Exploring the Effects of Culturally Responsive Instruction on Reading Comprehension, Language Comprehension, and Decoding with Bayesian Multilevel Models

Abstract

Reading comprehension (RC) can be predicted from language comprehension (LC) and decoding, and all three constructs are responsive to structured teaching. Culturally responsive instruction, which explicitly connects students’ lived experiences with school experiences, can also effectively support literacy learning. However, little is known about how structured and culturally responsive approaches work in tandem, and whether positive effects may occur through the path of LC or decoding, or directly on RC. Further, does culturally responsive teaching support transfer from local, personalized learning materials to standardized measures? This study investigates the impact of structured and culturally responsive teaching on standardized measures of RC, LC, and decoding among 263 students in grades 1 through 3. Participants were assigned to one of three groups: (1) generic structured teaching approach that used mainstream materials, (2) a structured culturally responsive approach that centered students’ interests, cultures, and sense of belonging, and (3) a waitlisted business-as-usual control group. Over 10 weeks, students received small-group teaching focused on decoding and LC. Bayesian multilevel ANCOVA models indicate all groups grew, with differential positive effects for LC for the culturally responsive treatment group. The findings suggest benefits to integrating cultural relevance into structured literacy teaching and that a multifaceted approach may be effective. Implications and limitations are discussed.

1. Introduction

Reading comprehension (RC) can be explained by two broad constructs: decoding and language comprehension (LC) (Hoover & Gough, 1990; Hoover & Tunmer, 2020). Decoding is the ability to convert the “code” of printed language to speech, while LC is the ability to understand the meaning of language in spoken form. The Simple View of Reading predicts that RC is the product of decoding and LC; that is, one cannot compensate for the other (Gough & Tunmer, 1986; Hoover & Gough, 1990). Decoding and LC together have explained 94–100% of the variance in children’s RC (Y.-S. G. Kim, 2017; Lonigan et al., 2018).

In contrast to the “simple” name, each construct in the Simple View of Reading is complex and comprises many skills. As LC contributes to RC, it draws on one’s entire (multi)linguistic repertoire: vocabulary, grammar, pragmatics, discourse-level skills, and knowledge. Skilled and fluent decoding is preceded by development of phonemic awareness and knowledge of letter–sound correspondence. Young readers learn to decode increasingly large orthographic patterns within words, and with practice, effortful decoding evolves into automatic word recognition. However, the process of successful word recognition is not a one-way path of unlocking the code from print to speech. It also involves matching the decoded word to one’s oral vocabulary (Ehri, 2017), which solidifies the word’s form, sound, and meaning in memory. A robust oral lexicon can support decoding development, as decoding and LC overlap in their prediction of RC (Sinclair et al., 2025; Foorman et al., 2020; Hoover & Tunmer, 2022).

The Simple View of Reading has been explored and tested for over 30 years. Researchers have proposed additions including reading fluency (Silverman et al., 2013; Tilstra et al., 2009), vocabulary (Protopapas et al., 2013), and a decoding + LC (addition) term (Chen & Vellutino, 1997), among others. Duke and Cartwright (2021) offer the Active View of Reading, which articulates bridging processes between LC and decoding including print concepts, reading fluency, vocabulary, morphological awareness, and graphophonological–semantic flexibility (the ability to quickly switch between and simultaneously consider written words’ form and meaning). The Active View of Reading also incorporates active self-regulation and, relevant to the present study, cultural knowledge (within the language comprehension factor) and sociocultural context. Educators may find this model more adaptable to classroom practice than the Simple View of Reading, as it articulates the constructs that constitute decoding and LC along with additional factors and bridging processes. However, the Active View of Reading is a relatively new model and remains to be fully validated (Hoover & Tunmer, 2022). This paper considers both models by exploring the effects of culturally responsive instruction on decoding, LC, and RC.

The Simple View of Reading forms the theoretical foundation for popular evidence-based materials for reading instruction, yet commercial programs tend to focus on decoding and often lack fulsome operationalization of LC (Cervetti et al., 2020; Goldenberg, 2020). LC is a malleable construct that is responsive to instruction, which can support linguistic, cognitive, and conceptual development. Teaching children conceptual knowledge and comprehension skills through oral language may be more efficient than teaching these through written language (Sweller et al., 2011). In one of the only published studies on a structured intervention that incorporated both oral LC and decoding, Clarke et al. (2017) found parallel instruction of decoding and LC to significantly improve the RC of 11- and 12-year-olds. However, their study utilized one-on-one instruction, resulting in somewhat lower ecological validity, due to most schools’ limited resources. The present study uses small-group instruction, which is a common approach to intervention in contemporary schools.

In this study, I explore the effects of small-group componential instruction through a culturally responsive lens. Culturally responsive instruction explicitly honors students’ personal and family histories, cultures, interests, and values, and tailors classroom learning to connect with students’ lives (Gay, 2018). Does a culturally responsive approach, interwoven with structured decoding and LC instruction, produce added effects beyond those of a generic structured approach?

1.1. Decoding as a Predictor of Reading Comprehension

The relationship between decoding and RC is well documented. García and Cain’s (2014) meta-analysis found a corrected correlation of 0.74 between decoding and RC, which decreases by 0.5 standardized units every year across childhood. Efficient decoding facilitates semantic activation to enable RC (Hoover & Gough, 1990), largely through phonological processing (Stanovich, 1982). Phonological awareness and grapheme–phoneme knowledge enable beginning readers to decode simple words. Later, multi-letter patterns, syllables, and morphemes are consolidated. Over time, the phonological, orthographic, and semantic properties of words are internalized, enabling automatic recognition and self-teaching of new words (Ehri, 2005; Share, 1995). Rapid naming, letter knowledge, phonemic awareness, and kindergarten word reading predict Grade 1 decoding, with inconsistent orthographies (such as English) showing more variance than consistent ones (Caravolas et al., 2019). Structured teaching of decoding (i.e., phonics) has a positive effect on typically developing learners (Torgerson et al., 2019) as well as children at risk for word-level reading disability (Lovett et al., 2017).

Successful word recognition requires decoding, or “lifting the words off the page”, but also matching the decoded phonological form to a known word in one’s oral vocabulary (Ehri, 2017). This process strengthens connections between the word’s sounds, printed form, and meaning in long-term memory through orthographic mapping (Ehri, 2005). As children make connections between the semantic, written, and spoken forms of words, they develop automaticity in word reading.

Children learn the meaning of new words more easily from listening than reading (Geva et al., 2017). In young children, LC is an indirect predictor of RC through the path of decoding (Sinclair et al., 2025). Thus, instruction that includes both language and decoding emphases can potentially create a linguistic and cognitive environment that nurtures robust spoken and semantic lexical representations, enabling newly acquired written representations of words to easily fit into existing schema.

1.2. Language Comprehension as a Predictor of Reading Comprehension

LC and RC skills have parallelism in terms of task demands (Hagtvet, 2003). Depending on the type of task (story retell or cloze), Hagtvet (2003) found oral (LC) and written (RC) versions drew on similar subskills—story retell tasks drew on vocabulary, and cloze tasks drew on syntax—across LC and RC. Wolf et al. (2019) found similar results, where LC and RC were highly predictive of each other, and oral vocabulary additionally predicted both LC and RC. Hagtvet (2003) notes that her team was surprised to find that “… listening and reading comprehension of equivalent types of tasks would share so many similarities in terms of underlying oral language skills …[This] underscores the importance of task demands, to some extent over and above modality [print or spoken]…” (pp. 525–526).

Hagvet’s comment aligns with Gough and Tunmer’s (1986) early paper describing the Simple View of Reading, where they postulate that once “printed matter is decoded, the reader applies to the text exactly the same mechanisms which he or she would bring to bear on its spoken equivalent… It would be falsified [by] someone who could decode and listen, yet could not read” (p. 9). Of course, differences exist between LC and RC (Oakhill et al., 2015). These include typically denser language in RC than in LC; the ability to clarify and ask questions of the speaker, which is not possible in RC; and rereading, which LC does not permit and thus makes greater demands of working memory (Dufva et al., 2001; Language and Reading Research Consortium et al., 2018).

Recent research has highlighted LC as applied to instruction (Pearson et al., 2020). LC has an increasingly strong association with RC over time, as children mature and academic texts and cognitive demands become more challenging (Catts et al., 2005; Tilstra et al., 2009). Oral language interventions, that is, targeted supports that focus on vocabulary, syntax, pragmatics, and/or discourse-level comprehension through spoken language, have shown improvement of RC in both early and middle childhood (Bianco et al., 2010; Clarke et al., 2010). The Reading for Understanding initiative (Cervetti et al., 2020) has called for further research to unpack LC to “determine which skills and knowledge are malleable to instruction, and to develop and test instructional interventions for young readers” (p. S169).

The constructs of language, listening, and linguistic comprehension are similar but not identical. Santoro (2012) holds that listening comprehension refers to the ability to understand connected text that is spoken orally, while language/linguistic comprehension is an umbrella construct that can include listening comprehension, vocabulary, oral expression, and broader verbal proficiency. In their writings on the Simple View of Reading, Gough and Tunmer (1986) and Hoover and Gough (1990) refer to the understanding of oral language as linguistic comprehension and listening comprehension, while Hoover and Tunmer (2020) define language comprehension as including background knowledge, inferencing skill, and linguistic knowledge (phonological, syntactic, semantic). They note that listening comprehension is often associated with a specific measurement approach that uses story retell tasks. The present study uses language comprehension because my focus is on the broader construct that employs knowledge, vocabulary, and syntax rather than parsing these into narrower components.

1.2.1. Language Comprehension and Knowledge Instruction

Relevant background knowledge has long been understood as an important factor in LC and RC, and it can partially compensate for a reader’s weaker verbal and decoding skills (Hattan & Lupo, 2020; Recht & Leslie, 1988; Schneider et al., 1989). A recent book by leading educational scholars (Surma et al., 2025) highlights the importance of knowledge across academic learning, and that RC, critical thinking, and problem-solving are all more efficient and easier when domain knowledge is strong. From a cognitive perspective, well organized background knowledge (schema) can be called into working memory as a cohesive unit. Working memory is limited in how many information units it can process simultaneously, with estimates ranging from 3 to 5 units (Cowan, 2010). However, if information is well organized, each unit can be quite large. Sweller et al. (2011) note that working memory is “limited in capacity and duration if dealing with novel information but unlimited in capacity and duration if dealing with familiar information” (p. vii). Thus, a reader’s well-organized knowledge can free up working memory resources for higher-level comprehension such as inference generation and comprehension monitoring.

Knowledge of word meanings is clearly important for LC and RC (Perfetti & Stafura, 2014), and while discrete word knowledge is important, measures of vocabulary reveal “the exposed tip of the conceptual iceberg” (Anderson & Freebody, 1981; as cited in J. S. Kim et al., 2021a); that is, robust word knowledge is an indicator of robust conceptual knowledge. Coyne et al. (2010) found direct instruction of vocabulary improved kindergarten students’ targeted word knowledge as well as general receptive vocabulary and LC. Similarly, Melby-Lervåg et al. (2019) examined the effects of a language intervention for pre-school children that included story-based narrative and listening instruction as well as explicitly trained vocabulary words. Improvements in both intermediate transfer (with some trained words) and distal transfer (no trained words) measures of language were mediated by students’ ability to define the explicitly taught vocabulary. Cervetti et al. (2016) similarly found that reading a set of conceptually coherent knowledge-building texts (i.e., on a similar topic) led to improved retell on related topics and marginally greater incidental learning of general academic words than did reading texts across a variety of topics.

Recent work by Kim and colleagues highlights the effects of scaffolded and structured conceptual knowledge instruction on near and far transfer. J. S. Kim et al. (2021a) employed structured, spiraling, content-focused literacy instruction, using science-focused nonfiction texts that were more challenging than typical materials. Instruction followed a thematic cycle that intentionally wove the creation of learning goals, interactive read-alouds and discussions, concept mapping, argumentative writing, collaborative research, and other activities. Grade 1 students’ vocabulary (largely explicitly taught words) showed positive gains, as well as in science domain-specific LC and argumentative writing, and general RC. J. S. Kim et al. (2021b) performed a replication study in grades 1–2, adding social studies content literacy instruction along with science. Each unit consisted of one theme and organizing question, e.g., “How do animals survive in their habitat [Grade 1 science]?… How do inventors solve problems [Grade 2 social studies]?” (p. 1942). They found similar gains as the previous study in vocabulary and argumentative writing in both domains, but not in RC. However, possibly due to the simultaneous teaching of two domains, students in the treatment condition demonstrated significant growth on non-taught words. J. S. Kim et al. (2023) found that from Grade 1 to 2, domain specific content literacy instruction improved near and intermediate vocabulary transfer. J. S. Kim et al. (2024) showed longitudinal effects through Grade 3 and Grade 4 follow-up on far transfer to RC and even mathematics. Collectively, Kim et al. sought to improve vocabulary, LC, and RC through both robust oral and written activities. These studies indicate that the greater the exposure to carefully structured teaching of conceptual knowledge and vocabulary, the better the near and far literacy and knowledge outcomes.

1.2.2. Language Comprehension and Strategy Instruction

In recent decades, trends in reading instruction have tended to emphasize teaching generic skills and strategies (e.g., monitoring, predicting, questioning, clarifying, summarizing, re-reading) over domain knowledge (Afflerbach et al., 2020), in part because the influential National Reading Panel Report (NICHD, 2000) advocates for strategy instruction but does not strongly connect this instruction to student knowledge. McNamara and O’Reilly (2009) argue that reading strategy use is, fundamentally, the reader strategizing how to effectively bring their knowledge to the text. Willingham (2017) argues that while strategy instruction can be effective, it typically produces a one-time gain and thus does not require extensive repetition:

“Comprehension is a product of connecting ideas. Strategy instruction emphasizes that ideas should be connected, but it cannot tell a reader how to make those connections… So the injunction “make inferences!” may convince a reader who has not previously done so to begin making them, but that will be a one-time improvement”.

(p. 125)

Willingham maintains that long-term instruction should focus on improving domain knowledge because it more strongly predicts successful RC. Indeed, a recent metanalysis found that strategy instruction is only effective in improving RC when combined with background knowledge instruction (Peng et al., 2024).

Nonetheless, certain forms of strategy instruction have demonstrated positive effects on comprehension. One is reciprocal teaching (Palinscar & Brown, 1984), which focuses on comprehension strategies of summarizing, questioning, clarifying, and predicting, and scaffolding children’s participation and responsibility through a gradual release of responsibility. The cycle begins with the teacher leading, explaining, and modeling, with the goal that over time students lead the discussion and application of strategies. Esposito et al. (2025) used reciprocal teaching in a 20-week LC intervention for 8–9-year-old children with weak language abilities and found an effect size of d = 0.40 (RC was not measured). Johnson-Glenberg (2000) found 28 sessions of reciprocal teaching over 10 weeks significantly improved the comprehension skills of students in grade 3–5 who were poor comprehenders (but adequate decoders) with an effect size of 1.32 for listening recall. Connecting LC to RC, Mesa et al. (2020) designed an intervention with direct instruction in LC and decoding, and reciprocal teaching for RC strategies for a wide age range. They found gains across all three constructs, and that decoding gains were maintained after nine months. Clarke et al. (2010) found oral-focused reciprocal teaching had a larger positive effect on RC than text-focused reciprocal teaching (or balanced text/oral teaching) for 8–9-year-old children with large discrepancies between reading fluency and RC. This finding suggests oral language-medium strategy interventions may be effective in supporting a range of challenges relating to language and reading.

1.2.3. Language Comprehension and Transfer

The previous sections discussed the importance of knowledge and strategy instruction for comprehension. Knowledge is required to make inferences, and further, well-organized knowledge can facilitate the efficiency of working memory, which has limited capacity. Strategy instruction can be beneficial, as questioning, clarifying, summarizing, and predicting, as described by Palinscar and Brown (1984), are “the kinds of active and aggressive interactions with texts that poor readers do not engage in readily” (p. 121).

The transfer of knowledge, skills, and strategies from a taught domain to another domain (formal or informal) has long been a pressing concern for educators and researchers. Hoover and Gough (1990) highlight how knowledge represented in LC can transfer to RC: “as an individual’s knowledge base increases [and] is reflected in linguistic comprehension, so should it be reflected in reading comprehension, and vice-versa… the greater the knowledge base expressible through linguistic comprehension, the greater the reading comprehension (assuming non-zero decoding skills)” (p. 153). Clarke et al.’s (2010) study showed that LC-focused strategy instruction was more effective in promoting RC than text-focused instruction, which suggests strategies can transfer from LC to RC without much explicit instruction in such transfer.

Yet, the transfer of knowledge and strategies from one domain to another is not guaranteed. Melby-Lervåg et al. (2019) cite Sternberg and Frensch’s (1993) discussion of the mechanisms of transfer. Encoding specificity pertains to how narrow students perceive the application of learning to be; narrower beliefs suggest transfer is less likely. Sternberg and Frensch suggest teaching in a way that requires multiple applications through teacher-provided and student-found opportunities. Organization is how clearly and logically content is presented. Students benefit when teachers provide a framework that organizes the material in relation to the curriculum and students’ lives outside of school. Discrimination refers to helping students understand which strategies and knowledge can be applied to novel applications; i.e., what is general and what is specific. The mental set refers to students’ understanding about when, where, and how transfer can occur appropriately, with mindsets that are neither too rigid nor too flexible.

The studies cited above indicate that structured teaching that focuses on knowledge and/or strategies can effectively transfer to new domains. Still, less is known about drivers of change: is it the structured routines of learning activities, the emphasis on strategies that activate comprehension, or the focus on coherent topics and conceptual knowledge? This study aims to contribute to the evidence base in these areas by exploring a structured approach to LC and decoding instruction that is also culturally responsive.

1.3. Culturally Responsive Instruction

Leading scholars have called for the development of evidence-informed reading instruction that is also culturally responsive and builds students’ knowledge (Duke & Cartwright, 2021; Jensen, 2021). Culturally informed instruction maintains high academic expectations for students, supports their development of cultural competence, integrates their lived experiences and ways of knowing into learning activities, and sustains multiple languages and cultures (Kelly et al., 2021; Ladson-Billings, 1995). Cultural and background knowledge can form a bridge to comprehension by activating relevant schema, which allows information to be contextualized and understood. Students from diverse cultures can experience the success of reading on familiar topics and in turn generalize these successes to less familiar tasks (Cantor et al., 2019).

Pritchard (1990) provides an example of how cultural familiarity influences comprehension. Pritchard recruited proficient Grade 11 readers from the US and the Pacific island of Palau to read two passages describing a funeral: one described typical funeral events in Palau, and the other typical funeral events in the US. Students read both passages in their home language (Palauan or English). Students’ responses to the culturally familiar passage were more thorough, had a greater number of elaborations and fewer distortions, contained more connections, and the response rate was faster.

Reynolds et al. (1982) recruited Grade 8 students, half of whom identified as Black (living in Tennessee) and half identified as White (living in Illinois). All participants read a fictional letter written from a friend about an altercation in a school cafeteria. The letter could be interpreted in two ways: as a physical fight, or as playing The Dozens (i.e., roasting, trash-talking, or “sounding”), a verbal game in Black American culture. After reading, participants wrote down every detail they could remember. Finally, they rated a set of statements on their consistency with the text; statements framed the letter as either documenting a physical fight or a verbal roast. Participants’ responses to both tasks were statistically associated with their cultural background. Black participants from Tennessee interpreted the letter as a verbal roast (game), while White participants from Illinois interpreted it as a physical fight, although there was no evidence of physicality in the letter.

Nearly a century ago, Bartlett (1932) analyzed participants’ recall of passages across genres (folk tale, descriptions, arguments) and content areas. His most famous experiment asked British university students to read a story that originated in Native American culture about a battle, ghosts, and death. In their retellings, events outside participants’ cultures (e.g., culturally specific descriptions of ghosts and death) were flattened, changed, or omitted. Familiar events and concepts were emphasized, such as details from the battle. Further, culturally familiar morals were ascribed to the story, even if absent in the original. Participants modified settings or events in the original passage that were difficult to understand, in order to align them with their existing schema. Barlett attributes these changes to recall being a constructive and social act.

Culturally informed instruction acknowledges that students’ learning is contextualized by the funds of knowledge they bring to school and whether that knowledge is valued by educators (Moll et al., 1992). Schools differentiate students by their cultural, linguistic, and symbolic capital (Bourdieu, 1977), and linguistically and culturally diverse students benefit when they see themselves in classroom materials and instruction (Sims Bishop, 2009). This is not always prioritized, as K-12 teachers in the US, Canada, Europe, and UK are less diverse than their students (Donlevy et al., 2015; Ryan et al., 2009). As holders of capital, teachers can name that which is mainstream and that which is not. Bourdieu (1989) notes that we “…tend to perceive the world as natural and to accept it much more readily than one might imagine—especially when you look at the situation of the dominated through the social eyes of a dominant” (p. 18).

As an example contrasting this tendency, Reynolds et al. (1982), whose study cited above describes White students misunderstanding a culturally unfamiliar text, notes about their study,

“[F]or once, a reading passage was biased in favor of black [sic] inner-city students since it was based on their implicit knowledge and system of relevancies. The reaction that many White middle-class teachers and students have to inner-city black students trying to work their way through culturally loaded material was mirrored by one of our black male subjects. Upon being told that White children understood the letter to be about a fight instead of sounding [verbal roast], he looked surprised and said, “What’s the matter? Can’t they read?””

(p. 365)

Culturally informed instruction has evolved over the past 30 years. Culturally relevant pedagogy (Ladson-Billings, 1992, 1995) emerged after the U.S. Civil Rights movement to promote inclusion and achievement for students of colour by emphasizing high expectations, offering meaningful scaffolding, drawing on students’ cultural identities, and fostering critical awareness. Culturally responsive pedagogy (Gay, 2002, 2018) builds on this by explicitly teaching “to and through [students’] personal and cultural strengths” (2018, p. 32), centering identity, culture, and learning. This approach explicitly works against systemic disadvantage and discrimination toward students who are racialized or marginalized for culture or language (Villegas & Lucas, 2002). A core principle is that engagement and achievement increase when students’ cultural perspectives and languages are validated and valued (Gay, 2018).

Culturally responsive practice has effectively been applied to structured literacy interventions such as Reading Recovery (Hilaski, 2020), pre-school literacy activities (Dajani & Meier, 2019), and locally developed skill-focused intervention programs, which Gillispie (2021) describes as successful collaborative programs with American Indian and Alaska Native scholars, teachers, and children. Recent work by Hammond (2015, 2025) has led the field of education in the integration of cognitive science and culturally responsive, student-centered practice. Her work highlights that understanding the connection between culture and cognition is essential for building psychologically safe, nurturing classrooms and for empowering students to be independent learners.

The present study seeks to add to the research base about the integration of structured and culturally responsive literacy instruction by addressing the question: What changes in RC, LC and decoding skills are associated with a structured and scaffolded skills-based approach that focuses on decoding and LC, in comparison to a structured and scaffolded skills-based approach that is also culturally responsive?

2. Materials and Methods

2.1. Participants

The study was approved by (redacted) University Interdisciplinary Committee on Ethics in Human Research (Clearance #20230384-ED). Students in grades 1–3 were recruited from 15 classrooms in 6 schools in an eastern province of Canada, and urban and rural schools were represented. Informed consent was obtained from parents/guardians, ensuring they understood the voluntary nature of their child’s participation and data confidentiality. Data are cross-sectional (n = 263) across three grade levels: Grade 1 (n = 79; mean age = 6.48 years), Grade 2 (n = 127; mean age = 7.52), and Grade 3 (n = 57; mean age = 8.54); 49% were girls.

Approximately 20% of participants’ parents reported multilingual homes; languages included Arabic, French, Vietnamese, Tagalog, Malayalam, Nepali, Punjabi, Serbian, Spanish, Tamil, Ukrainian, Urdu, and Yoruba. Though English proficiency was not formally assessed, all participants had adequate proficiency to understand instructions and partake in the study, with the exception of one child who had recently immigrated from Ukraine and was excluded from analysis. The province in which these data were collected does not track students’ socioeconomic status or ethnicity, so those data are not available. Learning disabilities are typically not assessed in this context until students are at least 8 years old; thus, a reliable indicator for special learning needs was not available. At the request of district-level administrators, all children were invited to participate in this small-group instructional study, regardless of reading and language ability.

2.2. Measures

The Woodcock Reading Mastery Test Third Edition (WRMT; Woodcock, 2011) is a standardized, norm-referenced Level B measure of reading and related subskills used in education and research. Four measures were employed: decoding as measured by word attack (reading nonsense words out loud); LC (listening to a oral passage and verbally answering open-ended questions); passage comprehension (completing cloze-type comprehension items that require the entire passage to be read and understood), and word comprehension (reading individual words and providing synonyms or antonyms, or completing analogies). Passage comprehension and word comprehension are averaged to represent RC (Woodcock, 2011). Sample items were administered for all measures, and all measures are of graded difficulty with a ceiling rule of four consecutive incorrect responses. Only summed data, and not item-level data, are available for the present analysis, and therefore reliability values are reported from the WRMT manual. The ranges of reported split-half reliabilities for the two forms across grades 1–3 are word attack: 0.92–0.96; passage comprehension: 0.84–0.93, word comprehension: 0.91–0.96, LC: 0.73–0.86 (Woodcock, 2011).

Models were estimated using growth scale values (GSV). GSV scores are derived from Rasch ability scores that are jointly calibrated across the WRMT‘s two parallel forms. GSV scores are conceptualized as absolute rather than relative, and because they are on a vertical scale (a linear function of the Rasch scores), they can be readily utilized to understand growth, even across grades and ages. GSV are the WRMT derived score recommended for use in calculations (Pearson Education, 2022).

2.3. Design and Procedures

The study design is a quasi-experimental matched comparison of recipients (Reichardt, 2011) with one control group and two treatment groups. The WRMT was administered individually in a quiet and distraction-free environment in late January 2023 (pre-test) and April/May 2023 (post-test). Participating students were taken out of their classroom to complete the test one-on-one with a research assistant. Nine trained research team members followed standardized instructions to ensure consistency. Administration time varied from 30–60 min depending on the child’s ability and attention, and for some participants the sessions were split over two days. WRMT forms A and B were counterbalanced with no student assessed with the same form twice. WRMT pre-test results were used to place students in roughly homogenous small groups based on RC score (4 to 6 students each small group) in one of three conditions: Treatment 1 (structured decoding and LC instruction), Treatment 2 (culturally responsive structured decoding and LC instruction), or a waitlisted control group who received instruction once the trial was complete.

All participants engaged in small group learning and attended an average of 14.4 35-min sessions over the 10-week period. Treatment 1 was structured instruction of decoding and LC using generic materials, and Treatment 2 was Treatment 1 with the use of culturally relevant materials and opportunities for critical conversation. Treatment 1 and Treatment 2 both employed quality code-focused and meaning-focused activities. The teaching sessions were structured similarly to the program described by Clarke et al. (2017), with code-focused instruction in the first part (12 min) and LC instruction during the second part (23 min).

2.3.1. Structured Decoding and LC Instruction (Treatment 1)

The code-focused instruction for Treatment 1 utilized University of Florida Literacy Initiative (UFLI; Lane & Contesse, 2022) materials (word lists, slide shows, connected text, word chains); students’ instruction began at the point in the scope and sequence most appropriate to their current decoding skill level. Each research assistant (RA) received seven grapheme manipulative sets to align with UFLI instruction in order to provide interactive, multimodal activities. The second part of each learning session entailed reciprocal teaching of oral language comprehension skills through interactive and student-led activities of summarizing, questioning, and clarifying (Palinscar & Brown, 1984). Oral language activities were designed to develop LC and related subskills (vocabulary, syntactic awareness, discourse-level understanding). For the oral language comprehension instruction, RAs sourced a variety of materials including authentic children’s literature sourced from area libraries and media about real world events. Books were read aloud to the small groups and served as the stimuli for discussion and strategies. Comprehension instruction was entirely verbal and based on books that the RA read aloud.

2.3.2. Culturally Responsive and Structured Decoding and LC Instruction (Treatment 2)

For Treatment 2, materials for learning activities (books, words and sentences) were initially selected/developed based on parents’ responses to an interest, language, and culture questionnaire and discussions with the classroom teacher. Over time, as students expressed their interests while participating in small groups, LC instructional materials were explicitly selected from area libraries that aligned with their interests. As in Treatment 1, oral comprehension instruction was based on spoken language and read-alouds; students did not engage in RC but only LC. RAs who taught the groups used culturally relevant techniques for Treatment 2 such as creating a culturally safe learning environment, learning more about students’ backgrounds, and engaging students in critical dialogue within and beyond the text. As for decoding instruction, RAs were encouraged to use the UFLI scope and sequence and modify word lists and connected text sentences and passages to connect to students’ lives while maintaining the targeted orthographic focus.

Gay (2018)’s Culturally responsive teaching: Theory, research, and practice served as a core resource for RAs. To differentiate Treatment 2, with its focus on culturally responsive instruction, RAs focused on (1) maintaining high academic expectations for students, (2) supporting their development of cultural competence and critical cultural consciousness, (3) integrating students’ lived experiences and ways of knowing, doing, and expression into learning activities, and (4) engaging in practices that sustain multiple languages and cultures.

2.3.3. Business-As-Usual Control Group

The third group served as a waitlisted business-as-usual control group. This group did not receive the research team-delivered small-group intervention during the main trial period, but as a waitlisted group, they did participate after the trial period. Some students in this group may have engaged in school-provided small-group instruction as part of regular practice; however, these activities were not observed or controlled by the research team. Most students received standard whole-group instruction. Coincidentally, several schools were also beginning to implement UFLI during this time.

2.4. Fidelity

Nine RAs who were doctoral students or retired teachers completed criminal record and vulnerable sector checks and confidentiality agreements. RAs participated in 12 h of initial training sessions online through Webex, followed by weekly debriefing meetings during the initial month of treatment. Training focused first on WRMT administration and then on instruction, especially differentiating the two treatments.

There is some overlap between Treatment 1 and Treatment 2, namely that both were structured and both maintained high academic expectations for all students. A key focus for Treatment 2 was supporting students to make meaningful connections between the learned content and their own lived experiences, as Kourea et al. (2018) note that a culturally responsive approach to reading instruction, “should maximize student responsiveness by presenting examples connected to students’ lives.” RAs were asked to emphasize key differences between the two treatments, especially in how often and deeply they engage students about their lives, backgrounds, and interests, and how often they relate these to classroom-based learning activities (extensively in Treatment 2, and almost never in Treatment 1).

When talking with students about their personal histories, cultures, and interests, in Treatment 1 RAs were are pleasantly neutral and not strongly engaged. For example, they would respond to a student sharing a personal story with, “How nice. Thank you for telling me about that.” For Treatment 2, on the other hand, RAs were deeply engaged, asked follow-up questions, and related back with their own experiences with the goal of understanding what makes students “tick”: their key interests and concerns, i.e., issues facing their family, community, and school. In terms of materials for Treatment 1, RAs chose age- appropriate mainstream materials, but they had no specific relevance. For Treatment 2, RAs chose materials that aim to “click” with students in relation to their cultures, backgrounds, and interests. They also focused on social justice topics that are age-appropriate for children.

When teaching in Treatment 1, RAs made broad connections to learning, perhaps offering a loose and abstract sense of connection to the community. For Treatment 2, RAs explicitly connected taught ideas to children’s real lived experiences: past, present, and future. To do this, RAs needed to know about the students and to have strong relationships with them (even given the short time frame of the study). In Treatment 2, RAs read books relevant to students’ lives and made direct connections between students, the community, and taught content. RAs explicitly aimed to build students’ sense of belonging and social justice orientation, connecting to the books they read aloud. They also allowed space for students to share personal stories. RAs engaged in critical conversations that helped students affirm and appreciate their own and others’ cultures of origin, and supported students’ ability to identify, analyze, and solve real-world problems, especially those related to societal inequalities, in a child-appropriate way.

For decoding activities in Treatment 2, RAs chose example words that were locally meaningful and relevant, and used words, sentences, songs, poems, and rhymes from students’ cultures when possible. Treatment 2 honored students’ dialects and accents and their different pronunciations of phonemes, as well as regional vocabulary. For Treatment 2 decoding activities, RAs used students’ names and familiar topics and words that also aligned with the orthographic complexity of the current activity, e.g., for consonant digraphs: “This is how I chop the stick,” or “STUDENT got stuck in the muck. STUDENT wants to pick up a stick and a brick.” Treatment 1 decoding used the UFLI-provided words, such as, “The clock goes tick tock.”

For LC activities in Treatment 2, books to read aloud were selected based on students’ interests and backgrounds. For example, one Grade 3 Treatment 2 group included students with Filipino and African heritage. The RA selected books including Hand Over Hand (Fullerton, 2017) and A Crocodile’s Tale (Aruego & Aruego, 1972), both grounded in Filipino culture, as well as Hair Love (Cherry, 2019) and The Water Princess (Verde & Badiel, 2016) which celebrate African and Black cultures. Group conversations directly connected these stories to students’ lives. The RA also reported connected LC activities to local current events, reading Dolly’s Rescue (Langdon & Short, 2022) about dolphins who were trapped in a harbor in Eastern Canada. The same RA reported for Treatment 1 choosing mainstream materials such as books by Dr. Seuss, Chester’s Way, The Mother’s Day Mice, and Spaghetti and Meatballs.

After each session, RAs were asked to document how they differentiated treatments 1 and 2. Their responses for Treatment 1 included: “Kept discussion to the book, did not engage in personal conversations”, “stuck to the material”, “discussion was related to the text with no connection to what they wonder about”, “focus on strategies, book choice is generic”, “conversation was between students, I remained neutral and did not comment on their discussion”, “The Lorax addresses environmental issues however I did not try to relate the issues of the book to the everyday lives of the students”, “few and only cursory connections made to students’ lives and experiences”, and “kept script as was presented.”

Treatment 2 documentation included: “Used students’ names in the phonics activity”, “Used student name in the sentence for dictation”, “Used student names in sentences we read during UFLI”, “Opened up discussions about race and identity”, “Focused a lot of attention on cultural experiences of the students. Celebrations with family”, “Connected refugee themes in book to students’ experiences”, “The book for the Treatment 2 group was much more interesting for the students since it is about jigging fish and most of the students have gone fishing”, “I questioned if they had ever visited a lighthouse or a beach/shoreline and if they ever had picnics like the boy and his father did in this book… all students shared personal experiences that related to the book we read today”, “Today’s book about Bobby Orr focused on skating and hockey—2 activities which interest the students in this group!”

2.5. Data Analysis

This study uses Bayesian statistics, which are useful in educational research because they can facilitate meaningful interpretations and explicitly incorporate prior knowledge. While frequentist statistics ask: “What is the likelihood of observing these data, given this model?”, Bayesian statistics ask, “What is the probability of the model parameters being true, conditional on these data?” which is arguably easier to understand. Further, credible intervals (similar to confidence intervals in frequentist approaches) are straightforward to understand, as they answer the question: Can we be 95% confident that the parameter lies between these two values? (König & van de Schoot, 2018).

Using the brms package in R version 4.4.1 (Bürkner, 2017), I fit Bayesian multilevel ANCOVA models (Gaussian likelihood) predicting vertically scaled WRMT GSV outcomes (LC, decoding, RC) from pretest score (centered), treatment condition (factor), age (centered), and home language (binary, centered), with random intercepts for classroom and RA. Where $Y$^(d) is the outcome of interest (LC, decoding, or RC), for student i in classroom j, taught and assessed by RA k:

$$Y_{ijk}^d~ N\left( {\mu }_{ijk}^d, {\sigma }_d^2\right)$$

$${\mu }_{ijk}^d={\beta }_0^d+ {\beta }_1^d\mathrm{P}\mathrm{r}{\mathrm{e}}_{ijk}+ {\beta }_2^d{\mathrm{A}\mathrm{g}\mathrm{e}}_{ijk} + {\gamma }^d\mathrm{G}\mathrm{r}\mathrm{o}\mathrm{u}{\mathrm{p}}_{ijk}+ {\delta }^d\mathrm{L}\mathrm{a}\mathrm{n}{\mathrm{g}}_{ijk}+ u_j^d+ v_k^d$$

$$u_j^d~ N\left(0, {\tau }_{class,\left(d\right)}^2\right), v_k^d~ N\left(0, {\tau }_{RA,\left(d\right)}^2\right)$$

Pre_ijk is the centered pretest covariate and Age_ijk is the centered age in years. Group_ijk is the contrast vector for the control and two treatment groups, with γ^(d) its coefficients. Lang_ijk indicates monolingual and multilingual home language background (centered at 0) with coefficient δ^(d). σ²_(d) is the residual variance; and τ terms representing classroom- and RA-level variance components, with random intercepts u_j^(d) and v_k ^(d) are independent of each other and of residuals.

König and van de Schoot (2018) found that few published educational studies using Bayesian statistics take advantage of the use of informative priors. The use of informative priors is important in small data sets, and further, the use of such priors explicitly supports the cumulative development of knowledge generated through educational research. This study tested weakly informative and informative priors (

Table 1. Weakly informative and informative priors used in the models.

	Priors
Parameter	Weakly Informative	Informative	Rationale for Informative Prior
Intercept	μ = 470, σ = 50	μ = 487, σ = 15	WRMT manual population norms
Pre-test	μ = 0, σ = 10	μ = 0.5, σ = 1	(NICHD, 2000; Florit et al., 2014; Verhoeven & Van Leeuwe, 2012)
Age	μ = 0, σ = 10	--	Presumed to be accounted for by WRMT GSV scaled scores
Group (Treatment 1)	μ = 0, σ = 10	μ = 4, σ = 5	(Al Otaiba et al., 2023; McNeill & Gillon, 2025; Melby-Lervåg et al., 2019)
Group (Treatment 2)	μ = 0, σ = 10	μ = 4, σ = 5	(Al Otaiba et al., 2023; McNeill & Gillon, 2025; Melby-Lervåg et al., 2019; Portes et al., 2018)
Multilingual home	μ = 0, σ = 10	--	Inconclusive
Residual SD	Df = 3, μ = 0, σ = 20	--	Inconclusive
Random effects
Classroom SD	Df = 3, μ = 0, σ = 10	Df = 2, μ = 0, σ = 5	(Borman et al., 2008; Savage et al., 2015)
Research assistant SD	Df = 3, μ = 0, σ = 10	--	Inconclusive

). All the parameters in the table are represented in the model. Fixed effect parameters have a mean and standard deviation, while the random effects are distributed under half-student-t. The weakly informative priors are essentially random, while the informative priors are based on the literature cited in the rightmost column. Some informative priors were left the same as weakly informative priors due to lack of consensus in the literature. I performed a sensitivity analysis to compare the models using weakly informative and informative priors with expected log predictive density (ELPD) under leave-one-out cross-validation.

3. Results

3.1. Descriptives and Correlations

Distribution of participants was n = 79 in the Control group, n = 92 in Treatment 1, and n = 92 in Treatment 2. The control group was somewhat smaller than the other two groups due to homogeneity grouping constraints and attrition. Pretest performance did not differ across control and treatment groups for any variable (

Table 2. Means (standard deviations) of the three variables pre- and post-treatment.

Variable	Control	Treatment 1	Treatment 2
Language comprehension
Pre-treatment	488 (20.1)	488 (19.9)	485 (21.0)
Post-treatment	492 (17.6)	494 (18.4)	495 (20.4)
Decoding
Pre-treatment	462 (32.4)	461 (27.9)	462 (36.0)
Post-treatment	470 (31.1)	473 (28.1)	470 (32.8)
Reading comprehension
Pre-treatment	463 (27.0)	464 (23.9)	460 (28.5)
Post-treatment	472 (23.5)	475 (20.0)	471 (26.4)

). For LC, the effect of group at pretest was nonsignificant, F(2, 260) = 0.73, p = 0.48. Similarly, for decoding, group differences were negligible, F(2, 260) = 0.00, p = 1.00. RC also did not have group differences, F(2, 260) = 0.60, p = 0.55. Thus, groups were well-matched at pretest across all three variables.

The number of sessions attended for the two treatment groups was quite similar. For Treatment 1, mean attendance was 14.4 sessions (SD = 2.65), with participants attending on average 87% of scheduled sessions. For treatment 2, mean attendance was also 14.4 sessions (SD = 3.03), with an average attendance rate of 86%. Including attendance in the model did not significantly improve model fit for any of the three constructs (LC, decoding, or RC).

Table 3. Correlations among pre-treatment, post-treatment, and covariates within each of the three constructs (decoding, LC, and RC).

Variable	Pre-Treatment	Post-Treatment	Age	Language
Language comprehension
Pre-treatment	--
Post-treatment	0.57 **	--
Age	0.34 **	0.32 **	--
Multilingual home	−0.01	0.01	0.14 *	--
Decoding
Pre-treatment	--
Post-treatment	0.82 **	--
Age	0.39 **	0.22 **	--
Multilingual home	0.27 **	0.20 **	0.14 *	--
Reading comprehension
Pre-treatment	--
Post-treatment	0.87 **	--
Age	0.44 **	0.36 **	--
Multilingual home	0.17 *	0.11	0.14 *	--

* indicates p < 0.05; ** indicates p < 0.01.

reports on the correlations within each variable at pre- and post-test, along with the covariates of home language and age. Correlations across LC, decoding, and RC were not needed because the models were run within (not across) constructs. LC had the weakest correlation between pre-test and post-test. Age was significantly correlated with all WRMT measures, which is intuitive because the GSV scores are on a vertical scale. Notably, a multilingual home environment was positively correlated with both pre- and post-test in decoding, with pre-test only for RC, and for neither pre- nor post-test for LC. In this sample, students with multilingual home environments were older than those in monolingual home environments, perhaps due to an artifact of the recruitment process and the characteristics of parents who wanted their children to participate.

3.2. Model Fit and Sensitivity

In the Bayesian multilevel model, all $\widehat{R}$ values equaled 1.00 and effective sample sizes exceeded 3000, indicating good convergence. The sensitivity analysis for the weakly informative and informative priors was evaluated according to ELPD. For decoding, the informative priors improved ELPD negligibly, by 0.2 (SE = 0.4). For LC, informative priors also improved the model, but only negligibly with an increase of 0.4 (SE = 0.2). For RC, informative priors provided no difference in ELPD, which means both models predicted equally well.

The simple ANCOVA models using informative priors were compared with models containing an interaction term of pre-test score * treatment to determine if the treatments were equally effective for students at different levels of initial ability. For LC, the simple model was negligibly better, with ΔELPD of 0.8 (SE = 2.3). For decoding, the difference was 2.4 (SE = 0.9) also in favour of the simpler model. For RC, the difference was 2.1 (SE = 3.3) in favor of the interaction model, but the difference was quite small relative to its uncertainty, which suggests similar predictive performance. There was little meaningful difference between the simple models and the models with interaction, so I chose the more parsimonious simple model. This indicates that the treatment was assumed to be equally effective for students who started out with different levels of proficiency in LC, RD and decoding.

3.3. Fixed Effects

Table 4. Fixed effects results predicting post-test score.

Parameter	Estimate (SD)	CrI (Low)	CrI (High)
Language comprehension
Intercept	491.16 (2.09)	487.15	495.38
Pre-treatment	0.48 (0.05) *	0.38	0.58
Group (Treatment 1)	3.12 (2.13)	−1.09	7.25
Group (Treatment 2)	4.90 (2.13) *	0.74	9.07
Age	2.89 (1.21) *	0.48	5.23
Multilingual home	0.53 (2.43)	−4.23	5.29
Decoding
Intercept	469.03 (2.42)	464.40	473.90
Pre-treatment	0.81 (0.04) *	0.74	0.89
Group (Treatment 1)	3.52 (2.32)	−1.01	8.07
Group (Treatment 2)	0.81 (2.27)	−3.59	5.24
Age	−2.93 (1.39) *	−5.64	−0.11
Multilingual home	−0.92 (2.75)	−6.27	4.45
Reading comprehension
Intercept	471.31 (2.18)	467.22	475.83
Pre-treatment	0.79 (0.03) *	0.73	0.85
Group (Treatment 1)	2.14 (1.68)	−1.14	5.38
Group (Treatment 2)	1.09 (1.67)	−2.19	4.35
Age	0.37 (1.09)	−1.89	2.43
Multilingual home	−1.07 (1.91)	−4.78	2.74

* indicates >95% probability and CrI does not cross zero. CrI = credible interval.

presents the fixed effects predicting the post-test results for each of the three variables (LC, decoding, RC). Credible intervals that do not cross the zero threshold can be interpreted as strong evidence that the effect is different from zero. The only group that showed substantial evidence of differential change was LC for Treatment 2, with a positive effect. Age was a positive predictor of growth for LC, and a negative predictor of growth for decoding. As expected, pre-test was a significant predictor across all three variables, although as expected from the correlations (Table 3), LC demonstrates the weakest relationship between pre-test and post-test.

Table 5. Contrasts and effect sizes comparing treatment groups and control.

Contrast	Estimate	CrI (Low)	CrI (High)	Contrast p (>0)	Posterior Mean Cohen’s d	Cohen’s d CrI
Language comprehension
Treatment 2 vs. control	4.86	1.36	8.21	99%	0.32	[0.05, 0.58]
Treatment 1 vs. control	3.11	−0.47	6.52	92%	0.20	[−0.07, 0.48]
Treatment 2 vs. Treatment 1	1.78	−1.84	5.40	79%	0.11	[0.16, 0.40]
Decoding
Treatment 2 vs. control	0.83	−3.43	5.37	64%	0.05	[−0.21, 0.31]
Treatment 1 vs. control	3.50	−1.06	7.99	94%	0.21	[−0.06, 0.47]
Treatment 2 vs. Treatment 1	−2.69	−7.31	1.90	13%	−0.16	[−0.43, 0.11]
Reading comprehension
Treatment 2 vs. control	1.10	−1.62	3.80	75%	0.10	[−0.19, 0.38]
Treatment 1 vs. control	2.13	−0.59	4.93	90%	0.19	[−0.09, −0.47]
Treatment 2 vs. Treatment 1	−1.04	−3.70	1.61	26%	−0.09	[−0.37, −0.19]

CrI = credible interval.

shows the between-group contrasts and effect sizes for the two treatment groups and the control group. In alignment with the estimates in Table 4, Treatment 2 has a 99% posterior probability of increasing LC in relation to the control group, with an estimated effect size of 0.32. No other contrast met the 95% criterion, although Treatment 1 nearly did for decoding with a 94% posterior probability.

3.4. Residual Variation and Random Effects

In terms of individual residual variation and random effects at the classroom and RA level, for decoding the individual residual variation around predicted post-test was estimated at σ = 17.15 (95% CrI [15.68, 18.74]). Variability across classrooms was modest (random intercept SD = 2.37, 95% CrI [0.12, 6.09]), as was variability attributable to RAs (SD = 2.98, 95% CrI [0.14, 7.86]). These results suggest that most of the unexplained variability in decoding occurred at the individual level rather than between classrooms or RAs. For LC, the student-level residual variation was σ = 15.24 (95% CrI [13.96, 16.66]). Variability across classrooms was modest (random intercept SD = 1.80, 95% CrI [0.07, 4.97]), as was RAs (SD = 2.03, 95% CrI [0.09, 6.10]). Like decoding, these results suggest that most unexplained variability was at the individual level. For RC student-level residual variation was σ = 11.41 (95% CrI [10.46, 12.47]). Variability across classrooms was modest (random intercept SD = 2.02, 95% CrI [0.10, 5.16]), though a bit higher for RAs (SD = 3.78, 95% CrI [0.43, 8.83]).

4. Discussion and Limitations

This study examined the effects of 10 weeks of structured LC and decoding instruction with two conditions: a generic structured approach (Treatment 1) and a culturally responsive structured approach (Treatment 2). The constructs examined were LC, decoding, and RC, which were modeled by predicting post-test outcomes conditioned on pre-test, home language environment (monolingual or multilingual), and age.

The only treatment condition that showed a meaningful effect was Treatment 2, and only for LC, with a 99% posterior probability that the effect was greater than zero (Figure 1). The estimate of 4.90 GSV points more for Treatment 2 than the control group represents 35% of annual growth in LC for a student in Grade 1 and 50% of annual growth in LC for a student in Grade 2 (Woodcock, 2011). The effect of Treatment 1 on decoding neared the threshold of credibility, with a 94% posterior probability of a positive effect. These effects should be interpreted with caution as the study was preliminary and is based on a relatively small sample and rather minimal dose (mean attendance of 14.4 sessions) over a short time frame (10 weeks).

The improvement of LC in the culturally responsive treatment group may pertain to students becoming more attuned to meaning and language as they listened to and discussed stories that pertained to their lives. Expectancy-value theory (Eccles & Wigfield, 2002) predicts that people engage more deeply in tasks that they perceive as personally valuable and in which they expect they can be successful. Students in Treatment 2 may have experienced greater expectations of success in listening, due to the structured and scaffolded language comprehension instruction (Palinscar & Brown, 1984) and personal connection to the material and instructor. The instructional model encouraged personal value in listening and attending by connecting content to students’ lived experiences and by building a culturally safe learning environment. It is possible that participating in Treatment 2 increased students’ motivation and engagement, enabling them to attend and focus more deeply during learning activities.

Strategy instruction and content instruction have both shown in previous research to improve RC outcomes, as demonstrated by Clarke et al. (2010, 2017) and J. S. Kim et al. (2021a, 2021b, 2023, 2024), and many others. The present study offered strategy instruction, and it focused on content, but the goal was not to teach about specific curriculum content. Rather, Treatment 2 content was designed to be personally relevant for students. The goal was to determine if the foundational skills that predict RC (decoding and LC) or RC itself are differentially responsive to pedagogy that actively connects learning activities and content to students’ lived experiences. Indeed, this was found to be the case for Treatment 2 for LC, which demonstrated greater growth than for Treatment 1. The locally developed, highly personalized oral language instruction demonstrated transfer to a standardized measure, the WRMT. Because this transfer did not occur for LC in Treatment 1—though it did approach the threshold of credible evidence with a posterior probability of 92% in relation to the control group—the effect can be inferred to a cause of receiving culturally responsive instruction.

Hammond’s (2015) framework connects culturally responsive teaching with cognitive science. She emphasizes the importance of building learning alliances with students, grounded in trust and rapport. Without such positive relationships, attempts at rigor or constructive feedback can trigger an “amygdala hijack” in students (p. 91)—an activation of the fight or flight cortisol response. Social–emotional threats like microaggressions and stereotyping can also elicit the amygdala hijack, which causes “all other cognitive functions such as learning, problem-solving, or creative thinking” (p. 40) to halt. Hammond suggests teachers position themselves as allies and “warm demanders” who offer “both care and push” (p. 95). The rapport warm demanders build, their belief that all students can learn, and their high levels of pedagogical competence earn students’ respect and their investment in learning. Treatment 2 emphasized such a culturally responsive and rigorous approach.

The present findings align with the Active View of Reading (Duke & Cartwright, 2021) which articulates a “mechanism by which the sociocultural context and the reader’s social identity, including racial, religious, socioeconomic, gender, and many other sociodemographic aspects of identity, impact the reading process” (p. S38). This study’s results represent preliminary evidence that students’ engagement with culturally relevant content improved LC as assessed by a standardized measure; in other words, transfer occurred from local and personalized instruction to standardized, population-level assessment. A potential explanation is offered by Cantor et al. (2019), who state that culturally responsive teaching primes the learner to make new conceptual connections by using cultural knowledge as a scaffold that connects current knowledge to new concepts. This process promotes effective information processing that draws on the learner’s long-term-memory as well as motivation from using their own expertise (p. 320). Exploring the role of motivation would present a fruitful next step in understanding the mechanisms for comprehension skill transfer.

Over 30 years ago, psychologists Sternberg and Frensch (1993) discussed the mechanisms that facilitate transfer from one domain to another, and they highlight the importance of connecting taught content to students’ lives. The mechanism of organization is the clarity and logic teachers use to frame the content, but it is not constrained to the school context:

“The various pieces to be learned should make sense in terms of each other, and in terms of other information, the student has about the world. Such connections are often not drawn. Oddly enough, it is the exception rather than the typical teacher who starts a course or a lesson with a discussion of why and how what is to be learned is important to the students’ lives. If the teacher does not know, the student cannot be expected to know either.”

(p. 35, emphasis added)

The specific nature of the mechanism for LC transfer in the present study is unknown. It could potentially pertain to attention, motivation and engagement, or schema and concept development—or all of these. Further research is needed to understand the mechanisms that generated these results. Another factor to consider is the relatively low published reliability for LC as measured by the WRMT, which for Grade 1 is 0.73. Future studies can explore more precise measurement instruments for LC for young children.

Early papers describing the Simple View of Reading suggest that transfer of comprehension skills from LC to RC is largely automatic (Gough & Tunmer, 1986; Hoover & Gough, 1990). Clarke et al. (2010) indeed found that oral LC intervention supported RC growth. The results in the present study, where LC was responsive to the culturally responsive oral LC instruction in Treatment 2, but RC was not, may pertain to the population sample. Clarke et al. (2010) recruited students with specific comprehension challenges—specifically students with a large discrepancy between RC (lower) and reading fluency (higher). Thus, the oral LC instruction provided in their study was arguably a good match for the sampled students’ challenges. The present sample includes students from a full range of reading achievement levels (as requested by the participating school board). Students were grouped by RC, and all students received the same proportion of LC and decoding instruction, rather than instruction tailored to their learning needs (e.g., more or less decoding and LC instruction). Further, coincidentally, two participating schools were enacting structured literacy instruction for the first time during the course of this study, which could potentially explain the growth seen by all three groups (Treatment 1, Treatment 2, and business-as-usual control group).

It is possible that culturally responsive instruction did not show differential improvement for decoding because decoding is a more discrete and constrained skill than LC (Paris, 2005). Contextualization may have a more positive influence on the learning of unconstrained skills (e.g., vocabulary and comprehension) than on constrained skills (e.g., learning letter sounds) (Arya & Maul, 2021; Roberts, 2021). However, beyond this possible theoretical explanation, there is a relevant limitation in the present study. RAs participated in initial and ongoing training, but the research activities drew significantly on their professional decision-making in localizing Treatment 2. Based on the documentation and rationales, RAs appears to be more adept at differentiating the two treatments for LC than for decoding. For the latter, a common refrain in the activity descriptions was that students’ names were used in the generation of connected text for Treatment 2. Given the highly regimented nature of the UFLI objectives (which proceed developmentally through individual letters, digraphs, etc.) and the use of high-frequency words in the UFLI lessons, RAs may have been challenged to generate sentences and words that were sufficiently tailored to students’ interests and backgrounds while also complying with UFLI’s specificity of orthography. It appeared to be easier for RAs to achieve fidelity in differentiating treatment conditions for LC than for decoding. In future studies, more support and higher dosage over a longer term could potentially reveal stronger effects for constrained skills. Working with students at somewhat higher level of decoding skills could also enable greater differentiation in decoding materials without the constraints of using only a small set of orthographic patterns.

Covariates of multilingual home and age were included in the models, as were random effects at the level of classroom and research assistant. Multilingual home environment did not emerge as a significant predictor in the models, which means that both multilingual and monolingual students responded similarly. Language proficiency was not measured, so inferences cannot be generated from these results as to the effectiveness of the interventions for children at different levels of language proficiency. Correlations (Table 3) suggest students from a multilingual home had stronger decoding skills than monolingual peers at both time points, stronger RC at pre-test, and essentially equivalent LC skills at both time points. However, bivariate correlations do not hold other variables constant, so no inferences should be generated from these correlation coefficients. Participants’ age was significant in the subskill models but not in RC: age was a positive predictor in LC and a negative predictor for decoding. This means that older students made greater gains in LC, while younger students made greater gains in decoding, which aligns with the literature on the relative growth and influence of these two subskills on RC as children mature (Catts et al., 2005; Foorman et al., 2020; Lonigan et al., 2018; Tilstra et al., 2009). As for the random effects, for all three focal constructs most unexplained variance was at the student level, with only modest variation attributable to classrooms or RAs. This pattern suggests differences in response to instruction were largely driven by individual factors, rather than which RA led the small group instruction or the nature of classroom instruction outside of the small groups.

All three groups grew in all three skills over the course of the study (10 weeks), which is good news, as it means the “business-as-usual” control condition was making a positive impact on students. The fact that two schools with the largest number of participating students had recently begun a structured literacy approach to decoding instruction is par for the course when performing interventions “in the wild”. The simultaneous improvement in control conditions may have driven the lack of differential impact in either treatment group on decoding and RC (as RC is largely driven by decoding in the early elementary grades). Future studies could employ a finer-grained approach to explore trajectories of students who begin the study with LC difficulties, decoding difficulties, or both.

Bayesian statistics offer interpretable inferences based on the posterior probabilities that the model parameters take on particular values given the observed data and prior information. The Bayesian approach can support modelling flexibility and rich inferences, even with small samples. König & van de Schoot (2018) suggest that beyond these, one of the most promising attributes of Bayesian modeling is that prior information can be explicitly incorporated into the analysis, enabling cumulative knowledge development. In this study, the use of informative priors did not improve the model substantially in comparison to the weakly informative priors. Thus, the lack of improvement with informative priors suggests that the observed data were strong enough to dominate the posterior, yielding stable parameters without substantial reliance on priors.

Generally speaking, LC explains only a small amount of variance in RC when children are young, but its influence grows over time. Catts et al. (2005) examined the prediction of decoding and LC for RC across grades 2–8, finding LC’s unique contribution steadily increased while decoding’s steadily decreased. In line with this, Foorman et al. (2020) reported that LC’s unique role in RC grew from grades 5–9. Lonigan et al. (2018) observed similar trends from grades 3–5. Y.-S. G. Kim and Wagner (2015) modelled RC as predicted by decoding and LC, mediated by fluency; LC increased in importance from grades 1–4. Thus, the finding that LC was malleable to culturally responsive and structured instruction, including transfer to a standardized measure, is potentially quite meaningful for educators, as LC is known to play an increasingly critical role in RC over time. Future studies can explore the attributes of context and instruction that may help LC growth transfer to RC growth.

While the research base on culturally relevant literacy instruction is mature, there remains a theory–practice gap, as many teacher preparation programs and school curricula may not address cultural relevance (Ahmed, 2019; Dukes et al., 2021). At the same time, many teacher preparation programs also lack strong integration with findings from cognitive science. The Active View of Reading offers a pedagogically useful enhancement of the Simple View of Reading, by specifically including cultural knowledge and sociocultural context, as well as cognitive processes. This study provides some evidence in support of the Active View of Reading by describing on how culturally responsive instruction may impact children’s reading development. These insights can potentially support the ongoing evolution of evidence-based reading instruction. Further research is needed, as this was a pilot study with the goal of exploring new and instructive ways of conceptualizing reading education. The rich diversity of contemporary classrooms calls for innovative research so all students can meet and exceed their potential, and this study sought to answer that call.