Spanish language proficiency in dual language and English as a second language models: the impact of model, time, teacher, and student on Spanish language development

ABSTRACT

In this study, we examine the Spanish proficiency progress of elementary Spanish-English bilinguals in three types of language program models: a dual language bilingual education (DLBE) model with a 90/10 language allocation, a DLBE model with a 50/50 language allocation, and an English as a Second Language (ESL) model over an academic year. In the domain of speaking, individual student characteristics were most predictive of progress. For listening and reading, time had the greatest impact on their Spanish language growth. With writing, the most impactful moderator was the teacher, and this was most predictive in the DLBE 90/10 model. Implications point to the need to provide immersive writing instruction and to arrange DLBE schedules to ensure the inclusion of writing. Oftentimes in 50/50 DLBE models, the domain of writing is least covered within the partner language with English taking precedence due to testing accountability. As provisions for dual language bilingual education are increasing, this study holds important implications for this model's impact on Spanish-English bilinguals’ Spanish proficiency. Future studies will examine outcomes across time among emergent bilinguals in their native language, along with the potential influences of the student, the teacher, and the demographics of the school and community.

KEYWORDS:

• Additive bilingualism
• bilingual education
• biliteracy
• dual language immersion
• language testing
• language proficiency

In this study, we examine the Spanish proficiency progress across an academic year of Spanish-native, bilingual students in three different language programs: English as a Second language (ESL) where students receive instruction in-class or in a pull-out fashion; dual language bilingual education (DLBE) when 90% of the instruction is in Spanish and 10% in English with one teacher (DLBE/90/10/1T); and DLBE with a language allocation of 50% in each language with one teacher for each language (DLBE/50/50/2T). The DLBE/90/10/1T model has a long history of implementation whereas the DLBE/50/50/2T model is newer. For both DLBE models, there is an equal constitution of language majority students of English and emergent Spanish-English bilinguals whose first and family language is Spanish. We examine students’ Spanish language assessment scores to determine how student, time, teacher, and language model moderates their Spanish language proficiency in the four language domains of listening, speaking, reading, and writing.

• What are the effects of student, teacher and time across different language program models (ESL, DLBE 50/50/2T, or DLBE 90/10/1T), on the Spanish language proficiency of emergent bilingual students?

As provisions for dual language bilingual education are growing exponentially, this study holds important implications for this language model’s impact on bilingualism, and thereby, addressing a long history of educational inequities that subtract languages from emergent bilingual students.

Statement of the problem

Dual language bilingual education (DLBE) promises opportunities for students to acquire two or more languages while improving their academic achievement, reclassification as English language proficient (fluent), sustaining and contributing to family kinship and educational reform that addresses historic inequalities for emergent bilinguals.¹ As English is the privileged language of testing and accountability in the US, DLBE studies often focus on the superior long-term outcomes using the outcomes of English standardized tests and English language proficiency assessments as they are required by US educational policy; however, few schools formally assess the partner/native/minoritized language, even when it is the primary language of instruction (Lindholm-Leary and Block 2010). Arteagoitia and Yen (2020) claim that not assessing the native language of the emergent bilinguals within DLBE is an equity issue. By privileging English as the only language worthy of measure, such hierarchies map subtractively onto emergent bilinguals. Further, dual language was originally designed as a form of bilingual education for emergent bilinguals and recognizing their language recenters their language identities and rights (Valdés 1997, 2021).

Dual language teachers often use performance-based measures to qualitatively assess their students’ minoritized language proficiencies (e.g. Spanish). While such measures provide an important view of students’ growing language proficiency and offers formative feedback to educators, standardized measures offer a more reliable evaluation by language domain. Further, an assessment in minoritized languages offers credibility and parity with the annual English language proficiency assessments required by US federal educational policy.

This study addresses this measurement gap by examining different language instructional programs for elementary emergent bilinguals. By assessing native Spanish-English students’ Spanish proficiency by language domain, we can attain greater specificity on what native/partner language proficiency means for emergent bilinguals in different types of program models, particularly since US law requires language programming for emergent bilinguals (ESSA 2015; Castañeda v. Pickard 1981).

Literature review and conceptual framework

Emergent bilinguals being instructed within a DLBE benefit from its long-term benefits which include cognitive benefits, (Bialystok, Craik, and Luk 2012; Grundy and Timmer 2017), greater academic achievement (Morita-Mullaney, Renn, and Chiu 2020a, 2020b; Collier and Thomas 2004; Steele et al. 2017; Thomas and Collier 2002; Valentino and Reardon 2015), better grades (Morita-Mullaney, Renn, and Chiu 2020a, 2020b) and higher rates of English-fluent reclassification (Burke, Morita-Mullaney, and Singh 2016; Umansky and Reardon 2014). Importantly, and often omitted, is the family kinship that is sustained when minoritized students’ language(s) are honored and promoted within schools (Baquedano-López, Alexander, and Hernandez 2013), positively contributing to their civic identities and the reclamation of students’ heritage languages (Palmer et al. 2019).

Building on Thomas and Collier's (1997, 2002) original longitudinal work on the effectiveness in English academic achievement within dual language, more recent studies have shown similar impacts over time with emergent bilinguals achieving parity with English majority students and better than emergent bilinguals taught in English as a Second Language (ESL) models in the subject areas of English language arts and mathematics (Morita-Mullaney, Renn, and Chiu 2020b; Tran et al. 2015; Valentino and Reardon 2015). Yet, critiques with the methodologies used within such studies persist as they do not consistently account for selection bias (Steele et al. 2017), where dual-language programs are often moderated by the school district's ‘choice’ mechanisms (Bernstein et al. 2021 Garrett Delavan, Freire, and Morita-Mullaney 2022; Valdez, Delavan, and Freire 2014) with school populations being artificially arbitrated.

Federal, state, and local initiatives are quickly influencing the expansion of DLBE programs with the expectation that the program itself will manifest de facto bilingualism and biliteracy (Morita-Mullaney, Renn, and Chiu 2020b). While measuring language proficiencies is a more accurate mechanism for assessing language growth, it is rarely conducted in the partner language within DLBE programs. Because measurement of English language proficiency is required within current federal educational policy (ESSA 2015) of any identified English Learner (EL) (that we reference herein as emergent bilinguals), having comparative language information in the partner/native language can assist with relative comparisons over time. Identifying such patterns across languages can provide the bilingual field with summative feedback to pinpoint language strengths, developing areas, and how each language informs the other.

Next, such measurements taken within certain language communities can map the ways that particular varieties of Spanish manifest within given communities (e.g. a Spanish-speaking population that is US-born vs. not). By examining a standardized Spanish language exam and students’ performances, particular patterns across the language domains can be assessed among different speech communities, providing interpretive space for teachers, families and students. At the school level, Spanish language assessment can detail effective approaches and strategies for fostering different language domains for emergent bilinguals in DLBE programs, as well as the articulation of language(s) and languaging across subject areas and other social contexts that contribute to bilingual language development over time (Morita-Mullaney, Renn, and Chiu 2020b). As multiple factors influence the language and academic development and achievement of students, an investigation of a range of variables is needed to more holistically examine other mediating impacts on native language development of emergent bilinguals within DLBE. Next, we examine these variables, including the role of (1) the student; (2) the teacher; and (3) time within distinct language program models, namely, ESL; DLBE 50/50/2T; and DLBE 90/10/1T.

We acknowledge that within these program models, the stated language allocation may not account for actual language use. Importantly, using language allocation to describe a dual language program model can serve as an abstraction dissuading us from actual language practices that may be mediated across and through various languages, a critique that persists within dual language (Hamman-Ortiz 2019; Henderson and Palmer 2019). Adopting a more dynamic and fluid interpretation of language use allows implementational space for a translanguaging stance (García and Wei 2009).

Student demographics

Passing an academic achievement test is moderated by an emergent bilingual student's level of English proficiency; those with higher levels of English perform better than those in the earlier stages of English language development (Burke, Morita-Mullaney, and Singh 2016). This demographic identifier also has informed more recent studies that found that initial level of English proficiency and age moderates the trajectory to English fluency (Cook 2013; Haas et al. 2016), a finding that now informs US federal educational accountability calculations for the English language learning trajectories required of each identified emergent bilingual (ESSA 2015).

Time

Changes in progress, language proficiencies, and academic achievement are best examined cumulatively, as they become more evident over time versus at one discrete time point. This longer look also allows for better measurement across language program differences (Morita-Mullaney, Renn, and Chiu 2020a; Thomas and Collier 2012). Over time and in a well-implemented DLBE model, language proficiencies and academic achievement of emergent bilingual youth is higher relative to youth who are in non-DLBE models or English medium language models. Emergent bilingual youth are reclassified as English proficient at a slower rate in the elementary years, but have higher rates overall rates of reclassification to English fluency and academic achievement into their secondary years (Cook et al. 2008). This more gradual progress is consistently observed in other longitudinal studies, demonstrating the slower move toward target English language proficiencies (Umansky and Reardon 2014). Our study differs as it looks at such progress in the partner/native language of Spanish.

Teacher impact

As defined by U.S. policy, a teacher is highly qualified when they acquire a Bachelor's degree at an accredited institution, is state-licensed in their subject area, and has subject teaching expertise (NCLB 2001). As teachers with more preparation and experience tended to teach high-performing students rather than those at risk (e.g. emergent bilinguals; Kini and Podolsky 2016), a nationwide, highly qualified teacher standard sought to raise the minimum quality of teachers to serve students regarded at risk. Despite this high-quality teacher effort, research over the last decade shows that years of teaching experience is most predictive of student outcomes (Harris and Sass 2011) as opposed to pre-service preparation (Kukla-Acevedo 2009). Teachers with more experience yield better student outcomes, and such effects begin to materialize most substantively after 7 years of teaching (Blazar 2015). While more experience teaching in a given subject or grade level accounts for this effect, this is also supported by teachers’ engagement in more in-service professional development, a helpful proxy for years of experience (Harris and Sass 2011).

Language program type

Language program typology is multifaceted with as many as 12 different functional descriptions ranging from English immersion to dual language with 90–100% of the instruction in a minoritized language (Wright 2019). For this study, we look more simply at English as a Second Language (ESL), and ‘two-way programs are populated by English majority and speakers of the partner language in equal proportion’ (Morita-Mullaney, Renn, and Chiu 2020b).

The Indiana Department of Education (2020), defines ESL programs as when ‘their instruction is based on a special curriculum that typically involves little or no use of the native language, focuses on language (as opposed to content) and is usually taught during specific school periods’ (9). Indiana uses the naming convention of ‘dual language immersion’, stating

English Learners are enrolled in classes with non-English Learners and receive a minimum of 50% of their school day with instruction in the English Learner's native language and the remaining percentage with academic instruction in English (Indiana Department of Education 2020, 10). The dual language program model definition goes on to include a language allocation description of 50/50, 80/20 and 90/10 models.

Language allocation is often calculated at the percentage of time ‘taught’ in a given language with the naming conventions of 90/10; 80/20; and 50/50 articulating the proportion of time taught in English and the partner language. Scholars have long demonstrated that the greater proportion in the minoritized language, the greater the language proficiencies and academic achievement for emergent bilinguals and English majority students over time (Morita-Mullaney, Renn, and Chiu 2020b; Thomas and Collier 2002; Steele et al. 2017). Scholars have found that a 50/50 model often results in academic achievement that is lower than a 90/10 model, but even lower than an ESL model, suggesting that a higher proportion of instruction in the partner language results in higher academic achievement relative to emergent bilinguals in English-medium models and DL models with a higher proportion of English use (Collier and Thomas 2017; Genesee et al. 2006; Goldenberg 2008).

Language allocation within DLBE in practice can also imbue a strict language separation policy (e.g. English time and Spanish time). Yet, even with such language policies of separation, invariably, students will take up the languages that they need to negotiate meaning within the classroom. To account for these more fluid interactions, scholars call for more of a translanguaging stance that enables students to draw upon all their language resources regardless of content area, accepting and honoring all language negotiations and productions (Morita-Mullaney, Renn, and Chiu 2020b; Garcia and Wei 2009; Sánchez, García, and Solorza 2018). With this more fluid language approach DLBE classrooms, the language allocation becomes more diffuse and complicated to measure (Potowski 2007; 2019). Scholars critique that when such practices are adopted, depending on the student constitution in the classroom and their locally generated language policies and ecologies, English can take over, reducing possibilities of target language development (Potowski 2019; Wright et al. in press).

Due to the more recent implementation of DLBE across the US, how schools frame instruction within a DLBE model differs. Crucially, DLBE differs from a foreign language model. Foreign language models are generally behavioristic in nature with the proficient and expert teacher of the target language delivering discrete language for memorization and acquisition among its students, and not on using language within the context of content area learning. In contrast, DLBE scholars advocate for an orientation of native language arts instruction, which draws from the effective practices used within language arts classrooms, reflecting a more constructivist approach (Potowski and Carreira 2004; Webb and Miller 2000).

Most literature on the longitudinal studies of the impact of DLBE programming on emergent bilinguals posits that the language program model itself has a specific impact on students’ achievement and English language proficiency, not explicitly addressing the role of instructional language allocation over time and teacher quality (Morita-Mullaney, Renn, and Chiu 2020a, 2020b; Sánchez, García, and Solorza 2018; Soltero 2016). For example, a teacher may translanguage throughout the day yielding a different language allocation not ascribed to the model. Further, students may negotiate or translanguage throughout their lessons disrupting the precision of language allocation (e.g. Spanish in the morning; English in the afternoon, Sánchez, García, and Solorza 2018, 37). As multiple factors influence the precise fidelity to a language model, such as staying within its bounds of 90/10 or 50/50, examining other variables is needed, allowing us to examine other mediating impacts on native language development among emergent bilinguals.

Methods

This study examines the degree to which student, teacher, time, and instructional program attributes affect emergent bilinguals’ Spanish proficiency in listening, speaking, reading, and writing using a multivariate outcome, multilevel analysis, focusing specifically on emergent bilinguals identifying as native Spanish speakers.

Data collection

Educators teaching in the DLBE models (50/50/2T or 90/10/1T) were part of a treatment intervention focused on effective instruction for oral language, vocabulary in use, and small group instruction to facilitate development in all Spanish language domains (Morita-Mullaney, Renn, and Chiu 2020b). Each DLBE teacher (n = 9), had a comparison teacher (n = 9) who taught in a general education classroom at the same grade level. Comparison teachers served in ESL-only models with emergent bilinguals pulled out or receiving itinerant push-in support within their general education classrooms.

The school buildings of focus include a total of five early childhood and/or elementary schools across Mapleton and Silton (pseudonyms) (Table 1). Each building had either a treatment condition (90/10/1T or 50/50/2T) or a comparison condition (English as a Second Language).

Table 1. Demographic information of target schools in the 2018–2019 school year.

District	School	Language program model	Total # of students	Total # of Emergent Bilinguals	% of emergent bilinguals	% students of color	% of special education	% of SES
Mapleton	Mapleton TX Elem.	DLBE 50/50/2T	518	234	45.2	65.6	9.8	61.0
	Mapleton Comp. Elem.	ESL	354	146	41.2	65.0	14.4	71.8
Silton	Silton TX Elem.	DLBE 90/10/1T	769	235	30.6	71.1	4.7	45.1
	Silton Comp. Elem.	ESL	602	58	9.6	74.6	14.1	65.6
	Silton Comp. Kinder center	ESL	162	25	15.4	74.6	18.7	59.6

In our study, 94 emergent bilinguals, all of whom are native speakers of Spanish, participated in the Spanish assessment before and after an academic school year (pre- and post-), yielding 188 test results. These data provided comparisons of growth across a year for either kindergarten to 1st grade, 1st to 2nd grade, or 3rd to 4th grade (Table 2). Seventeen (17) students did not have two data points and were thus omitted.

Table 2. School sites of investigation, N = 94.

School name (pseudonym)	Program model	# of tests (over 2 years)	# of teachers	Site type
Mapleton TX Elem.	DLBE 50/50/2T	42	4	Treatment
Mapleton Comp. Elem.	ESL	40	4	Comparison
Silton TX Elem.	DLBE 90/10/1T	76	5	Treatment
Silton Comp. Elem.	ESL	20	4	Comparison
Silton Comp. Kinder center	ESL	10	1	Comparison
Total		188	18

Instrument

We used the LAS Links Español Form B instrument to measure the Spanish proficiency of emergent bilinguals with a native language of Spanish. The instrument covers social and school language, which includes intercultural and instructional communication like school-related tasks (e.g. following directions), and language related to academic content, which includes communications in the context of language arts, mathematics, science, and other subject areas (CTB/McGraw-Hill LLC 2013a). It is organized into four language content ‘strands:’ (1) Social, Intercultural, and Instructional Communication; (2) Language Arts, Social Studies, and History; (3) Mathematics, Science, and Technical Subjects; and (4) Foundational Reading and Writing Skills (Grades K-3 only).

LAS Links Español assesses students’ receptive language proficiency in Listening (Escuchando) and Reading (Lectura), and their productive language skills in Speaking (Hablando) and Writing (Escritura). Most of the items assessing Speaking and Writing are constructed responses, whereas most of the items measuring Listening and Reading are multiple-choice. The test also provides composite scores, including Overall, Oral Language, Comprehension, Literacy, and Productive Language (CTB/McGraw-Hill LLC 2013a). Raw scores, scale scores, and an overall proficiency level between 1.0 and 5.0 are provided for each domain and composite score. Raw scores are converted to a common scale to allow for comparison within and across grades to track growth over time. The scale score ranges vary across language domains: listening (grade K-1: 245–550; 2–3: 280–580; 4–5: 335–655), speaking (K-1:290–590; 2–3: 290–610; 4–5: 290–675), reading (K-1: 215–560; 2–3: 280–630; 4–5: 330–720), and writing (K-1: 300–630; 2–3: 320-685; 4–5: 320–695) (Table 3). Descriptive statistics from the field test norming sample are furnished (Appendix).

Table 3. LAS Links Español score ranges by language domain and grade level.

	Kindergarten and Grade 1		Grades 2 and 3		Grades 4 and 5
	Min SS	Max SS	Min SS	Max SS	Min SS	Max SS
Listening	245	550	280	580	335	655
Speaking	290	590	290	610	290	675
Reading	215	560	280	630	330	720
Writing	300	630	320	685	320	695

We recognize that this instrument does not capture the full repertoire of emergent bilinguals’ capacities, but it does recognize the well-studied language domain constructs of listening, speaking, reading, and writing in Spanish and parallel the domains assessed in English.

Table 4 illustrates a Kindergarten Spanish language proficiency level descriptions by language domain and level.

Table 4. Descriptions of Spanish language proficiency levels by language domain and level (CTB/McGraw-Hill LLC 2013a, p. 10)

Grade-span level proficiency level descriptors Table D.2 Proficiency level descriptors, Kindergarten Please note that the performance level descriptors represent a progression of skills and abilities. Skills and abilities specified in lower-performance levels are likely demonstrated by students in the higher-performance levels and may not be noted in the higher-level descriptors for a grade or grade range.
Kindergarten	Hablando	Escuchando	Lectura	Escritura
1 Beginning	Beginning students are starting to develop receptive and productive skills in Spanish.	Beginning students are starting to develop receptive and productive skills in Spanish.	Beginning students are starting to develop receptive and productive skills in Spanish.	Beginning students are starting to develop receptive and productive skills in Spanish.
2 Early intermediate	Early Intermediate students typically use basic vocabulary and simple phrases to name or describe common objects and express opinions or preferences in social and academic situations. They narrate a story related to a sequence of pictures about school-related activities using basic vocabulary. Restricted vocabulary and developing grammar limit expression. Errors frequently impede communication.	Early Intermediate students typically follow some simple oral directions using knowledge of everyday tasks, basic academic vocabulary, and familiar locations. Errors frequently impede communication and comprehension.	Early Intermediate students typically identify capital and lowercase letters in isolation and identify beginning, middle, and ending sounds. They match text to pictures and recognize high-frequency words. Errors frequently impede comprehension.	Early Intermediate students typically copy simple words and sentences that describe pictures or other prompts. Errors frequently impede communication.
3 Intermediate	Intermediate students typically use appropriate words and phrases when conducting transactions, making requests, and asking for clarification. They narrate a story related to a sequence of pictures about school-related activities using mostly accurate, although limited, vocabulary. They provide mostly clear information although errors interfere with communication.	Intermediate students typically follow simple oral directions. They identify the main ideas and details and make some inferences in simple oral stories. Errors interfere with communication and comprehension.	Intermediate students typically decode words, identify some main ideas, and recall details and events from a passage read independently. Errors interfere with comprehension.	Intermediate students typically write one or more words to describe a picture or respond to other prompts. Students are beginning to recognize correct sentence format. Errors interfere with communication.
4 Proficient	Proficient students typically produce simple and accurate sentences when making requests and asking for clarifications. They use appropriate words and phrases to label and describe the purpose of less common objects. They narrate a story related to a sequence of pictures about school-related activities using accurate vocabulary. Minor errors do not interfere with communication.	Proficient students typically follow more complex oral directions and draw inferences from oral stories. They identify the main ideas in more complex stories.	Proficient students typically read sentences or short passages independently. They identify rhyming words and identify main ideas and recall details in a passage. Errors do not interfere with comprehension.	Proficient students typically use correct basic grammar and capitalize the beginning of a sentence. They identify standard sentence structure and generate descriptive and explanatory sentences. Errors do not interfere with communication.
5 Above proficient	Above Proficient students typically produce simple sentences and use correct grammar when making requests, asking for clarification, and describing situations. They narrate a story with extensive and accurate vocabulary and grammar appropriate to their age.	Above Proficient students typically recall sequence of events from oral stories that have advanced vocabulary.	Above Proficient students typically use context clues to determine word meanings. They recall subtle details, identify sequences, and make inferences in a passage read independently.	Above Proficient students typically write a complete sentence to describe a picture or respond to other prompts. They use correct punctuation in exclamatory, declarative, and interrogative sentences. Communication is clear and complete, although content may contain minor errors.

Data analysis problems and solutions

Analyzing these data requires solving several problems regarding data, outcomes, and independent variables (see Table 5). These data have missing data and selection bias problems. Missing data (less than 1% in this study) can distort results, lower estimation efficiency, and complicate analyses (Peugh and Enders 2004). Thus, we address this problem by estimating missing values with Markov Chain Monte Carlo multiple imputation (MCMC-MI, Peugh and Enders 2004). In computer simulations, MCMC-MI estimates the missing data more effectively than simple imputation, pairwise deletion, listwise deletion, and mean substitution (Peugh and Enders 2004).

Table 5. Statistics strategies to address each analytic difficulty.

Analytic difficulty	Statistics strategy
Data Set
Selection bias in comparison vs. experimental groups	Test matching methods (Propensity Score Matching, Mahalanobis Distance Matching, Coarsened Exact Matching, Mahalanobis frontier, L1 frontier) to maximize both the subsample size and balance of covariates (King et al. 2011; Stuart 2010)
Outcome variables
Nested data (students within classes within schools …)	Multilevel analysis (aka Hierarchical linear modeling, Goldstein 2011)
Multiple outcomes (Y1, Y2, … )	Multivariate outcome multilevel analysis (Goldstein 2011) Multilevel Structural equation model (Joreskog and Sorbom 2018)
Explanatory variables
Many hypotheses’ false positives	Two-stage linear step-up procedure (Benjamini, Krieger, and Yekutieli 2006)
Compare effect sizes (β1 > β2?)	Lagrange multiplier tests (Bertsekas 2014)
Consistency of results across data sets (Robustness)	Separate multilevel, single outcome models Analyses of subsets of the data (Kennedy 2008)

When comparison and experimental groups of students differed significantly (selection bias), we created matching subsets in five ways: coarsened exact matching, propensity score matching, L₁ frontier, Mahalanobis distance matching, and Mahalanobis frontier (King, Lucas, and Nielsen 2015). As the effectiveness of each matching method (maximal sample size and minimal imbalance) differs across data sets, we selected the best result according to three criteria: (a) L_1, (b), Mahalanobis matching discrepancy, and (c) mean difference (King et al. 2011). These statistical matching results yielded 188 tests by 94 students (from the original 222 tests by 111 students). For α = 0.05 and an effect size of 0.3, the statistical power of our 188 student tests is 0.99 (Cohen et al. 2003) (Table 5).

Our data might have outcome problems of multiple outcomes or nested data. Students taught by the same teacher in the same school at the same time might resemble one another more than those taught by different teachers in different schools at different times (namely, nested data, Hox, Moerbeek, and Van de Schoot 2017). As an ordinary least squares regression might bias the standard errors, we rand a multilevel analysis (Hox, Moerbeek, and Van de Schoot 2017). As the correlated residuals of multiple outcomes might bias standard errors, we use a multivariate outcome multilevel analysis (Hox, Moerbeek, and Van de Schoot 2017).

Our analysis might face explanatory problems such as (a) false positives caused by multiple hypotheses, (b) comparisons of effect sizes, or (c) robustness issues. Testing multiple hypotheses can raise the likelihood of a Type I error (false positive, Benjamini, Krieger, and Yekutieli 2006). To address this issue, we used the two-stage linear step-up procedure (2SLSP; Benjamini, Krieger, and Yekutieli 2006). In computer simulations, 2SLSP performed better than 13 other competitors (Benjamini, Krieger, and Yekutieli 2006). When determining whether multiple independent variables have different effect sizes, Wald and likelihood ratio tests fail at boundary points (Bertsekas 2014). By contrast, Lagrange multiplier tests apply to all data and have more greater statistical power for minor deviations away from a null hypothesis (Bertsekas 2014). To determine whether our results are similar across despite small differences in data or analysis specifications, we conduct the following robustness check (Kennedy 2008). In an analysis with multiple outcomes, an incorrect equation for one outcome can cause errors for outcomes in other correct equations, so we separately analyze each outcome (Kennedy 2008). Also, we analyze separate data subsets (Kennedy 2008).

Explanatory model

We modeled student outcomes in each class across time with a multivariate outcome, multilevel difference-in-differences analysis (Hox, Moerbeek, and Van de Schoot 2017). (1) $\mathbf{T}\mathbf{e}\mathbf{s}{\mathbf{t}}_{ijkyt}={\text{β}}_y+e_{ijkyt}+f_{ijky}+g_{jky}+h_{ky}+{\text{β}}_{xy}Demographic{s}_{ijy}+{\text{β}}_{vjky}Treatmen{t}_{ijky}$(1) For vector Test_yijkt, student i in class j in school k has outcome y (speaking, listening, reading, writing) at time t. The grand mean intercept is β_y. There are unexplained components (also called residuals) at the student-, class-, school-, and time-levels (e_ijkyt, f_ijky, g_jky, h_ky). Next, we entered Demographics variables for students (female, age, grade, some_lunch, free_lunch, special education) and for school (all staff, years of experience, proportion of teachers in staff, proportion of licensed teachers in staff). A nested hypothesis test shows whether a group of independent variables is significant (χ² log-likelihood, Kennedy 2008). Removing a non-significant isolated variable does not cause omitted variable bias, so we removed them to raise precision and lower multicollinearity (i.e. when independent variables are correlated) (Kennedy 2008). Then, we entered the vectors Treatment (comparison (ESL) vs. DLBE 90/10/1T vs. DLBE 50/50/2T).

Findings

The district of Silton has a well-developed 90/10 DLBE schoolwide model with a one teacher/both languages design. Silton was an early adopter of DLBE, implementing the 90/10 model in 2008 (Morita-Mullaney, Renn, and Chiu 2020b). Due to its long history, they have adopted a model that allows for translanguaging by teachers and students, ascribing to a native language arts instructional model (Potowski and Carreria 2004). Mapleton was a newer adopter of the DLBE model and chose a 50/50 in 2017, one teacher/one language model where students switch teachers midway through the day. Like Silton, it begins at kindergarten, but at the time of our study, only had matriculated to grade 3. Mapleton hosted the DLBE model as a strand in one school and strictly separated their instructional languages. The Mapleton staff conceived their program as a foreign language model, focusing mainly on listening and speaking domains.

Teachers in the treatment context were paired with a comparison teacher who taught the same grade level, had a similar classroom constitution of emergent bilinguals, and whose years of experience, and advanced degrees to the extent possible, were commensurate (Table 6). For example, a treatment teacher at Silton Elementary Treatment was paired with a teacher at Silton Elementary Comparison, to identify any differences associated with their years of service and types of treatments and/or comparisons they received. The comparison teachers at Silton and Mapleton were in elementary schools where there was no DLBE program and students received pull-out ESL services and no direct Spanish instruction. Table 6 presents demographic information, as well as self-reported language proficiency by teacher.

Table 6. Treatment and comparison teacher characteristics.

District	School	Teacher grade level	Language program model	Language(s) of instruction	Teacher's years of experience	Teacher's reported overall proficiency in English (1–5)*	Teacher's reported overall proficiency in Spanish (1–5)*
Mapleton treatments	Mapleton TX	Kinder	DLBE 50/50/2T	English	3	4	1
	Mapleton TX	Kinder	DLBE 50/50/2T	Spanish	2	5	5
	Mapleton TX	1st	DLBE 50/50/2T	English	35	5	Did not report
	Mapleton TX	1st	DLBE 50/50/2T	Spanish	14	Did not report	5
Mapleton comparisons	Mapleton Comp	Kinder	ESL	English	22	5	Did not report
	Mapleton Comp	Kinder	ESL	English	5	5	5
	Mapleton Comp	1st	ESL	English	5	3.5	Did not report
	Mapleton Comp	1st	ESL	English	21	5	4.5
Silton treatments	Silton TX elem	Kinder	DLBE 90/10/1T	Spanish and English	3	5	4
	Silton TX elem	1st	DLBE 90/10/1T	Spanish and English	5	5	3
	Silton TX elem	1st	DLBE 90/10/1T	Spanish and English	8	5	5
	Silton TX elem	3rd	DLBE 90/10/1T	Spanish and English	6.5	5	5
	Silton TX elem	3rd	DLBE 90/10/1T	Spanish and English	14	5	5
Silton comparisons	Silton comp ELC	Kinder	ESL	English	11	5	Did not report
	Silton Comp elem	1st	ESL	English	21	5	Did not report
	Silton Comp elem	1st	ESL	English	17	5	Did not report
	Silton Comp elem	3rd	ESL	English	16	5	Did not report
	Silton Comp elem	3rd	ESL	English	3	5	Did not report

Note: Teachers self-reported their language proficiency in English and/or Spanish (*), reporting for the language(s) they ‘used on a daily basis.’ Proficiency levels were defined based on the US Department of State's language proficiency definitions – 1: beginner; 2: intermediate; 3: advanced-low; 4: advanced-high; 5: native or native-like.

https://careers.state.gov/faq-items/language-proficiency-definitions/

Limitations

Matching by building was based on relative comparison of demographic constitution of the schools’ student and teacher population, but a precise match was not possible given the capacity of buildings to accommodate our study. We also recognize that this study is over one year and therefore is not longitudinal, but it still provides the field with information that may explicate the various impacts on Spanish proficiency progress over one school year.

Descriptive statistics

These participating emergent bilingual students ranged from Kindergarten to 3rd grade at the time of the first administration (year 1), and their mean age was 7 years (see details in Table 4). Nearly half of these students were female (47%), and most students (92%) received free or reduced lunch benefits. A few students were special education (4%). These participating emergent bilingual students taking the exam ranged from Kindergarten to 3rd grade at the time of the first administration (year 1), and the average student age was approximately 7 years old. Mean scale scores ranged from 383.94 in Reading to 468.45 in Speaking (Table 7).

Table 7. Summary statistics (N = 188).

Variable	Mean	SD	Minimum	Median	Maximum
Student
Female	0.457
Free/reduced lunch	0.920
Special ed	0.043
Grade	1.330	1.064
Age (years)	7.06	1.09	5.28	6.90	10.12
Speaking score	468.45	39.507	290	475	590
Listening score	419.81	41.609	280	414	550
Reading score	383.94	72.772	215	377	566
Writing score	402.77	74.221	300	412	588
School
Total staff	34.683	6.186	17	39	40
Years of experience (%)
0–5	0.284	0.112	0.175	0.286	0.455
6–10	0.185	0.058	0.107	0.154	0.25
11–15	0.219	0.052	0.176	0.2	0.321
16–20	0.114	0.052	0.061	0.077	0.175
20+	0.198	0.034	0.152	0.2	0.294
% Teachers	0.799	0.059	0.714	0.8	1
% Licensed teachers	0.705	0.123	0.575	0.714	1

Note: Possible scale score ranges on Las Links Español varied depending on the language domain and the grade-level band. For example, possible score ranges for Reading were 215–560 for K-1, 280–630 for 2–3, and 330–720 for 4–5; for speaking, the ranges were 290–590 for K-1, 290–610 for 2–3, and 290–675 for 4–5.

At the building where treatment and comparison teachers taught, the teachers ranged in their years of experience from 0–5 to 20+ years as elementary teachers with a median distribution of 0.286 for 0–5 years; 0.154 for 6–10 years; 0.2 for 11–15 years; 0.077 for 16–20 years; 0.2 for 20+ years. Across the treatment and comparison schools, a median of 0.80 of educators in the building were trained teachers; and a median of 0.714 were licensed.

Raw score overview

An initial review of the scores across the testing intervals indicated that the speaking domain was strongest across the years in the ESL program, but not in either of the DLBE models (Table 8). Students in the DLBE 90/10/1T program outgained other students in listening, reading, and writing.

Table 8. Overall Spanish proficiency levels and differences across program type.

	Speaking			Listening			Reading			Writing
Program	2018	2019	Diff	2018	2019	Diff	2018	2019	Diff	2018	2019	Diff
ESL	3.24	3.28	0.03	2.80	2.93	0.12	1.51	1.48	−0.04	1.27	1.13	−0.14
DLBE 50/50/2T	3.64	3.52	−0.12	2.93	2.72	−0.21	1.39	1.60	0.21	1.14	1.12	−0.02
DLBE 90/10/1T	3.98	3.65	−0.33	2.94	3.07	0.13	1.70	2.35	0.65	1.72	1.87	0.15

Explanatory model

The outcomes (Spanish speaking, listening, reading, writing) all varied significantly across teachers, students, and time, but not across schools. Treatments and demographics accounted for differences in some outcomes, but not others, as detailed below (Table 6).

Speaking and listening

Speaking test scores differed most across students (47%), less across time (36%), and least across teachers (17%). The weak impact of the teacher suggests that speaking is the least malleable of the Spanish domains. Compared to other students, those who received free or reduced lunch or were in the DLBE/90/10/1T program had higher speaking scores (+32 and +20 points, respectively; see Table 6), accounting for over 17% of the differences in speaking scores.

Listening scores differed most across time (46%), less across teachers (30%), and least across students (24%). Students in higher grades had higher listening scores (+28 points), accounting for nearly 41% of the variance.

Reading and writing

Reading scores differed most across time (51%), less across teachers (28%), and least across students (24%). Older students had slightly higher reading scores (+0.1 point per year). Students who received free/reduced lunch or were in DLBE 50/50/2T programs had lower reading scores (−44 and −27 points respectively), accounting for over 37% of the variance.

Writing scores differed most across teachers (45%), less across time (36%), and least across students (19%), suggesting a strong teacher impact. In writing, girls outperformed boys (+22 points), and older students slightly outperformed younger students (+0.07 points). Students in the DLBE 90/10/1T programs had higher writing scores than other students (+23 points), especially compared to those in the DLBE 50/50/2T programs (+50 points = +23 – (−27)). Together, these variables accounted for over 44% of the writing score variance. Notable within the area of writing is the strong impact of the teacher on writing scores within the DLBE 90/10/1T model (Table 9).

Table 9. Summary of multivariate outcome, multilevel results modeling speaking, listening, reading and writing test scores.

Explanatory variable	Speaking	Listening	Reading	Writing
Female				22.352*
				(9.513)
Age			0.105***	0.075***
			(0.011)	(0.009)
Grade		27.810*
		(10.953)
Free or Reduced Lunch	32.080***		−44.192*
	(6.910)		(17.923)
DLBE 90/10/1T	19.910**			23.402**
	(6.211)			(9.055)
DLBE 50/50/2T			−26.916***	−27.018**
			(7.235)	(8.623)

Level	Variance	Expl	Variance	Expl	Variance	Expl	Variance	Expl
Teacher	17%	0.964	30%	1.000	28%	0.658	45%	0.831
Student	47%	0.018	24%	0.000	24%	0.000	19%	0.000
Time	36%	0.000	46%	0.231	51%	0.358	36%	0.189
Total variance explained		0.172		0.406		0.367		0.442

*p < .05, **p < .01, ***p < .001.

Note: The bold is used to bring light to the findings of importance.

Discussion

The working theory within DLBE programming is that any type of DLBE program will yield significant results in the native language of emergent bilinguals. Yet, these findings show that emergent bilinguals in DLBE 50/50/2T fared lower in Spanish writing than the comparison students in an ESL program model. Because the time devoted to Spanish is so limited in the DLBE 50/50/2T model, the domain of writing is often omitted from the Spanish language arts block, depriving students of opportunities for Spanish writing. Secondly, the 50/50 model follows the Spanish as a Foreign Language approach (Potowski and Carreira 2004), in which the Spanish part of the day is devoted almost entirely to listening and speaking or oracy. As our team continues to work with DLBE 50/50/2T educators within this study, we raise important conversations about the structure of their program and how more time and explicit focus on Spanish writing is needed, as well as how a reframing toward a paradigm of Spanish language arts would assist with the complementary development of all language domains, greatly benefiting emergent bilinguals within DLBE.

Language and content allocation fidelity

In Mapleton's treatment elementary with a 50/50/1T model, use of Spanish was allocated to math, science and Spanish language arts with social studies and English language arts conducted in English. Mapleton adheres to a strict language separation policy with each teacher staying and expecting the given language from students during their allocated part of the day. This decision and policy are informed by their adherence to a ‘Utah model’ that suggests that the 50/50 model is superior to the 90/10 (Freire and Delavan 2021). The English teachers within the treatment site of the DLBE program, described their desperation to integrate writing and found that they spent as much time as they could to include it. In contrast, the Spanish portion of the day did little to no writing across the grade levels because the focus was on speaking and listening. Thus, the Spanish half of the day reflected a foreign language model consistent with observed practices in high school programs (Potowski and Carreira 2004). The audiolingual method (repeat after me) was commonplace with few opportunities for interaction among students and little to no writing instruction.

In contrast, Silton's elementary's treatment/DLBE actively used a translanguaging and bridging stance, allowing students to work across their multiple languages during instruction. Further, teachers would translanguage to negotiate meaning with students. Students in Silton also spent the entire day with one teacher, so there was no physical transition to a new classroom and because the model is a 90/10 language allocation, English time was brief.

Not only did these two programs adhere to different language allocation models, they were girded by different philosophies of language learning. Mapleton is a newer and developing DLBE strand within a school, whereas Silton is a well-established program and is a school wide program. Due to these differing histories, their professional development was guided by this language allocation/philosophical stance. In its genesis, Mapleton participated in professional development with rigid adherence to the 50/50 model and language separation, which left little time for the inclusion of all language domains in English and Spanish. Silton participated in its own internal professional development, driven by district DLBE coaches and DLBE teacher experts from within. Given that Silton has a 20+ year history of doing a two-way model, they examined how the languages intersected and how translanguaging served as a bridge for meaning making among its students, particularly their Spanish-English bilinguals. Further, they focused more of their time on all language domains, teaching Spanish more as the content area of Spanish language arts (Potowski and Carreira 2004). Its longevity as well as its school wide focus, Silton had a school wide ecology of Spanish language development, whereas Mapleton treated Spanish time as a foreign language model with oral recitation and repetition. Further, Mapleton teachers who taught the 50% of the day in English felt rushed, feeling that they could not attend to writing given the need to cover all of the curriculum, whereas the Spanish teachers focused on Spanish listening and speaking.

Reading = literacy, not writing

Both treatment and comparison schools were hyper focused on reading during designated Spanish and English times. Reading time focused on discrete letter identification and phonics, all accompanied by recitation, song and/or movement in the earlier grades (K-1st). While this type of reading instruction was less evident in the upper grades (Grades 3–4), there was still a heavy emphasis on reading content in the partner language and less inclusion of writing instruction, demonstrating that the local construction across all sites, treatment and comparison, conceived and practiced literacy as reading, and not writing.

Our study shows that writing was best moderated and influenced by the teacher, demonstrating that when it is an explicit focus, the teacher can have an impact on its development. Writing is a more complex language domain, the last to develop an academic gatekeeper, and seldom employed in the native language within our observed DLBE classrooms (Morita-Mullaney, Renn, and Chiu 2020b).

Implications

As much of the globe is expanding dual language offerings for emergent bilinguals and English-language majority students, our study shows how thoughtful language planning for new, developing, and established programs is needed. Newer programs, like Mapleton swiftly identified dual language models, adopted them and employed them with little attention to measurement of partner language. The assumption was that any proportion of instruction in Spanish would yield biliteracy, and that assessment was only necessary in the majority language of English, the language of testing. For Silton, with a much longer history, their rationale for not measuring Spanish was driven by their superior academic achievement outcomes over time (Morita-Mullaney, Renn, and Chiu 2020a, 2020b), which served as a proxy for its bilingual success, as reinforced by their highest achieving status in the district. While both districts celebrate the maintenance and development of Spanish among Spanish-emergent bilinguals, measurement is always done in relationship to English, but not in parity to Spanish; a much needed equity action within DLBE K-12 education.

Program models in this new immigrant state are mostly adopting 50/50 models as informed by limitations with staffing (Morita-Mullaney, Renn, and Chiu 2020a). Finding one English teacher and one Spanish teacher is feasible and gradually prepares a school building to shift its staffing over time. Further, Indiana leaders are most inclined to select a 50/50 DLBE model, as they are concerned about testing and accountability outcomes in English that begin in grade 3, despite what current research suggests about more time spent in the minoritized language yielding better academic achievement outcomes over time. Invariably, leaders aspire for swift and tangible results that materialize quickly.

Lastly, measuring and attending to Spanish proficiency or the partner language demonstrates value toward the minoritized language and thereby, biliteracy. If it is not centered as a value toward equity for emergent bilinguals, then it is assumed as a form of social capital that privileges language majority students, foreclosing on the genesis of bilingual education being centered on equity for emergent bilinguals.

Next steps

In Part 2 of our study, we will examine different types of bilingual students and how such student variables may moderate language proficiency trajectories in Spanish. Not all Spanish-bilinguals are the same. Some are simultaneous bilinguals, sequential bilinguals and/or heritage speakers of Spanish. These differences may inform other variables that may moderate the trajectories for Spanish language development among Spanish-English bilinguals, increasing our understanding of how Spanish proficiency develops within the context of a dual language education with varying language allocations.

A newly proposed study intends to follow cohorts of emergent bilinguals and English majority students in DLBE for four years to identify the varying impacts of different DLBE program models, meeting the criterion of a longitudinal study. At each interval of progress, we will share students’ results with their teachers to identify the possible variables and language conditions that inform language growth over time across different student groups, providing interpretive value to the outcomes.

Assessing students’ Spanish proficiency reveals how the constructs of listening, speaking, reading, and writing are performed by different student groups and bring greater clarity to how the tool defines such language constructs. The Seal of Biliteracy, which awards students an insignia upon their graduation for biliteracy (Californians Together 2016) has been critiqued as the native language assessments used for the Seal's recognition employs a paradigm of foreign language (Carreria and Potowski 2004), and thus, English majority students perform at a superior level compared to emergent bilinguals (Heineke, Davin, and Bedford 2018). Studies like ours can analyze and shape the ways in which language assessments are designed, considering a fuller repertoire of what language proficiency means in Spanish, and reforming current languages policies, specifically to benefit and recenter emergent bilinguals of Spanish.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Additional information

Funding

This work was supported by Office of English Language Acquisition [grant number T365Z170213].

Notes on contributors

Trish Morita-Mullaney

Trish Morita-Mullaney is an Associate Professor at Purdue University. Her research focuses on the influences of educational policy at the federal, state, district, school and classroom levels and how this informs the practices of educators who work with and among emergent bilinguals. Framed by critical language, critical race, and feminist theories, she employs a variety of methods to accomplish these aims. She uses participatory and constructivist methods with her study participants, as they unpack, critique and analyze their orientations towards emergent bilinguals and multilingualism. Her most recent work focuses on how equity can be trapped for emergent bilinguals within dual language programming.

Jennifer Renn

Jennifer Renn an Associate Research Scientist at Purdue University. Her research lies at the intersection of sociolinguistics and education. Dr. Renn uses mixed methods approaches to study a range of topics related to language and education, most recently examining the impact of educator training on instruction, language attitudes, and ideologies.

Ming Ming Chiu

Ming Ming Chiu is Chair Professor of Analytics and Diversity (Honor) at The Education University of Ming Hong Kong. He invented statistical discourse analysis, multilevel diffusion analysis, artificial intelligence Statistician, and online detection of sexual predators. He studies inequalities, culture, and learning in 65 countries, and automatic statistical analyses.

Notes

1 We use the term, Emergent bilingual (EB) to describe students who are gaining English as an additional linguistic resource (García 2009).

Appendix. Las Links Español Form B Scale Score Descriptive Statistics from the field test norming sample (CTB/McGraw-Hill LLC 2013b).

Table

Grades	Mean	SD	Median
Speaking
K	485	46.91	492
1	503	49.61	509
2–3	512	53.75	508
4–5	532	69.02	525
Listening
K-1	406	70.27	409
2–3	469	61.34	470
4–5	498	66.74	501
Reading
K-1	376	69.73	374
2–3	469	58.90	472
4–5	488	67.25	488
Writing
K	378	58.43	368
1	454	60.47	461
2–3	488	62.98	495
4–5	503	72.04	504