Written language development depends on phonological information processing, which relates – among other things – to the processing of acoustic signals of the spoken language (Ptok et al., 2008; Wagner and Torgesen, 1987) and is often described as a compound of four sub-domains: phonological awareness (PA), phonological working memory (PWM), rapid automatized naming (RAN) and letter knowledge (LK) (Vellutino et al., 2004; Wagner and Torgesen, 1987; Wolf and Bowers, 1999). In addition to PA, language abilities influence spelling achievement (Ennemoser et al., 2012; Fricke et al., 2016; Lervåg and Hulme, 2010).

Phonological awareness is a metalinguistic skill defined as awareness of the sound structure of language (Wagner and Torgesen, 1987) and the ability to analyze, segment, and manipulate phonological units (e.g., Schnitzler, 2008; Tunmer and Hoover, 1992). The term phonological awareness is used extensively in research and focuses on phonological units of various sizes: larger units are word, syllable, and rhyme, while the single phoneme is a small unit (Skowronek and Marx, 1989). Smaller units are acquired later than larger units and are more difficult to manipulate.

In addition to the size of the phonological unit, the operation process matters. Analysis is easier than segmentation or synthetization, and both are easier than manipulation (Schnitzler, 2008). Addressing the question of whether children with limited phonological competencies benefit in the long term from training, Schneider et al. (1998) reported significant training effects, particularly in spelling, when the training adhered precisely to instructions.

Phonological working memory allows temporary sound-based representations or linguistic information to be kept available until cognitive processing is complete (Baddeley, 1992). It is interrelated with PA and influences spelling development (Gindri et al., 2007; Steinbrink and Klatte, 2008). Working memory is described as a “capacity-limited system that serves the short-term storage of information” (Mähler and Schuchardt, 2014, p. 77). The capacity of working memory seems particularly relevant for children at the beginning of the reading acquisition process (Schneider, 2009). It is considered a stable trait-like ability that shows resistance to systematic training and has primarily predictive characteristics (Mayer, 2016, p. 77).

Rapid automatized naming (RAN) relates to the fast pronunciation of objects, colors, or digits (Georgiou et al., 2009). In their Double-Deficit Hypothesis, Wolf and Bowers (1999) described children with dyslexia not only in terms of deficits in phonological awareness but also in terms of differences in naming speed. There is strong agreement that children with reading difficulties, especially in consistent orthographies, can be characterized by deficits in naming speed (Brizzolara et al., 2006; Mayer, 2016). The most significant specific predictor for spelling, however, remains phonological awareness (Landerl and Wimmer, 2008). Korhonen (1995) presented RAN as a personality trait, which indicates that difficulties in rapid naming persist into early adulthood. The specific components of the complex construct of naming speed, which is determined by the interplay of visual processing speed and speed of access to phonological information, are subject of intense discussions (cf. Mayer, 2016). Given the well-established association between RAN and literacy acquisition, it is reasonable to assume that RAN-trainings have the potential to improve literacy, nevertheless, empirical evidence of the effectivity of RAN-based training methods is rare (Berglez, 2003; Mayer, 2016). Berglez (2003) posited that children exhibit slow naming speed due to insufficient practice, and suggested that their actions in early childhood have been insufficiently accompanied by speaking or naming. “This weakness is to be compensated by training children to name different categories of objects faster” (p. 123). In a study in preschool children, she observed significant improvements in naming speed, which indicated its fundamental trainability. However, these improvements could not be attributed to specific training effects and even affected the control groups without training. Importantly, to date there have been no reports on the effects of training to improve rapid naming on later reading and writing.

Letter knowledge defines the explicit knowledge of letter names (e.g., Ennemoser et al., 2012; Foulin, 2005) and is an important predictor of reading (Foorman et al., 2016; Malling et al., 2022; Schatschneider et al., 2004). Its predictive accuracy depends on the transparency of the orthography (Martínez and Goikoetxea, 2019). Further, the relative contributions of rapid automatized naming, letter sound knowledge, and phonological awareness differ before and after school entry (Schmitterer and Schroeder, 2019).

There is ample evidence that – together with working memory (Berninger et al., 2010) – phonological awareness is the strongest predictor of later spelling abilities (e.g., for German-speaking children, see Ennemoser et al., 2012; Pfost, 2015 for a review). Wolf and Bowers (1999) emphasized the close connection between RAN and reading, and Ennemoser et al. (2012) postulated for RAN a persistent but lower correlation to spelling than to reading.

The vocabulary is built and expanded by recognizing new words and storing them in the mental lexicon together with the semantic, syntactic and morphological information and the phonetic-phonological form for each lexical unit in a network (Levelt, 1998). This organization and structure ensures efficient word retrieval in comprehension and in production and in the oral and written modalities (Aitchison, 1997). Although the size of the mental lexicon increases dramatically at preschool age (Clark, 1995; Kauschke, 2000 for German), it takes several years to establish the lexical representations in an adult-like manner because storage of detailed phonological, morphosyntactic, and semantic information requires children to hear words in various situations (Oller and Eilers, 2002). Two characteristics of lexical development are important with regard to our study: First, lexical development largely takes place on an item-by-item basis; that is, learners must hear the words in order to acquire them. Vocabulary size is thus strongly influenced by socio-cultural factors (Hoff, 2003, but see, e.g., Zaretsky and Lange, 2017; see section 1.2). Second, lexical development is never complete, because new words are integrated into the mental lexicon and existing representations are modified throughout one’s entire life (Engelkamp and Rummer, 1999).

In both of the above regards, vocabulary differs from grammatical knowledge, which is largely based on rules that are mastered at a certain point in development (Schulz, 2007). Receptive and productive grammatical knowledge grows dramatically at preschool age. Between 2 and 3 years of age, German-speaking children produce main clauses with correct subject-verb agreement, and embedded clause structure is acquired between 2;6 and 4 years of age (e.g., Clahsen, 1986; Rothweiler, 1993). Due to complex rules and a high numbers of exceptions, other phenomena in morphosyntactic development, such as case and gender marking, plural formation, and inflection of irregular verbs, remain difficult until primary school age and beyond (Schulz, 2007). German-speaking children can rarely infer plural formation, gender marking, and irregular verbs by applying rules; instead, many derivations must be learned item by item (Clahsen, 1986). Given that written language encodes grammatical information (Bredel, 2015; Fuhrhop and Peters, 2023; Maas, 2015), grammatical (morphosyntactical) knowledge influences reading development (Ennemoser et al., 2012) and word spelling, even in first graders (Fricke et al., 2016; Lervåg and Hulme, 2010; Author et al., under review).

The spelling achievement of children is strongly associated with the socioeconomic status (SES) of their parents. SES – referring to the relative social position of individuals – is a multifaceted construct that is frequently measured using one or more of the following three indicators: income, education and occupational status (Bradley and Corwyn, 2002; Conger and Donnellan, 2007; Sirin, 2005). It has repeatedly been found that children from low-SES families show lower verbal academic achievement than children from high-SES backgrounds (Sirin, 2005; White, 1982). With regard to spelling, there is also consistent evidence that low-SES children have poorer orthographic skills than their peers (see Breit et al., 2016; Niemietz et al., 2023 for German speaking countries). On average, children from high-SES backgrounds have a larger vocabulary, can use more varied and complex grammatical forms and show greater phonological awareness than children from low-SES backgrounds (Fernald et al., 2013; Gilkerson et al., 2017; Hoff, 2003; Huttenlocher et al., 2010; Li et al., 2023; Noble et al., 2007; Pace et al., 2017; Rowe, 2008). This gap in language-related skills develops before school enrollment, and it influences academic achievement throughout elementary school (Hindman et al., 2010; Morgan et al., 2015; Von Stumm et al., 2020).

Pace et al. (2017) described several pathways by which SES may affect language development in general and consequently also spelling. In line with general models of the influence of SES on child development (e.g., Bradley and Corwyn, 2002; Conger and Donnellan, 2007), SES is considered to be associated with the quantity and quality of generic (e.g., quality and quantity of general parenting behaviors) and language-specific (e.g., quality and quantity of language input provided by parents) parent–child interactions. It is assumed that low-SES parents show less positive parenting (e.g., less sensitivity) due to, amongst others, strain associated with a low SES (Conger and Donnellan, 2007), and provide less (qualitative) language input, which in turn affects language development (e.g., vocabulary and grammatical knowledge). Limited economic resources of low-SES parents are also associated with limited availability of learning materials (e.g., number of books) relevant for language development. Finally, SES may be associated with other skills that are important for language learning (e.g., processing efficiency, fast mapping skills). Low-SES children may therefore be characterized by their lower levels of the skills needed for language development (Pace et al., 2017).

According to Grosjean (2012), children are bilingual if they use two or more languages in their daily life. Learning the language of instruction as a second language is a further factor associated with spelling development. Overall, international large-scale studies such as PISA (Programme for International Student Assessment; OECD, 2023) indicate that children, who learn the language of instruction as a second language (and speak another language at home) often have a migration background. A migration background and the use of a heritage language at home are strongly associated with lower academic achievement, particularly in German-speaking countries (e.g., OECD, 2023) – the context of the current study. There is consistent evidence that migration background and heritage language use are associated with lower reading scores in primary and secondary education. Notably, some of these differences are explained by the lower SES of migrant families (e.g., Breit et al., 2016; Henschel et al., 2023). Likewise, studies suggest that students who acquire German as their second language (Lenhart et al., 2019) and students with migration background (Henschel et al., 2023) lag behind their peers in their spelling skills. Again, lower SES accounts for some of the association of spelling achievement with migration background (Niemietz et al., 2023).

Bilinguals can differ from age-matched monolingual children not only in SES, but also in their oral language abilities. Due to dual language exposure, bilinguals can show accelerated or decelerated language development (Paradis and Genesee, 1996). Bilingualism can foster the acquisition of language and cognitive abilities, as expressed, for example, by the notion of a ‘bilingual advantage’ (Bialystok, 2011; Blom et al., 2014). A bilingual advantage has sometimes been reported for phonological awareness, which is a key predictor of written language achievement (Bialystok et al., 2003; but see Antoniou, 2019; Cat et al., 2018; Ross and Melinger, 2017 for recent evidence opposing a bilingual advantage). In addition, typological differences in the oral marking of word and sentence structure can increase the awareness of word boundaries (Veldhuis, 2015), which is relevant for spelling. However, at pre-school and primary-school age, bilinguals often lag behind monolinguals in their lexical (Czapka et al., 2019; Klassert et al., 2014; Seifert et al., 2019; for an international review see Bialystok and Luk, 2012) and grammatical (e.g., Schulz and Grimm, 2019) abilities. A decelerated acquisition is reported in particular for the sub-group of successively bilingual children, that is, children who start to acquire the ambient language with entry to kindergarten or with school enrolment (Paradis et al., 2021).

The oral language abilities are associated with the spelling outcome in several ways. Lexical and morphosyntactic abilities are required to conceptualize the text and to produce written morphemes and sentences (Ehri, 2017; Silverman et al., 2015). Due to dual language development, many bilingual children cannot rely on the same language resources as monolingual children when learning to spell, for example because they started to acquire the majority language successively to the heritage language (Schulz and Grimm, 2019). Bilingualism can thus indirectly – modulated by the oral language abilities – have a negative impact on spelling achievement. We emphasize that the difficulties are not caused by the factor bilingualism per se and that they do not represent a deficit of the child. Rather, spelling difficulties observed in bilingual children result from the education system (here: the Austrian), which takes monolingualism as the norm (‘monolingual habitus’, Gogolin, 2008), and from inadequate spelling instruction (Bredel et al., 2017). Taken together, bilingualism matters in spelling achievement; not because there is a general difficulty in learning to spell in an L2, but due to the lower SES, and because the Austrian (and similarly the German) education system does not take typical manifestations of dual language exposure into account.

As outlined above, there is evidence that bilingualism and SES are associated with spelling (see Figure 1, path c on the left; e.g., Breit et al., 2016; Henschel et al., 2023; Lenhart et al., 2019; Niemietz et al., 2023). There is also evidence that bilingualism and SES are associated with spelling precursors (see Figure 1, path a on the left). While low-SES children are considered to lag behind their peers in lexical and grammatical abilities (Gilkerson et al., 2017; Pace et al., 2017), for bilinguals differential effects are expected, specifically, lower language skill levels (grammar and vocabulary; Czapka et al., 2019; Schulz and Grimm, 2019; Seifert et al., 2019). Since findings regarding a potential bilingual advantage in phonological awareness are inconsistent (e.g., Antoniou, 2019; Cat et al., 2018; Ross and Melinger, 2017, see Section 1.3), we had no specific hypothesis in this regard. Finally, there is a large body of research highlighting the importance of language skills (grammar and vocabulary) and phonological information processing (Bigozzi et al., 2016; Ennemoser et al., 2012; von Goldammer, 2010; Kim et al., 2013; Pinto et al., 2015) to the development of spelling skills (see path b on the left side of Figure 1). Integrating these three paths into a single combined model produces a mediation model (MacKinnon, 2013; shown on the right-hand side of Figure 1) that – to the best of our knowledge – our study was the first to test. This model assumes that parental education (as a measure of SES) and bilingual status are associated with language skills and strongly with specific precursors of phonological information processing at the time of school enrollment (path a); these skills in turn predict spelling skills assessed at the end of grade 1 (path b). Path c’ reflects the effect of bilingual status and parental education on spelling that is not accounted for by the mediators. Notably, we expected that – in line with a mediation hypothesis – there are significant indirect effects of bilingual status and parental education on spelling. Considering similar research into the prediction of reading (Li et al., 2023), we hypothesized that at least for parental education the c’-path is significant.

This study used data from two cohorts (2021/22 and 2022/23) of first graders from Upper Austrian schools, mainly from the school district Linz-Land. Recruitment proceeded as follows: First, we obtained written permission from the regional school board for Upper Austria to contact schools for participation. Second, we informed the head teachers of 11 schools about the project and invited them to participate. Since all head teachers agreed to join the project, we asked the first-grade teachers to distribute letters to parents that included general information about the project, a consent form, and a written questionnaire for them to complete and return to the school. 1,070 questionnaires were collected. For this study, the analysis excluded 58 children for the following reasons: 20 children were ill during both testing periods (i.e., at school entry and at the end of grade 1), and parental consent was not received for 38 children. Thus, data on 1,012 children (from a total of 61 classes) were used for this study. The sample comprised 48.5% girls, which reflects the proportion of girls among Upper Austrian first graders [46.7%; χ²(1) = 0.736, p = 0.391].

73.0% of the children grew up monolingually with Austrian German. This proportion of monolinguals corresponds to the rate of Upper Austrian first graders that speak exclusively German in their everyday life [78.9%; χ²(1) = 0.014, p = 0.907]. Reference values for the percentages of girls and bilinguals, respectively, were taken from the Austrian School Statistics (Statistik Austria, 2024).

The overall sample consisted of parents from all educational backgrounds. Using the highest level of parental education (i.e., that of the parent with the higher level of education), in 44.6% of families at least one parent had a university degree. About a fifth (19.6%) had a university entrance qualification. Another 30% reported a vocational education or training, and in 5.9% of the families neither parent had a qualification beyond compulsory schooling. Compared to the Upper Austrian parent population of 4^th graders in the school year 2017/18,¹ parents with a university degree were overrepresented (44.6% vs. 26.5%), and parents with vocational education or training were underrepresented (30.0% vs. 45.8%; χ²(3) = 170.85, p < 0.001). The sample was thus representative in terms of child gender and bilingual status, but not in terms of parental education, which is probably due to the sample including mostly schools in urban areas, where more parents with a higher level of education live (see also Elliott, 2018; Yulianti et al., 2023).

The analyses described are part of a longitudinal project (SCHNAPP) that aims to assess and monitor early reading and writing development of children in primary schools. Two cohorts were assessed (a) at the very beginning of the first year of schooling (in 2021 and 2022) in terms of precursors of spelling (language and phonological information processing) and (b) at the end of first grade (in 2022 and 2023) by a spelling test.

The Schnapp spelling test (Schöfl, Steinmair, et al., 2023) was administered in class via tablet PCs. The items of this test are embedded in a cloze task, and sentences have been pre-recorded and presented as audio via headphones. Children were asked to write down the target words on the tablet using a digital pen. Unlimited repetitions and corrections were allowed. The mean duration was 15 min (SD = 7.2). Teachers were present but were asked not to interact with the children during the test. Written feedback about the performance of each class and a video-explanation were given to head teachers at the end of the assessment phase.

Oral language-related precursors and child-related information were assessed in a one-on-one setting (i.e., individual administration). Informed consent was gained from all parents, and only children whose parents had given permission participated. The assessments were carried out by trained students and project staff. The test was embedded in a magical story, presented on a tablet, to keep the children’s attention and maintain motivation. Assessors adapted to the working pace of the children; the test took 19 min on average (SD = 5.2). A questionnaire was developed to collect information from parents about the child’s age and whether s/he was growing up monolingually or bilingually, attested language difficulties, language use within the family, and the parent’s socio-economic background (assessed via parental education).

We applied a structural equation modeling (SEM) approach (e.g., González-Valenzuela et al., 2023) using Mplus 8 (Muthén and Muthén, 1998-2017) to test the mediation model outlined above. Notably, as the current study used clustered data (i.e., children in classes), variance in the spelling test might be due not only to bilingualism, parental education and precursors of spelling, but also to differences in the first year of schooling (e.g., differences in instruction and other teacher effects). However, our hypotheses focused on the individual level (level 1): We sought to explain individual differences in spelling that are not due to class-level variables (level 2). Thus, we applied multilevel SEM (see, e.g., Preacher et al., 2010), where we focused on the mediation process at level 1 to explain variation in spelling within school classes². Notably, although classes differed in terms of parental background, proportion of bilinguals, language skills and phonological information processing (see Supplementary Table A1), these differences were not due to schooling, as these variables were assessed at the very beginning of formal schooling and some are time-invariant (parental education, bilingualism). Therefore, these variables were not decomposed into individual and class-level components, but treated as individual-level variables.³ The corresponding model is shown in Figure 2. The statistical significance of the indirect effects – supporting the proposed mediation hypotheses – was calculated using the delta method (MacKinnon, 2013). The preferable bias corrected bootstrap approach is not available for multilevel models in Mplus 8.

The rate of missing data ranged from 0% (bilingual status) to 16.4% (spelling). Little’s Missing Completely at Random (MCAR; see, e.g., Enders, 2010) test as implemented in SPSS 29 was significant [χ²(80) = 302.049, p < 0.001], which indicates that nonresponse depended on the observed variables [i.e., Missing at Random (MAR)]. In detail, supplementary analyses indicated systematic dropout at time 2 (i.e., the spelling assessment). Children with missing data on the spelling test were more likely bilingual, had parents with lower education, and showed comparably lower scores on the subtests for language skills and phonological information processing. Since these results suggested a MAR mechanism, we applied a Full Information Maximum Likelihood Estimation to appropriately deal with missing values.