Speaking English Language Learners - Wiley Online Library

Child Development, March/April 2013, Volume 84, Number 2, Pages 630–646

The Language Growth of Spanish-Speaking English Language Learners Raúl Rojas

Aquiles Iglesias

University of Texas at Dallas

Temple University

Although the research literature regarding language growth trajectories is burgeoning, the shape and direction of English Language Learners’ (ELLs) language growth trajectories are largely not known. This study used growth curve modeling to determine the shape of ELLs’ language growth trajectories across 12,248 oral narrative language samples (6,516 Spanish; 5,732 English) produced by 1,723 ELLs during the first 3 years of formal schooling (M age at first observation = 5 years 7 months). Results indicated distinct trajectories of language growth over time for each language differentially impacted by summer vacation and gender, significant intra- and interindividual differences in initial status and growth rates across both languages, and language-specific relations between language growth and initial status. Implications of ELLs’ language growth are discussed.

Children in the United States who come to school speaking a language other than English and whose English language skills are insufficient to function successfully in English-only classrooms are typically classified as English Language Learners (ELLs). Unlike monolingual children, whose language development follows a continuous and orderly trajectory (Gleason & Ratner, 2009), ELLs undergo varying patterns of first- and second-language acquisition (Kohnert, Windsor, & Ebert, 2009). Current research has begun to focus on language growth trajectories, but primarily in monolingual English children (e.g., Farkas & Beron, 2004; Hadley & Holt, 2006; Pan, Rowe, Singer, & Snow, 2005; Rice, Hoffman, & Wexler, 2009). An emerging body of research on ELLs has demonstrated that their overall language development is positive, and within-language associations between lexical and morphosyntactic development have been identified (e.g., Carias & Ingram, 2006; Conboy & Thal, 2006; Kohnert, Kan, & Conboy, 2010; Vagh, Pan, & Mancilla-Martinez, 2009). However, the shape, direction, and rate of ELLs language growth trajectories are largely not known (Hammer, Lawrence, & Miccio, 2008).

This research has been supported in part by Grant R305U010001 (jointly funded by the National Institute of Child Health and Human Development, and the Institute of Education Sciences), and by a New Century Scholars Doctoral Scholarship from the American Speech-Language-Hearing Foundation. Correspondence concerning this article should be addressed to Raúl Rojas, Callier Center for Communication Disorders, University of Texas at Dallas, 1966 Inwood Road, Dallas, TX 75235. Electronic mail may be sent to [email protected].

To understand the growth trajectories of ELLs’ languages and the factors that influence those trajectories, we must characterize change over multiple points in time within each language as a whole, and within the individual domains of each language. Studies that measure language development at only two points in time contribute limited information concerning trajectories of growth (Huttenlocher, Haight, Bryk, Seltzer, & Lyons, 1991), relative to studies that measure change more frequently. In essence, what must be measured to understand language development in the ELL population is the growth, over multiple observations, of a continuously changing and complex system composed of two languages and their individual linguistic domains. The purpose of this study, which is the first in a series of planned longitudinal studies, was to model the language growth of Spanish-speaking ELLs in Spanish and English (independently of one another) during the early school years (kindergarten to second grade) using an approach that determined: (a) the functional form (henceforth shape) of language growth including time-varying and time-invariant predictors, (b) the individual variability of ELLs language growth, and (c) the impact of initial status on language growth. The present study advances our knowledge of language growth in Spanish-speaking ELLs by estimating initial status and growth rates in Spanish © 2012 The Authors Child Development © 2012 Society for Research in Child Development, Inc. All rights reserved. 0009-3920/2013/8402-0019 DOI: 10.1111/j.1467-8624.2012.01871.x

The Language Growth of Spanish-Speaking ELLS

and English during the first 3 years of formal schooling, and determining the impact of summer vacation and gender on language growth. Shape of Language Growth Elman et al. (1999) developed a growth trajectory taxonomy composed of three nested dimensions that could account for varying and complex growth patterns over time: linearity, direction, and continuity. Linearity specifies whether growth is gradual, periodic, and uniform (linear) or whether it is characterized by instantaneous, exponential accelerations or decelerations (curvilinear). Direction specifies whether growth steadily increases or decreases (monotonic) or whether it is characterized by alternating periods of positive and negative growth (nonmonotonic). Continuity specifies whether quantitative changes in growth are consistent (continuous) or whether there are periods characterized by sudden shifts of inconsistent positive or negative growth (discontinuous). Time-varying predictors can influence the overall shape of language growth trajectories, sometimes resulting in discontinuous growth (e.g., Perkins, Brutten, & Grass, 1996; Wu, West, & Hughes, 2008). One such time-varying predictor is the gap in schooling that occurs between the spring and fall semesters (summer vacation). Evidence suggests that the summer vacation has negative effects on the growth of math and reading skills (e.g., Cooper, Nye, Charlton, Lindsay, & Greathouse, 1996; McCoach, O’Connell, Reis, & Levitt, 2006), with math being more adversely affected than reading. A 2-year study with Spanish-speaking preschool ELL children (Hammer et al., 2008) suggested a more complex relation between the summer vacation schooling gap and receptive language growth. Hammer et al. (2008) found that summer vacation positively impacted receptive language skill growth in Spanish and English (standard score growth) for participants who had demonstrated negative growth of their languages during the 1st year, and negatively impacted the growth in both languages (standard score decline) for those that had previously demonstrated positive language growth. However, the preceding patterns of growth during the fall and spring semesters of the 1st year continued during the 2nd year. These contrasting patterns of change over time resulted in trajectories with periods of discontinuous growth during the summer. The present study considered the effect of not having had schooling during the summer vacation as a time-varying predictor.

631

Gender is often used as a time-invariant predictor in studies of early language development in children. In monolingual children, evidence suggests that girls exhibit advanced linguistic development relative to boys, a pattern that appears to attenuate around age 3;0 (Bornstein, Hahn & Haynes, 2004), and to reappear between 10;0 and 11;0 (Coates, 1993). Longitudinal work on ELLs’ language growth suggests different gender-based patterns. In a 3-year, longitudinal study of the impact of maternal language use on vocabulary and early literacy skills of Spanish-speaking children during Head Start and kindergarten, Hammer, Davison, Lawrence, and Miccio (2009) found that gender was not significantly related to vocabulary or early literacy skill growth in either Spanish or English. Hammer et al. (2009) concluded that the lack of gender-based differences might have been associated with the participants’ immersion in English-only programs. In contrast, Uchikoshi (2006) found that in kindergarten, ELL boys had relatively higher initial levels and growth of English receptive and expressive vocabulary skills than girls. Uchikoshi speculated that the boys’ vocabulary advantage was related to the increased frequency of interactions with English speakers in the community, and with Hispanic parents’ emphasis on the importance of learning English. Such inconsistent findings establish gender as a critical timeinvariant variable for the present study. Intra- and Interindividual Variability Individual differences over time can be reflected within (intraindividual variability) and between (interindividual variability) individuals. Child development research has traditionally underemphasized the importance of individual differences (Siegler, 2002; van Geert & van Dijk, 2002), but a number of studies have highlighted the importance of studying individual differences in language growth. A seminal study (Fenson et al., 1994) of variability in early communication development based on a large cross-sectional sample of MacArthur Communicative Development Inventories (CDI; Fenson et al., 1993) showed significant individual differences in English-speaking infants and toddlers, and some studies have addressed intra- and interindividual variability in growth curve modeling of monolingual longitudinal data. Hadley and Holt (2006) reported significant interindividual differences in the onset and growth rates of tense marking in slowly developing monolingual English children. Rice, Wexler, and Hershberger (1998),

632

Rojas and Iglesias

Rice, Redmond, and Hoffman (2006), and Rice et al. (2009) found differential patterns of intra- and interindividual differences in morphosyntactic measures of children with specific language impairment (SLI). Such findings highlight the need to consider the heterogeneity of language learners, and to extend language growth findings beyond initial status and growth rates. In contrast, the majority of growth curve modeling studies of bilingual language development (e.g., Kan, 2010; Uchikoshi, 2006; Vagh et al., 2009), have focused on interpreting fixed effects (initial status and growth rates) without interpreting variance components (intraand interindividual variability).

ing into adolescence, and might never catch up to their TD peers. However, given that the interceptslope covariance was negative and nonsignificant (i.e., the initial status and growth trajectories were not systematically related), it can be speculated that the SLI participants might not fall any further behind. In another study of children with SLI, Rice et al. (2006) reported another negative and nonsignificant intercept–slope covariance for mean length of utterance in morphemes (MLUm). The work of Rice and colleagues (Rice et al., 1998; Rice et al., 2006; Rice et al., 2009) demonstrates the importance of considering the impact of initial status on growth trajectories and of reporting intercept–slope covariance data to gauge this relation.

Impact of Initial Status on Growth Differences at the onset of growth can result in systematically different growth trajectories over time. This phenomenon is detected by the intercept–slope covariance in growth curve modeling data, which estimate the strength and direction of the relation between the initial status of growth (intercept) and the rate of growth (slope), after controlling for group or program membership (Singer & Willett, 2003). The impact of initial status on growth trajectories is often found in the educational research literature (e.g., Ding & Davison, 2005; Seltzer, Choi, & Thum, 2003). For example, Ding and Davison (2005) compared math skills in low- versus high-achieving students and found a negative and significant intercept–slope covariance between math skills at initial status and growth over a 4-year period in two cohorts of primarily monolingual children. These findings suggested that low-achieving students at initial status demonstrated faster growth rates, on average, than high achievers. The findings also suggested that the initial differences in achievement would diminish over time. Examples of intercept–slope covariance data can also be found in the language development literature. Rice et al. (1998) found that verb tense marking was significantly higher for typically developing (TD) than for SLI children at the beginning of data collection, and that this difference in initial status was maintained over time. Although Rice et al. (1998) did not report the covariance data, the findings presume a positive and significant intercept–slope covariance. In a follow-up study (Rice et al., 2009), TD and SLI participants again demonstrated parallel growth trajectories, and TD participants again had significantly higher initial status. Rice et al. (2009) concluded that older SLI participants show slower growth of finiteness mark-

Summary The development of language is a dynamic and complex process, involving linguistic domains that change over time that result in varying developmental (growth) trajectories. However, most language growth research, whether on monolingual or ELL children, has focused on initial status and growth rates (fixed effects) within one linguistic domain, and with less emphasis on inter- and intraindividual differences (variance components) and the systematic relation between initial status and growth. The present study was designed to advance understanding of ELLs’ oral language development by examining the change over time (initial status, growth rate, direction, and shape) of multiple linguistic domains in Spanish and English and the predictive roles played by summer vacation and gender. This study also addresses individual differences in the language growth rates and the extent to which initial levels of language predict growth over time in specific oral language skills. Three nested research questions were addressed. First, what are the overall shapes of growth in Spanish and English, and what predictive roles are played by summer vacation and gender? Second, are there significant intra- and interindividual differences in initial status and growth rates of Spanish and English? Third, what impact does initial status have on the growth trajectories in Spanish and English?

Method Parent Study The present study used secondary data from the longitudinal subproject of the Biological and Behavioral Variation in the Language Development of


Spanish-Speaking Children study (Francis et al., 2005), a large-scale, multilevel investigation of factors that might influence the development of literacy and language skills of native Spanish-speaking ELLs during the initial years of formal schooling. Participants of the longitudinal subproject were enrolled in 1 of 40 schools and 93 classrooms in California (Long Beach area) and Texas (Austin, Houston, and Brownsville) and met the following inclusionary criteria: (a) designated by the school district as ELL; (b) attended a school in which at least 40% of the school population was Latino, representative of typical educational environments where the majority of Spanish-speaking ELLs are enrolled (Swanson, 2009); (c) attended a school in which at least 30% of the kindergarten population was identified as Limited English Proficient, as designated by district-based, language proficiency testing; (d) had never received nor were receiving special education services; and (e) were enrolled in specific programs/ models of bilingual instruction (e.g., structured English immersion; transitional bilingual). The proportion of participants across program type was consistent across waves of observation. At school entry (fall of kindergarten), 42.3% of participants’ mothers voluntarily completed a survey to gauge relative language use (1 = only Spanish, 2 = mostly Spanish, 3 = Spanish and English equally, 4 = mostly English, 5 = only English) at home. The mean relative language use at home was 1.49 (SD = 0.81), indicating that mothers communicated with their children mostly in Spanish. Longitudinal Sampling Narrative language samples were collected over a 3-year period, involving six waves of data collection between the fall of kindergarten to the spring of second grade. Language sampling occurred during October and May, to ensure that participants had been exposed to at least 1 month of academic instruction. Oversampling, common in large-scale longitudinal studies (McCoach et al., 2006), was implemented to compensate for participant attrition. After the first two waves, every child in every classroom who met the criteria for the subproject was sampled, rather than one-for-one replacements for each attriter. No attempt was made to replace participants with children who had identical educational experiences. Thus, the number of children in each target classroom who met the criteria for the sub-project determined the number of children added throughout the duration of the study.

633

Although the longitudinal sampling strategy differs from the definition of a classical and complete longitudinal data set (i.e., the same measures collected from the same participants during the same number of waves), it resulted in a large-scale, longitudinal data set in which the same measures were collected from a range of participants who provided data across a differing number of waves. This type of longitudinal data set was appropriate for the analytic method (growth curve modeling) used in this study, which accommodates varying schedules of data collection and missing data. Participants Additional inclusionary criteria for participants of the present study included: (a) gender information was available, (b) not retained in any academic grade, (c) within the expected age range during the fall semester of each grade (4;9–7;1–kindergarten; 5;9–8;1–first grade; and 6;9–9;1–second grade), and (d) contributed at least one narrative language sample containing at least five complete and intelligible utterances, with 75% or more of the total number of different words (NDW; uninflected word roots) produced in the target language. A total of 328 participants from the longitudinal subproject were excluded based on these criteria: 226 who lacked gender information, 94 who failed one or more academic grades, and 7 who exceeded the age range (2 at kindergarten; 5 at first grade; 1 at second grade). Excluding these 328 participants resulted in the exclusion of 1,492 narrative language samples. An additional 72 samples were excluded from the data set: 37 had fewer than five complete and intelligible utterances, and 35 samples had less than 75% NDW in the target language. This study’s final longitudinal data set consisted of 12,248 oral narrative language samples (6,516 Spanish; 5,732 English) produced by 1,723 ELLs. The majority of the participants contributed at least three or more language samples (three or more waves) in either Spanish (69% of participants) or English (63% of participants). The mean age for the participants was 5;7 years (SD = 0;4) at initial status (fall of kindergarten). The total gender distribution across the longitudinal data set was 50.3% girls (n = 867) and 49.7% boys (n = 856). Table 1 provides descriptive statistics for each wave. Table 1 demonstrates that the percentage of participants who provided retells in both target languages steadily increased from the fall of kindergarten (42%) to the spring of second grade (nearly 90%). This pattern is important, as it indicates that the majority of participants developed the ability

634

Rojas and Iglesias

Table 1 Story, Participant (Chronological Age and Percentage of Participants That Contributed Retells in Spanish and English, Spanish Only, and English Only), and Language Sample Descriptive Statistics per Semester in Spanish and English

Story Participants Age (years; months) Retell%: Sp. and Eng. Retell%: Sp. only Retell%: Eng. only Total samples Spanish samples: M (SD) MLUw NDW WPM English samples: M (SD) MLUw NDW WPM

K-Fall

K-Spring

1st-Fall

1st-Spring

2nd-Fall

2nd-Spring

FWAY 1,070 5;7 (0;4) 42.43 45.14 12.43 1,524

FGTD 1,120 6;2 (0;4) 62.86 27.41 9.73 1,824

FOHO 1,302 6;7 (0;4) 71.74 20.28 7.99 2,236

OFTM 1,325 7;2 (0;4) 82.64 11.47 5.89 2,420

FWAY 1,178 7;7 (0;4) 85.23 8.15 6.62 2,182

FGTD 1,088 8;2 (0;4) 89.52 4.5 5.97 2,062

4.69 (0.83) 55.26 (18.66) 56.87 (19.75)

5.68 (1.0) 71.16 (20.06) 63.57 (21.05)

5.53 (0.84) 80.14 (23.01) 65.87 (20.54)

5.96 (0.96) 74.87 (18.84) 64.76 (19.8)

5.8 (0.84) 79.75 (18.35) 67.48 (20.64)

7.06 (1.11) 91.78 (19.54) 82.11 (21.65)

5.55 (1.18) 53.98 (21.83) 68.24 (24.06)

6.47 (1.28) 65.38 (24.27) 69.12 (25.49)

6.3 (1.12) 73.03 (27.34) 73.83 (25.73)

6.93 (1.05) 73.12 (23.33) 77.7 (23.83)

6.79 (0.82) 81.71 (22.37) 88.37 (24.0)

7.97 (1.12) 99.95 (23.14) 93.37 (22.67)

Note. FWAY = Frog, Where Are You? (Mayer, 1969); FGTD = Frog Goes to Dinner (Mayer, 1974); FOHO = Frog on His Own (Mayer, 1975a); OFTM = One Frog Too Many (Mayer, 1975b); Sp. = Spanish; Eng. = English; MLUw = mean length of utterance in words; NDW = number of different words; WPM = words per minute.

to provide retells in both languages over time. Miller et al. (2006) who reported findings from a cross-sectional subproject of the same large-scale, parent study, noted that ELLs contributed a retell only in one language because they were either absent on the date of data collection, were unable to provide a retell in the other language, refused to comply with the retell task, or the retell sample was unusable due to technical complications. As with Miller et al. (2006), the specific reason why some participants did not provide retells in both languages was not available in the longitudinal data set used in this study. It is important to reemphasize that the purpose of this study was to model the language growth of Spanish-speaking ELLs in Spanish and English independently of one another. The mean differences reported in Table 1 are provided for descriptive purposes only; comparisons across languages are not appropriate due to language-specific differences such as PRO-Drop in Spanish (Bedore, 1999) and the ubiquitous use of prepositional satellites in English (Talmy, 1991), which have an impact on the language measures analyzed. Narrative Language Sampling Protocol Participants were asked to retell, in Spanish and in English, one of a series of wordless picture storybooks commonly known as Frog Stories, that have been shown to be nonbiased and valid for eliciting narrative language samples from participants of various

linguistic and cultural backgrounds (e.g., Özçalişkan & Slobin, 1999; Papafragou, Massey, & Gleitman, 2002; Pavlenko, 2009) including Spanish speakers (e.g., Heilmann, Miller, and Nockerts, 2010; Sebastián & Slobin, 1994). Four Frog Stories were used: (a) Frog, Where Are You? (Mayer, 1969); (b) Frog Goes to Dinner (Mayer, 1974); (c) Frog on His Own (Mayer, 1975a); and (d) One Frog Too Many (Mayer, 1975b). To promote task familiarity in the native language of the children, all initial sessions were administered in Spanish. The data collection procedure was replicated in English approximately 1 week later. All examiners were trained in the narrative elicitation protocol, and were proficient Spanish-English speakers. The narrative language elicitation protocol required the examiner to sit across from the child participant, and provide a prescripted narrative of the story in the target language while the child viewed the book. The examiner then gave the book to the child and asked her or him to retell the story (“Ahora, cuéntame lo que pasó en este cuento”/ “Now, tell me what happened in the story”). Examiner feedback was limited to backchannel responses (e.g., “aha”/“sí”/“yes”) with rising inflection or the repetition of the participant’s preceding last utterance to encourage the participant to continue narrating. Examiners were instructed not to provide any additional information or answer any questions from the participants. The narrative language samples were recorded onto digital media and transcribed orthographically


by trained Spanish-English bilingual transcribers following the conventions for bilingual oral language samples (Heilmann et al., 2008; Rojas & Iglesias, 2009) using the Systematic Analysis of Language Transcripts Research 2010 (SALT; Miller & Iglesias, 2010) software program. Modified communication units (MC-units), originally proposed by Gutiérrez-Clellen and Hofstetter (1994) for Terminable Units in Spanish, were used to segment narratives in Spanish to accommodate for the PRO-Drop nature of Spanish. Although English is not a PRODrop language, narratives in English were also segmented using the MC-unit segmentation method to maintain segmentation consistency across measures in both languages. Protocol accuracy (adherence to SALT transcription conventions) and transcription accuracy (segmentation of words and utterances) were calculated for 20 English and 20 Spanish language samples, randomly selected from the parent study. Based on independent ratings, protocol accuracy ranged from 98% to 100% in English and 94% to 99% in Spanish. Transcription accuracy values on orthographic transcriptions ranged from 90% to 98% in English and 91% to 99% in Spanish (Miller et al., 2006). Outcome Measures: Expressive Language Skills Three dependent variables (outcome measures) were selected to quantify the morphosyntactic domain, the lexical domain, and the global integration of these and other domains during narrative production. These outcome measures were generated from the narrative language sample transcripts in each language using SALT Research 2010 software (Miller & Iglesias, 2010): (a) mean length of utterance in words (MLUw-Spanish; MLUw-English), (b) number of different words (NDW-Spanish; NDWEnglish), and (c) words per minute (WPM-Spanish; WPM-English). MLUw, NDW, and WPM have consistently been shown to be developmentally sensitive measures (Heilmann, Miller, et al., 2010); they also are positively related to the bilingual reading achievement of ELLs (Miller et al., 2006). MLUw was used as a general measure of the morphosyntactic domain. Relative to MLU in morphemes (MLUm), MLUw is preferred in crosslinguistic and bilingual research as it is unaffected by cross-linguistic variation in inflectional morphology (e.g., Gutiérrez-Clellen, Restrepo, Bedore, Peña, & Anderson, 2000). Relative to other morphosyntax measures, MLUw has been found to be particularly useful for assessing Spanish morphosyntax (Bedore & Peña, 2008). This measure also enabled cross-lan-

635

guage consistency to be maintained. MLUw was calculated by summing the total number of intelligible words produced in the sample, and dividing that total by the number of complete and intelligible utterances. NDW, the total number of different uninflected word roots, was used to represent the lexical domain. NDW reflects the diversity of participants’ vocabularies (Golberg, Paradis, & Crago, 2008) and is significantly correlated (r = .71) with age (Miller, 1987; Miller & Chapman, 1981). In addition, NDW is a developmentally sensitive measure of narrative productivity for Spanish-English bilingual children (Uccelli & Paéz, 2007), and it is positively associated with MLUw in bilingual speakers (Bedore, Fiestas, Peña, & Nagy, 2006; Kohnert et al., 2010; Simon-Cereijido & Gutiérrez-Clellen, 2009). NDW was calculated by adding the total number of different uninflected word roots in the target language for that sample; word roots in the nontarget language were excluded. The number of WPM, a measure of verbal fluency, was used to represent the overall integration of multiple domains (e.g., articulatory speech rate, morphosyntax, lexicon) necessary for oral language production (Miller et al., 2006). Although WPM is not as well established as MLU or NDW, it is increasingly being used in language sample analyses (e.g., Heilmann, Nockerts, & Miller, 2010; Price, Hendricks, & Cook, 2010; Tilstra & McMaster, 2007), and it is positively correlated with the age of ELL children (Miller & Heilmann, 2004). WPM was calculated by dividing the total number of words produced in the target language by the duration of that language sample in minutes and seconds. Covariates: Summer Vacation and Gender One of the principal research aims of this study was to examine whether or not summer vacation and gender optimized model fit and predicted growth over time. Summer vacation, a time-varying covariate, modeled discontinuous growth over time by accounting for the potential effect of not having had schooling during each summer preceding each fall semester, and having had schooling preceding each spring semester. Gender (the effect of being a girl), a timeinvariant covariate, was included to identify potential gender-based differences in initial status and growth rates from kindergarten to second grade. Analytic Approach Growth curve modeling was conducted by using IBM SPSS Statistics 18.0 for Mac (SPSS Inc., 2010),

636

Rojas and Iglesias

using the maximum likelihood estimation method to handle missing data and to estimate the fixed effects (initial status; growth rates) and variance components (intra- and interindividual variance; intercept–slope covariance) for each outcome measure in Spanish and English. Growth curve models (GCMs) offer several advantages for longitudinal data analyses including: (a) the shape of growth trajectories can be modeled to be linear or curvilinear, (b) growth trajectories can indicate positive growth or decay (negative growth) over time, (c) intra- and interindividual patterns of change are simultaneously estimated, (d) descriptive and predictive (with time-varying or time-invariant covariates) information is provided, (e) time-structured and time-unstructured data collection schedules can be accommodated, and (f) all participants, even those with only one observation are included in the analyses as missing data are handled extremely well by using either maximum likelihood or multiple imputation (Baraldi & Enders, 2010; Singer & Willett, 2003). For the present study, growth curve modeling was used to profile the growth of ELLs’ oral language skills in Spanish and in English over a 3-year period. Academic semester was the time metric and waves of measurement were centered relative to the fall of kindergarten as initial status (the starting point of language growth). A series of GCMs of continuous and discontinuous growth were tested for each outcome variable (MLUw, NDW, and WPM) in each language. Model testing for each measure involved the construction of a series of GCMs, calculation of the within- and between-person variances accounted for by each model, and the comparison of goodness-of-fit indices across models to determine the best fitting model for each outcome measure in Spanish and English (Singer & Willett, 2003). The final GCM for each measure accounted for the most variance (i.e., highest pseudo-R2s, representative of the highest proportional variance reductions) and/or achieved the best fit (i.e., lowest goodness-of-fit indices) to explain the longitudinal data. The fixed effects from the final GCM for each outcome measure were used to generate the prototypical growth curve trajectories for each language. Prior to growth curve model testing, exploratory analyses consisting of linear regression individual and fitted growth plots were examined for each outcome measure (MLUw, NDW, and WPM) in Spanish and English. These exploratory analyses informed GCM construction by suggesting which additional covariates (time varying and time invariant) could account for wide ranges of observed

residual variation in intercepts and slopes. Furthermore, the individual and fitted growth plots suggested modeling of linear and curvilinear growth across the study’s six waves of measurement that could account for varying patterns of positive and negative growth. Curvilinear growth was modeled using quadratic and cubic polynomial functions, incorporated as time-varying covariates. In addition, discontinuous growth (linear growth with expected discontinuities) was modeled using summer vacation as a time-varying covariate. The first model tested for each outcome measure in Spanish and English was the unconditional means (UM) model. The UM model, an interceptonly model without the effect of time, served as the baseline model to test the proportional variance reduction and model fit of nested, unconditional growth (UG) models. UG models were growth curve models that included the effect of time (academic semester) as a time-varying covariate, without any additional covariates. Linear, quadratic, and cubic UG models were tested as nested models (the UG-linear model was nested within the UGquadratic model; the UG-quadratic model was nested within the UG-cubic model) with all possible combinations of fixed and randomly varying slopes to determine which UG model demonstrated the greatest proportional variance reduction (relative to the baseline UM model) and best model fit. If any UG model failed to converge, indicating that the sample data were highly unbalanced or there was an overabundance of missing data, the slopes were set as fixed (Singer & Willett, 2003; Vagh et al., 2009). The smallest 2 log-likelihood deviance statistic (2LL), which is intended to compare model fit and the mean change in deviance across nested GCMs (Nakamoto, Lindsey, & Manis, 2007; Singer & Willett, 2003), determined the best fitting UG model, which in turn served as the baseline model to test the proportional variance reduction and model fit of conditional growth (CG) models. The 2LL differences between models were confirmed using v2 distribution testing. Two sets of CG models were tested to determine whether or not additional proportional variance reduction and better model fit, relative to the best fitting UG model, could be achieved by adding time-invariant covariates and/or time-varying covariates. The first set was a series of CG models with gender (time-invariant covariate) added to determine its effect on initial status, and every potential interaction of Gender 9 Slope was also tested to determine gender’s effect on linear, quadratic, and cubic growth. The second set was a


series of discontinuous CG models with summer vacation (time-varying covariate) added to determine the best fitting discontinuous CG model first, by testing all possible combinations of fixed and randomly varying slopes. The effect of gender was then added to the best fitting discontinuous CG model to determine its effect on initial status, and every potential interaction of Gender 9 Slope was also tested to determine gender’s effect on linear and summer vacation growth. The final CG model (determined from these two sets of non-nested, CG models) achieved the smallest Schwarz’s Bayesian information criterion (BIC), which is recognized as the most stringent goodness-of-fit index and is intended to compare model fit across non-nested GCMs (Singer & Willett, 2003).

637

Results The growth of language will be described first for Spanish, then for English, followed by the prototypical growth trajectories in each language. Growth in Spanish Growth curve model testing of MLUw, NDW, and WPM in Spanish resulted in a number of common outcomes, summarized in Table 2. First, the CG model of cubic growth with the effect of gender (CG – CB + G; see Model E in online supporting information Appendix S1 and Tables S1.1–S1.3) was the best fitting model for all outcome measures, indicating curvilinear growth over time of Spanish.

Table 2 Final Growth Curve Model Parameter Estimates for Mean Length of Utterances in Words (MLUw), Number of Different Words (NDW), and Words per Minute (WPM) in Spanish MLUw Fixed effects: c (SE) Intercept Linear slope Quadratic slope Cubic slope Gender (G) G 9 Linear Slope G 9 Quadratic Slope G 9 Cubic Slope Variance components: r (SE) L1: Within-person variance L2: B/w-person intercept L2: B/w-person linear slope L2: B/w-person quadratic slope L2: B/w-person cubic slope Covariance (r02, r12) Covariance (r02, r22) Covariance (r02, r32) Covariance (r12, r32) Covariance (r22, r32) Covariance (r22, r32) Proportional variance reduction L1: Within-person variance L2: B/w-person intercept L2: B/w-person linear slope L2: B/w-person quadratic slope L2: B/w-person cubic slope Goodness-of-fit 2LL BIC

c00 c10 c20 c30 c01 c11 c21 c31

4.56*** (0.04) 1.30*** (0.07) 0.54*** (0.03) 0.07*** (0.004) 0.21*** (0.06) 0.21* (0.09) 0.14** (0.05) 0.02** (0.006)

re2 r02 r12 r22 r32 r01 r02 r03 r12 r13 r23

0.56* (0.01) 0.30* (0.03)

Re2 R02 R12 R22 R32

0.003 (0.003) 0.0001 (0.0001) 0.01 (0.007) 0.003* (0.001)

NDW

50.13*** 25.69*** 9.44*** 1.16*** 5.46*** 5.17* 2.87** 0.38**

(0.88) (1.29) (0.63) (0.08) (1.21) (1.78) (0.87) (0.11)

211.74* (4.83) 200.64* (11.93) 0.006 (0.03)

0.85

(0.45)

0.0007 (0.0005) 48% 6%

38% 5%

0% 50%

0%

WPM

53.54*** (0.89) 11.31*** (1.42) 4.96*** (0.69) 0.76*** (0.09) 2.13† (1.23) 4.62* (1.97) 2.64** (0.97) 0.34** (0.13) 226.15* 167.85* 150.22* 38.51* 0.64* 19.40 16.64 2.23 73.22* 9.22* 4.94*

(7.34) (19.62) (57.16) (14.10) (0.24) (25.001) (11.83) (1.53) (27.71) (3.51) (1.83)

31% 1% 2% 2% 2%

16757.6*

55748.8*

56577.6*

16889.3

55854.2

56744.5

Note. L1 = Level 1 submodel; L2 = Level 2 submodel; 2LL = 2 log-likelihood deviance statistic; BIC = Schwarz’s Bayesian information criterion. † p < .10. *p < .05. **p < .01. ***p < .001.

638

Rojas and Iglesias

The final model (Model E) demonstrated the highest overall proportional variance reduction of within-person (pseudo-Re2) residual variance and of between-person intercept (pseudo-R02) residual variance and the lowest goodness-of-fit indices (2LL for nested models; BIC for non-nested models) for all outcome measures. Second, the fixed effects from the final model for each outcome measure indicated that boys and girls had a positive and significant instantaneous linear rate of change (c10; c11) at the fall of kindergarten, their expected rate of deceleration in the linear slope over time (c20; c21) was negative and significant, and their expected rate of acceleration in the quadratic slope over time (c30; c31) was positive and significant. Third, the effect of gender (being a girl) was positive and significant for initial status (c01) and for linear (c11), quadratic (c21), and cubic (c31) growth for all outcome measures in Spanish, except WPM, where gender’s effect on initial status was nonsignificant (c01 = 2.13, p = .08). Fourth, the variance components from the final model for each outcome measure indicated that there were significant intraindividual differences over time (re2), and that there were significant interindividual differences in initial status (r02) after controlling for gender. It should be noted that the final models for MLUw and NDW did not account for an initial status, linear slope covariance (r01), as their linear slopes were fixed. However, the final model for WPM demonstrated a nonsignificant initial status, linear slope covariance (r01 = 19.40, p > .05). The following model served as the final GCM for MLUw in Spanish (the final GCMs for NDW and WPM in Spanish are delineated in online supporting information Appendix S2): Level 1: Within-person variation MLUw-Spanishij ¼p0i þp1i ðTIME-1ij Þþp2i ðTIME-1ij Þ2 þp3i ðTIME-1ij Þ3 þeij Level 2: Between-person variation p0i ¼ c00 þ c01 ðGenderi Þ þ f0i p1i ¼ c10 þ c11 ðGenderi Þ p2i ¼ c20 þ c21 ðGenderi Þ þ f2i p3i ¼ c30 þ c31 ðGenderi Þ þ f3i The fixed effects of this model indicated that boys’ expected average initial status of MLUw-Spanish was c00 = 4.56, p < .001, with a positive and significant instantaneous linear rate of change (c10 = 1.30, p < .001) at the fall of kindergarten. The boys’

expected rate of deceleration in the linear slope over time was negative and significant (c20 = 0.54, p < .001), but their expected rate of acceleration in the quadratic slope over time was positive and significant (c30 = 0.07, p < .001). There was a positive and significant effect of gender (being a girl) on initial status (c01 = 0.21, p < .001) and on instantaneous linear rate of change (c11 = 0.21, p < .05) during the fall of kindergarten. The girls’ expected rate of additional deceleration (relative to boys) in the linear slope over time was negative and significant (c21 = 0.14, p < .01), but their expected rate of additional acceleration (relative to boys) in the quadratic slope over time was positive and significant (c31 = 0.02, p < .01). The variance components of this model indicated that there were significant intraindividual differences over time (re2 = 0.56, p < .05), and that there were significant inter-individual differences in initial status (r02 = 0.30, p < .05) after controlling for gender. The covariance components of this model indicated that there were positive and significant individual differences in the initial status, cubic slope covariance (r03 = 0.003, p < .05) after controlling for gender. Growth in English GCM testing of MLUw, NDW, and WPM in English also resulted in a number of common outcomes, summarized in Table 3. First, the CG model of discontinuous growth with the effect of gender (CG – Dc + G; see Model G in online supporting information Appendix S3 and Tables S3.1–S3.3) was the best fitting model for all outcome measures, indicating discontinuous growth over time of English. Although the final model (Model G) did not have the highest overall proportional variance reduction (pseudo-R2s), it had the lowest goodnessof-fit indices (2LL for nested models; BIC for nonnested models), which accounted for the fixed effects and variance components for all outcome measures. Second, the fixed effects from the final model for each outcome measure indicated that boys had a positive and significant linear growth rate (c10) during each semester, and an additional positive and significant linear growth rate (c20) during each spring semester. Third, the effect of gender (being a girl) was negative and significant for the initial status (c01) of WPM, but nonsignificant for the initial status of MLUw (c01 = 0.01, p = .86) and NDW (c01 = 1.02, p = .49). However, gender had an additional positive and significant effect on the linear growth rate (c21) of MLUw and NDW during each spring semester, relative to boys. Fourth, the variance components from the final model for each


outcome measure indicated that there were significant intraindividual differences over time (re2), that there were significant interindividual differences in initial status (r02) and in the linear slope (r12) after controlling for summer vacation and gender. In addition, the covariance components indicated that there were negative and significant individual differences in the initial status, linear slope covariance (r01) after controlling for summer vacation and gender, for each outcome measure. The following model served as the final GCM for MLUw in English (the final GCMs for NDW and WPM in English are delineated in online supporting information Appendix S4): Level 1: Within-person variation MLUw-Englishij ¼ p0i þ p1i ðTIMEij Þ þ p2i ðSUMMMERij Þ þ eij Level 2: Between-person variation p0i ¼ c00 þ c01 ðGenderi Þ þ f0i p1i ¼ c10 þ c11 ðGenderi Þ þ f1i

639

lish was c00 = 5.33, p < .001, with a positive and significant linear growth rate (c10 = 0.76, p < .001) during each semester, and an additional positive and significant linear growth rate (c20 = 0.88, p < .001) during each spring semester. There was an additional positive and significant effect of gender (being female) on the linear growth rate (c21 = 0.10, p < .05) during each spring semester, relative to boys. The variance components of this model indicated that there were significant intraindividual differences over time (re2 = 0.69, p < .05), and that there were significant interindividual differences in initial status (r02 = 0.99, p < .05) and in the linear slope (r12 = 0.15, p < .05) after controlling for summer vacation and gender. The covariance components of this model indicated that there were negative and significant individual differences in the initial status, linear slope covariance (r01 = 0.311, p < .05) after controlling for summer vacation and gender. Prototypical Growth Trajectories

p2i ¼ c20 þ c21 ðGenderi Þ The fixed effects of this model indicated that boys’ expected average initial status of MLUw-Eng-

Figure 1 summarizes the prototypical growth trajectories for MLUw, NDW, and WPM in Spanish and in English. The prototypical growth trajectories

Table 3 Final Growth Curve Model Parameter Estimates for Mean Length of Utterances in Words (MLUw), Number of Different Words (NDW), and Words per Minute (WPM) in English MLUw Fixed effects: c (SE) Intercept Linear slope Summer vacation slope Gender (G) G × Linear Slope G × Summer Vacation Slope Variance components: r (SE) L1: Within-person variance L2: B/w-person intercept L2: B/w-person linear slope Covariance (r02, r12) Proportional variance reduction L1: Within-person variance L2: B/w-person intercept L2: B/w-person linear slope Goodness-of-fit 2LL BIC

c00 c10 c20 c01 c11 c21

5.33*** 0.76*** 0.88*** 0.01 0.02 0.10*

re2 r02 r12 r01

0.69* 0.99* 0.15* 0.31*

Re2 R02 R12

50% 0% 0%

(0.05) (0.03) (0.03) (0.08) (0.04) (0.05) (0.02) (0.07) (0.02) (0.04)

NDW

46.17*** 19.11*** 9.52*** 1.02 0.12 3.25***

WPM

(1.05) (0.46) (0.60) (1.48) (0.64) (0.83)

58.53*** (1.10) 15.81*** (0.49) 4.36*** (0.56) 4.37** (1.54) 0.91 (0.68) 1.35† (0.78)

207.97* (5.44) 484.94* (28.58) 18.25* (4.68) 46.17* (9.88)

179.62* (4.78) 575.01* (31.11) 41.52* (5.60) 85.00* (11.33)

57% 0% 2%

51% 1% 0%

15237.8*

46608.7*

46359.4*

15323.6

46694.4

46445.2

Note. L1 = Level 1 submodel; L2 = Level 2 submodel; 2LL = 2 log-likelihood deviance statistic; BIC = Schwarz’s Bayesian information criterion. † p < .10. *p < .05. **p < .01. ***p < .001.

640

Rojas and Iglesias

Figure 1. Prototypical growth trajectories by gender for mean length of utterance in words (MLUw), number of different words (NDW), and words per minute (WPM) in Spanish and English.

for the outcome measures in Spanish were curvilinear (cubic), nonmonotonic, and continuous, whereas they were linear, nonmonotonic, and discontinuous for English. The prototypical growth trajectories indicated that in Spanish, girls began and ended with significantly higher outcome measures than boys. In English, the prototypical growth trajectories for MLUw and NDW did not demonstrate genderbased differences, but those for boys began and ended with significantly higher WPM than that of girls. The prototypical growth trajectories indicated common findings within each language. In Spanish,

the growth rates for all measures accelerated during kindergarten and second grade, with girls’ acceleration outpacing boys. In contrast, all Spanish outcome measures decelerated during first grade. In English, boys and girls demonstrated constant, linear growth in their outcome measures during kindergarten, first grade, and second grade, with girls’ linear growth outpacing boys in MLUw and NDW in English. During the summers, boys and girls demonstrated reduced growth in MLUw (negative growth) and NDW in English (decrease in growth rate, but overall positive growth). In contrast, WPM in English demonstrated increased growth during the summers,


with boys’ growth outpacing girls. In short, although the prototypical growth trajectories demonstrated similarities within each language, they also indicated unique shapes of growth across each language.

Discussion This study examined the language growth of Spanish-speaking children (classified as ELLs) who were acquiring English as a second language. The initial status and the growth rates of specific oral language skills (MLUw, NDW, and WPM in Spanish and English), and the predictive roles of summer vacation and gender, were estimated during the first 3 years of formal schooling. The findings of this study are discussed in the context of the three research questions. What Are the Overall Shapes of Growth in Spanish and English, and What Predictive Roles Are Played by Gender and Summer Vacation? The first research question focused on the growth of ELLs’ language skills over time. Based on the existing literature on the language development of ELLs (e.g., Hammer et al., 2008; Hammer et al., 2009; Kohnert et al., 2010; Simon-Cereijido & Gutiérrez-Clellen, 2009; Uchikoshi, 2006; Vagh et al., 2009), it was hypothesized that: (a) the outcome measures would demonstrate positive growth over time, (b) the shapes of growth would demonstrate within-language similarities, and (c) summer vacation and gender would predict language growth. The first two hypotheses were supported by the findings from this study, which indicated that all outcome measures demonstrated positive growth overall and that the shapes of growth were similar within each language. Language growth in Spanish was predicted by academic semester and gender, while language growth in English was predicted by academic semester, gender, and summer vacation. Shape of growth and summer vacation. For Spanish, the shape of the prototypical growth trajectories across outcome measures was curvilinear, nonmonotonic, and continuous. The fit of the GCMs with summer vacation was inferior to the fit of the GCMs without reflecting the continuous growth of the Spanish outcome measures. The prototypical growth trajectories in Spanish accelerated during kindergarten and second grade, but decelerated to an almost-flat trajectory during first grade, which was unexpected. However, it is possible that during first grade these children’s Spanish language sys-

641

tems underwent a temporary restructuring or reorganization (Larsen-Freeman, 2006; Perkins et al., 1996) as a function of simultaneous growth in other Spanish domains such as phonology, or as a result of the simultaneous development of the second language system. It is important to note that the Spanish deceleration during first grade is most likely not a sampling artifact; otherwise, a drastically distinct rate of growth would probably have been also observed in the prototypical language growth of English during first grade. For English, the shape of the prototypical growth trajectories across outcome measures in English was linear, nonmonotonic, and discontinuous. The discontinuous growth of the English outcome measures was due to the time-varying effect of the summer vacation schooling gap. The growth trajectories in English reflected outcome measures, with constant, linear growth during each academic year. During the summers, however, MLUw and NDW in English demonstrated reduced growth, whereas WPM in English demonstrated increased growth. Although the reduced growth of MLUw and NDW during the summers was expected (e.g., Hammer et al., 2008; McCoach et al., 2006), the increased growth of WPM was not. If the constant, linear growth of English morphosyntactic and lexical domains during the academic year is associated with systematic support and exposure to English in the classroom, it is possible that the reduced English growth is associated with lack of support during summer vacation. Interestingly, the increased growth of WPM during the summer may be related to the negative growth of MLUw. Some have observed the inverse pattern in ELLs (Bedore et al., 2006; Carias & Ingram, 2006), where an increase in MLUw is associated with a decrease in verbal fluency, with the production of more pauses, repetitions, and revisions. The decline in MLUw during the summer may have been associated with fewer such disfluencies and an increase in WPM. Shape of growth and gender. Boys and girls demonstrated near parallel shapes in their growth trajectories in Spanish (curvilinear and continuous) and in English (linear and discontinuous). In Spanish, girls began with significantly higher initial status and their growth rates across outcome measures significantly outpaced boys’ throughout the duration of the study. In English, there were no significant gender-based differences in the initial status of MLUw or NDW. Girls showed faster growth for MLUw and NDW in English, but only during the spring semesters. Boys’ initial status was significantly higher than girls’ for WPM in English, but

642

Rojas and Iglesias

their growth rate was not significantly different from girls’ throughout. The female growth advantage in Spanish found in this study was consistent with prior gender-based findings of monolingual early language development (e.g., Bornstein et al., 2004; Fenson et al., 1994). However, an important difference is that although the growth advantage for monolingual girls attenuates around the age of 3;0, the present study showed that for ELLs this advantage continued until at least the end of second grade (mean age = 8;2) in Spanish. It should be noted that the female growth advantage in Spanish throughout and the temporary female growth advantage in English during the spring were not consistent with prior findings of ELLs’ language growth (e.g., Hammer et al., 2009; Uchikoshi, 2006). Methodological differences across studies of ELLs’ language growth, however, make it difficult to compare findings across studies. For instance, the present study used expressive language outcomes instead of receptive language outcomes, and it used oral language samples instead of formal assessment instruments. Are There Significant Intra- and Interindividual Differences in Initial Status and Growth Rates of Spanish and English? The second research question focused on the variability of individuals’ performance. Although prior findings on the individual differences of ELLs’ language growth are limited, varying degrees of interand intraindividual variance across measures and across languages were hypothesized. These hypotheses were supported by the findings from this study, which indicated significant individual differences that were not accounted for by gender or summer vacation. For both languages and across outcome measures, significant intraindividual (within-person) differences were identified across each wave of measurement. In addition, the significant variability at the intraindividual level suggests that the rate of language growth of individual ELLs, irrespective of their gender or the effect of summer vacation, may differ significantly from one academic semester to the next. This significant intraindividual variability underscores the considerable degree of heterogeneity of ELLs’ language growth in Spanish and English. Although prototypical growth trajectories summarize language growth, this growth is highly variable over time within each child, in each of their languages. This finding also suggests the need for alternative approaches to language assessment that can capture

the intraindividual variability of ELLs, which traditional assessment methods (typically administered at one point in time) do not detect. With respect to the significant interindividual variability identified in this study, three patterns were found. First, regardless of gender, participants differed significantly in their initial status (intercepts) at the fall of kindergarten, for all outcome measures across both languages. Second, regardless of gender or summer vacation, ELLs demonstrated significantly different rates of growth for all outcome measures in English and of WPM in Spanish. Third, the MLUw and NDW growth rate differences in Spanish were largely accounted for by gender. Researchers and professional service practitioners can use these findings as empirical evidence of the broad, and highly variable range of language skills that ELL children demonstrate at formal school entry and throughout kindergarten to second grade. What Impact Does Initial Status Have on the Growth Trajectories in Spanish and English? The third research question focused on the extent to which initial status affects growth trajectories in Spanish and English, such that participants with lower performance at school entry “caught up” to higher performers. The findings from this study provided mixed support for the hypothesis that the morphosyntactic domain (MLUw), the lexical domain (NDW), and the integration of these linguistic domains (WPM) would demonstrate nonsignificant intercept–slope covariances in Spanish and English, based on the work of Rice et al. (2006, 2009). Specifically, in English the growth of MLUw, NDW, and WPM was systematically related to initial status, as demonstrated by negative and significant intercept–slope covariances. In Spanish, the growth of NDW and WPM was unrelated to initial status (nonsignificant intercept–slope covariances), but the growth of MLUw in Spanish was systematically related to initial status as demonstrated by a positive and significant intercept–slope covariance. The nonsignificant intercept–slope covariances of NDW and WPM in Spanish were consistent with prior findings (Rice et al., 2006; Rice et al., 2009), and suggest that the growth rates of these outcome measures may be expected to remain consistent over time. Therefore, ELLs with initially lower levels of NDW and WPM in Spanish may or may not catch up to their peers who perform at higher levels on these measures during the fall of kindergarten. The positive and significant intercept–slope covariance of MLUw in Spanish suggests that ELLs with


higher levels of MLUw in Spanish at initial status tend to have faster rates of growth, on average, than those with lower levels at initial status. Furthermore, lower performing ELLs at initial status may not catch up to higher performers at initial status, as the initial differences in Spanish MLUw will tend to become more pronounced over time. The negative and significant intercept–slope covariances of MLUw, NDW, and WPM in English suggest that ELLs with higher levels of these measures in English at initial status tend to have slower rates of growth, on average, than those with lower levels at initial status—perhaps indicating a “topping out” phenomenon for these particular outcome measures. These findings also suggest that lower performing ELLs at initial status may catch up to higher performers, as the initial differences in English will tend to become less pronounced over time. Taken together, the intercept– slope covariance findings for Spanish and English are encouraging. With regard to Spanish, the positive and significant intercept–slope covariance findings suggest that although some ELLs will invariably undergo first-language attrition, ELLs with stronger native language skills at the beginning of kindergarten are likely to experience continued growth in their native language. For English, the negative and significant intercept– slope covariance findings suggest that ELLs of most concern, those with weaker second language skills at the fall of kindergarten, are likely to experience faster second language growth. Limitations and Future Directions Despite the fact that the present study has advanced our understanding of language development in ELLs by specifying the shape and variability of growth for MLUw, NDW, and WPM in Spanish and English, it had a number of limitations. For instance, 16% of participants (328 ELLs) from this study’s final longitudinal data set were excluded. Although most participants (226 ELLs) were excluded due to missing gender information, an additional 94 were excluded because they were grade-retained. Although grade-retained participants represented only 4.6% of the longitudinal subproject, they nevertheless warrant future consideration. Language acquisition history prior to school entry, a potentially important covariate, was not considered since it was unavailable in the longitudinal subproject of the parent study. An important methodological limitation is that the bilingual growth of MLUw, NDW, and WPM was modeled

643

using the final longitudinal data set as one population. The possibility of distinct subgroups of ELLs (other than boys and girls) contributing unique language growth trajectories was not tested, which seems probable given the unexpected deceleration of Spanish during first grade and the significant intra- and interindividual differences identified in initial status and growth rates. These limitations serve to guide and motivate future research that can augment the precision and the applicability of the current findings. The findings of this study suggest that future work examining the Spanish and English language growth of ELLs needs to consider the possibility of ELL subgroups and their unique growth trajectories, the systematic exposure to the native and second language throughout the duration of the study, and the impact of additional covariates on growth. Results from this study may be extended with more advanced analytical approaches for modeling growth such as growth mixture modeling, which permits modeling the growth of multiple subgroups (Mueller & Hancock, 2010; Wang & Bodner, 2007). Considering the significant heterogeneity of ELL children’s oral language skills found in the present study, modeling ELL subgroup growth trajectories is important for future research. Additional variables such as the proportion of academic instructional time provided in each language, participants’ language experiences, and exposure with different interlocutors (e.g., parents, older siblings, peers), and maternal education will be investigated in future, planned studies of ELLs’ language growth. Apart from the incorporation of additional covariates to model language growth, modeling the growth of more fine-grained oral language measures may provide further insight into the growth trajectories found in this study. The findings from this study advance our understanding of bilingual language growth over time, and its heterogeneity in ELL children, as they navigate through their constantly changing educational and linguistic environments. In addition, this study provides the foundational information needed to guide and support the construction or validation of theoretical frameworks of bilingualism (Larsen-Freeman, Schmid, & Lowie, 2011).

References Baraldi, A. N., & Enders, C. K. (2010). An introduction to modern missing data analyses. Journal of School Psychology, 48, 5–37.

644

Rojas and Iglesias

Bedore, L. M. (1999). The acquisition of Spanish. In O. L. Taylor & L. B. Leonard (Eds.), Language acquisition across North America: Cross-cultural and cross-linguistic perspectives (pp. 157–208). San Diego, CA: Singular. Bedore, L. M., Fiestas, C., Peña, E. D., & Nagy, V. (2006). Cross-language comparisons of maze use in Spanish and English in functionally monolingual and bilingual children. Bilingualism: Language and Cognition, 9, 233–247. Bedore, L. M., & Peña, E. D. (2008). Assessment of bilingual children for identification of language impairment: Current findings and implications for practice. International Journal of Bilingual Education and Bilingualism, 11, 1–29. Bornstein, M. H., Hahn, C. S., & Haynes, O. M. (2004). Specific and general language performance across early childhood: Stability and gender considerations. First Language, 24, 267–304. Carias, S., & Ingram, D. (2006). Language and disfluency: Four case studies on Spanish-English bilingual children. Journal of Multilingual Communication Disorders, 4, 149–157. Coates, J. (1993). The acquisition of gender-differentiated language. In J. Coates (Ed.), Women, men, and language: A sociolinguistic account of gender differences in language (2nd ed., pp. 143–167). London, England: Longman. Conboy, B. T., & Thal, D. (2006). Ties between lexicon and grammar: Cross-sectional and longitudinal studies of bilingual toddlers. Child Development, 77, 712– 735. Cooper, H., Nye, B., Charlton, K., Lindsay, J., & Greathouse, S. (1996). The effects of summer vacation on achievement test scores: A narrative and meta-analytic review. Review of Educational Research, 66, 227– 268. Ding, C. S., & Davison, M. L. (2005). A longitudinal study of math achievement gains for initially low achieving students. Contemporary Educational Psychology, 30, 81–95. Elman, J. L., Bates, E. A., Johnson, M. H., KarmiloffSmith, A., Parisi, D., & Plunkett, K. (1999). Rethinking innateness: A connectionist perspective on development. Cambridge, MA: MIT Press. Farkas, G., & Beron, K. (2004). The detailed age trajectory of oral vocabulary knowledge: Differences by class and race. Social Science Research, 33, 464–497. Fenson, L., Dale, P. S., Reznick, J. S., Bates, E., Thal, D. J., & Pethick, S. J. (1994). Variability in early communicative development. Monographs of the Society for Research in Child Development, 59, 40–185. Fenson, L., Dale, P. S., Reznick, J. S., Thal, D., Bates, E., Hartung, J. P., et al. (1993). The MacArthur Communicative Development Inventories: User’s guide and technical manual. San Diego, CA: Singular. Francis, D. J., Carlson, C. D., Fletcher, J. M., Foorman, B. R., Goldenberg, C. R., Vaughn, S., et al. (2005). Oracy/ Literacy development of Spanish-speaking children: A multi-level program of research on language minority children and the instruction, school and community

contexts, and interventions that influence their academic outcomes. Perspectives: The International Dyslexia Association, 31, 8–12. Gleason, J. B., & Ratner, N. B. (2009). The development of language (7th ed.). New York, NY: Pearson. Golberg, H., Paradis, J., & Crago, M. (2008). Lexical acquisition over time in minority first language children learning English as a second language. Applied Psycholinguistics, 29, 41–65. Gutiérrez-Clellen, V. F., & Hofstetter, R. (1994). Syntactic complexity in Spanish narratives: A developmental study. Journal of Speech and Hearing Research, 37, 645–654. Gutiérrez-Clellen, V. F., Restrepo, M. A., Bedore, L., Peña, E., & Anderson, R. (2000). Language sample analysis in Spanish-speaking children: Methodological considerations. Language, Speech, and Hearing Services in Schools, 31, 88–98. Hadley, P. A., & Holt, J. K. (2006). Individual differences in the onset of tense marking: A growth-curve analysis. Journal of Speech, Language, and Hearing Research, 49, 984–1000. 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, 99–121. Hammer, C. S., Lawrence, F. R., & Miccio, A. W. (2008). The effect of summer vacation on bilingual preschoolers’ language development. Clinical Linguistics & Phonetics, 22, 686–702. Heilmann, J., Miller, J. F., Iglesias, A., Fabiano-Smith, L., Nockerts, A., & Andriacchi, K. D. (2008). Narrative transcription accuracy and reliability in two languages. Topics in Language Disorders, 28, 178–187. Heilmann, J., Miller, J. F., & Nockerts, A. (2010). Using language sample databases. Language, Speech, and Hearing Services in Schools, 41, 81–75. Heilmann, J., Nockerts, A., & Miller, J. F. (2010). Language sampling: Does the length of the transcript matter? Language, Speech, and Hearing Services in Schools, 41, 393–404. Huttenlocher, J., Haight, W., Bryk, A., Seltzer, M., & Lyons, T. (1991). Early vocabulary growth: Relation to language input and gender. Developmental Psychology, 27, 236–248. Kan, P. F. (2010). Measuring word learning ability in sequential bilingual children. Perspectives on Communication Disorders and Sciences in Culturally and Linguistically Diverse Populations, 17, 25–32. Kohnert, K., Kan, P. F., & Conboy, B. T. (2010). Lexical and grammatical associations in sequential bilingual preschoolers. Journal of Speech, Language, and Hearing Research, 53, 684–698. Kohnert, K., Windsor, J., & Ebert, K. D. (2009). Primary or “specific” language impairment and children learning a second language. Brain and Language, 109, 101–111. Larsen-Freeman, D. (2006). The emergence of complexity, fluency, and accuracy in the oral and written produc-

The Language Growth of Spanish-Speaking ELLS tion of five Chinese learners of English. Applied Linguistics, 27, 590–619. Larsen-Freeman, D., Schmid, M. S., & Lowie, W. (2011). From structure to chaos: Twenty years of modeling bilingualism. In M. S. Schmid & W. Lowie (Eds.), Modeling bilingualism: From structure to chaos (pp. 1–11). Amsterdam, Netherlands: Benjamins. Mayer, M. (1969). Frog, where are you? New York, NY: Dial. Mayer, M. (1974). Frog goes to dinner. New York, NY: Dial. Mayer, M. (1975a). Frog on his own. New York, NY: Dial. Mayer, M. (1975b). One frog too many. New York, NY: Dial. McCoach, D. B., O’Connell, A., Reis, S. M., & Levitt, H. (2006). Growing readers: A hierarchical linear model of children’s reading growth during the first 2 years of school. Journal of Educational Psychology, 98, 14–28. Miller, J. F. (1987). A grammatical characterization of language disorder. In J. Martin, P. Fletcher, R. Grunwell, & D. Hall (Eds.), Proceedings of the first international symposium on specific speech and language disorders in children (pp. 100–113). London, England: Association for All Speech Impaired Children. Miller, J. F., & Chapman, R. S. (1981). The relation between age and mean length of utterance in morphemes. Journal of Speech and Hearing Research, 24, 154–161. Miller, J. F., & Heilmann, J. (2004). Bilingual language project update. Paper presented at the Department of Communicative Disorders Colloquium, Madison, WI. Miller, J. F., Heilmann, J., Nockerts, A., Iglesias, A., Fabiano, L., & Francis, D. J. (2006). Oral language and reading in bilingual children. Learning Disabilities Research and Practice, 21, 30–43. Miller, J. F., & Iglesias, A. (2010). Systematic analysis of language transcripts (SALT), Research Version 2010 [Computer software]. Madison, WI: SALT Software. Mueller, R. O., & Hancock, G. R. (2010). Structural equation modeling. In G. R. Hancock & R. O. Mueller (Eds.), The reviewer’s guide to quantitative methods in the social sciences (pp. 371–383). New York, NY: Routledge. Nakamoto, J., Lindsey, K. A., & Manis, F. R. (2007). A longitudinal analysis of English language learners’ word decoding and reading comprehension. Reading and Writing, 20, 691–719. Özçalişkan, S., & Slobin, D. I. (1999). Learning how to search for the frog: Expression of manner of motion in English, Spanish, and Turkish. In A. Greenhill, H. Littlefield, & C. Tano (Eds.), Proceedings of the 23rd annual Boston University conference on language development (Vol. 2, pp. 541–552). Somerville, MA: Cascadilla. Pan, B. A., Rowe, M. L., Singer, J. D., & Snow, C. E. (2005). Maternal correlates of growth in toddler vocabulary production in low-income families. Child Development, 76, 763–782. Papafragou, A., Massey, C., & Gleitman, L. (2002). Shake, rattle, ‘n’ roll: The representation of motion in language and cognition. Cognition, 84, 189–219.

645

Pavlenko, A. (2009). Verbs of motion in L1 Russian of Russian-English bilinguals. Bilingualism: Language and Cognition, 13, 49–62. Perkins, K., Brutten, S. R., & Grass, S. M. (1996). An investigation of patterns of discontinuous learning: Implications for ESL measurement. Language Testing, 13, 63–82. Price, L. H., Hendricks, S., & Cook, C. (2010). Incorporating computer-aided language sample analysis into clinical practice. Language, Speech, and Hearing Services in Schools, 41, 206–222. Rice, M. L., Hoffman, L., & Wexler, K. (2009). Judgments of omitted BE and DO in questions as extended finiteness clinical markers of specific language impairment (SLI) to 15 years: A study of growth and asymptote. Journal of Speech, Language, and Hearing Research, 52, 1417–1433. Rice, M. L., Redmond, S. M., & Hoffman, L. (2006). Mean length of utterance in children with specific language impairment and in younger control children shows concurrent validity and stable and parallel growth trajectories. Journal of Speech, Language, and Hearing Research, 49, 793–808. Rice, M. L., Wexler, K., & Hershberger, S. (1998). Tense over time: The longitudinal course of tense acquisition in children with specific language impairment. Journal of Speech, Language, and Hearing Research, 41, 1412– 1431. Rojas, R., & Iglesias, A. (2009). Making a case for language sampling. The ASHA Leader, 14, 10–11, 13. Sebastián, E., & Slobin, D. I. (1994). Development of linguistic forms: Spanish. In R. A. Berman & D. I. Slobin (Eds.), Relating events in narrative: A crosslinguistic developmental study (pp. 239–284). Hillsdale, NJ: Erlbaum. Seltzer, M., Choi, K., & 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, 263–286. Siegler, R. S. (2002). Variability and infant development. Infant Behavior and Development, 25, 550–557. Simon-Cereijido, G., & Gutiérrez-Clellen, V. F. (2009). A cross-linguistic and bilingual evaluation of the interdependence between lexical and grammatical domains. Applied Psycholinguistics, 30, 315–337. Singer, J. D., & Willett, J. B. (2003). Applied longitudinal data analysis: Modeling change and event occurrence. New York, NY: Oxford University Press. SPSS Inc. (2010). IBM SPSS Statistics 18.0 for Mac [Computer software]. Somers, NY: IBM. Swanson, C. B. (2009). Perspectives on a population: Englishlanguage learners in American schools. Bethesda, MD: Editorial Projects in Education. Talmy, L. (1991). Paths to realization: A typology of event conflation. In L. Sutton, C. Johnson, & R. Shields (Eds.), Proceedings of the seventeenth annual meeting of the Berke-

646

Rojas and Iglesias

ley Linguistics Society: General session and parasession on the grammar of event structure (pp. 480–519). Berkeley, CA: Berkeley Linguistics Society. Tilstra, J., & McMaster, K. (2007). Productivity, fluency, and grammaticality measures from narratives: Potential indicators of language proficiency? Communication Disorders Quarterly, 29, 43–53. Uccelli, P., & Paéz, M. M. (2007). Narrative and vocabulary development of bilingual children from kindergarten to first grade: Developmental changes and associations among English and Spanish skills. Language, Speech, and Hearing Services in Schools, 38, 225–236. Uchikoshi, Y. (2006). English vocabulary development in bilingual kindergartners: What are the best predictors? Bilingualism: Language and Cognition, 9, 33–49. Vagh, S. B., Pan, B. A., & Mancilla-Martinez, J. (2009). Measuring growth in bilingual and monolingual children’s English productive vocabulary development: The utility of combining parent and teacher report. Child Development, 80, 1545–1563. van Geert, P., & van Dijk, M. (2002). Focus on variability: New tools to study intra-individual variability in development data. Infant Behavior and Development, 25, 340– 374. Wang, M., & Bodner, T. E. (2007). Growth mixture modeling: Identifying and predicting unobserved subpopu-

lations with longitudinal data. Organizational Research Methods, 10, 635–656. Wu, W., West, S. G., & Hughes, J. N. (2008). Effect of retention in first grade on children’s achievement trajectories over 4 years: A piecewise growth analysis using propensity score matching. Journal of Educational Psychology, 100, 727–740.

Supporting Information Additional supporting information may be found in the online version of this article: Appendix S1. Tables S1.1–S1.3. Appendix S2. Final GCMs for NDW and WPM in Spanish. Appendix S3. Tables S3.1–S3.3. Appendix S4. Final GCMs for NDW and WPM in English. Please note: Wiley-Blackwell is not responsible for the content or functionality of any supporting materials supplied by the authors. Any queries (other than missing material) should be directed to the corresponding author for the article.