowledge in Spanish-speaking language minority learners: A parallel process latent growth curve model. Applied Psycholinguistics, 33(1), 23–54. https://doi.org/10.1017/S0142716411000099

33.

Kim

Burton

J. L.

Ahmed

Bale

(2019). Linguistic hierarchisation in education policy development: Ontario’s Heritage Languages Program. Journal of Multilingual and Multicultural Development, 41(4), 320–332. https://doi.org/10.1080/01434632.2019.1618318

34.

King

K. A.

Fogle

L. W.

(2006). Bilingual parenting as good parenting: Parents’ perspectives on family language policy for additive bilingualism. International Journal of Bilingual Education and Bilingualism, 9(6), 695–712. https://doi.org/10.2167/beb362.0

35.

King

K. A.

Fogle

L. W.

(2017). Family language policy. In T. L. McCarty & S. May (Eds.), Language policy and political issues in education (pp. 1–13).00 Springer.

36.

Kolen

M. J.

Brennan

R. L.

(2004). Test equating, scaling, and linking: Methods and practices (3rd ed.). Springer.

37.

Larsen-Freeman

D. L.

Cameron

(2008). Research methodology on language development from a complex systems perspective. The Modern Language Journal, 92(2), 200–213. https://doi.org/10.1111/j.1540-4781.2008.00714.x

38.

Lervåg

Aukrust

V. G.

(2010). Vocabulary knowledge is a critical determinant of the difference in reading comprehension growth between first and second language learners. Journal of Child Psychology and Psychiatry, 51(5), 612–620. https://doi.org/10.1111/j.1469-7610.2009.02185.x

39.

Mancilla-Martinez

Lesaux

N. K.

(2010). Predictors of reading comprehension for struggling readers: The case of Spanish-speaking language minority learners. Journal of Educational Psychology, 102(3), 701–711. https://doi.org/10.1037/a0019135

40.

Mancilla-Martinez

Lesaux

N. K.

(2011). Early home language use and later vocabulary development. Journal of Educational Psychology, 103(3), 535–546. https://doi.org/10.1037/a0023655

41.

McGloin

(2011). An achievement gap revealed: A mixed method research investigation of Canadian-born English language learners [Unpublished master’s thesis]. University of Toronto.

42.

Muthén

L. K.

Muthén

B. O.

(1998–2017). Mplus user’s guide (8th ed.). Muthén & Muthén.

43.

Nguyen

Shin

Krashen

(2001). Development of the first language is not a barrier to second-language acquisition: Evidence from Vietnamese immigrants to the United States. International Journal of Bilingual Education and Bilingualism, 4(3), 159–164. https://doi.org/10.1080/13670050108667725

44.

Park

S. M.

Sarkar

(2007). Parents’ attitudes toward heritage language maintenance for their children and their efforts to help their children maintain the heritage language: A case study of Korean-Canadian immigrants. Language, Culture and Curriculum, 20(3), 223–235. https://doi.org/10.2167/lcc337.0

45.

Pease-Alvarez

(2002). Moving beyond linear trajectories of language shift and bilingual language socialization. Hispanic Journal of Behavioral Sciences, 24, 114–137. https://doi.org/10.1177/0739986302024002002

46.

Portes

Rumbaut

R. G.

(2001). Legacies: The story of the immigrant second generation. Russell Sage Foundation.

47.

Preacher

K. J.

Wichman

A. L.

MacCallum

R. C.

Briggs

N. E.

(2008). Latent growth curve modeling (Qualitative applications in the social sciences, 157). SAGE Publications.

48.

Preacher

K. J.

(2010). Latent growth curve models. In G. R. Hancock & R. O. Mueller (Eds.), The reviewer’s guide to quantitative methods in the social sciences (1st ed., pp. 185–198). Routledge.

49.

Satorra

Bentler

P. M.

(2001). A scaled difference chi-square test statistic for moment structure analysis. Psychometrika, 66(4), 507–514. https://doi.org/10.1007/BF02296192

50.

Scheff

T. J.

(1965). Changes in public and private language among Spanish-speaking migrants to an industrial city. International Migration, 3(1–2), 78–85. https://doi.org/10.1111/j.1468-2435.1965.tb00732.x

51.

Seltzer

Choi

Thum

Y. M.

(2003). Examining relationships between where students start and how rapidly they progress: Using new developments in growth modeling to gain insight into the distribution of achievement within schools. Educational Evaluation and Policy Analysis, 25(3), 263–286. https://doi.org/10.3102/01623737025003263

52.

Seltzer

M. H.

Frank

K. A.

Bryk

A. S.

(1994). The metric matters: The sensitivity of conclusions about growth in student achievement to choice of metric. Educational Evaluation and Policy Analysis, 16(1), 41–49. https://doi.org/10.3102/01623737016001041

53.

Sinclair

Jang

E. E.

Vincett

(2019). Investigating linguistically diverse adolescents’ literacy trajectories using latent transition modeling. Reading Research Quarterly, 54(1), 81–107. https://doi.org/10.1002/rrq.220

54.

Singer

J. D.

Willett

J. B.

(2003). Applied longitudinal data analysis: Modeling change and event occurrence. Oxford University Press.

55.

Slavkov

(2018). What is your ‘first’ language in bilingual Canada? A study of language background profiling at publicly funded elementary schools across three provinces. International Journal of Bilingual Education and Bilingualism, 21(1), 20–37. https://doi.org/10.1080/13670050.2015.1126217

56.

Sneddon

(2000). Language and literacy: Children’s experiences in multilingual environments. International Journal of Bilingual Education and Bilingualism, 3(4), 265–282. https://doi.org/10.1080/13670050008667711

57.

Stevens

Ishizawa

(2007). Variation among siblings in the use of a non-English language. Journal of Family Issues, 28(8), 1008–1025. https://doi.org/10.1177/0192513X07301433

58.

Swidinsky

(1997). The determinants of heritage language continuity in Canada: Evidence from the 1981 and 1991 census. Canadian Ethnic Studies, 29(1), 81–98.

59.

Tannenbaum

Howie

(2002). The association between language maintenance and family relations: Chinese immigrant children in Australia. Journal of Multilingual and Multicultural Development, 23(5), 408–424. https://doi.org/10.1080/01434630208666477

60.

Wang

(2012). Structural equation modeling: Applications using Mplus. John Wiley & Sons.

61.

Wong-Fillmore

(1991). When learning a second language means losing the first. Early Childhood Research Quarterly, 6(3), 323–346. https://doi.org/10.1016/S0885-2006(05)80059-6

62.

Yeo

Park

(2014). Developmental differences in curriculum-based measurement (CBM) reading aloud growth rates between English-speaking students and English language learners in grade 8. Exceptionality, 22(2), 91–110. https://doi.org/10.1080/09362835.2013.865528