Predictors of Reading and Spelling Difficulties in Italian Children: Specific Language and General Cognitive Skills

Domain-Specific Predictors

The early screening protocols were mainly based on a theoretical model that posits phonology at the base of reading success and failure (Griffiths & Snowling, 2001; Vellutino et al., 2004). Research has also recognized the critical role of phonological development in acquiring written language (Hulme et al., 2015; Snowling et al., 2019). According to these researchers, reading descends from the increasingly automatic identification of groups of linguistic units (i.e., graphemes, syllables). Indeed, accumulating evidence suggests that different phonological structures influence the development of phonological awareness (Melby-Lervåg et al., 2012). For example, Snider (1995) noted: “Children become aware of larger linguistic units (words and syllables) before they become aware of smaller linguistic units (phonemes)” (p. 444). Thus, phonemic awareness is the last ability to develop, with onset-rime awareness serving as an intermediate stage (Konza, 2011). The causal link from phonological awareness to success in learning to read has been extensively discussed, with strong evidence supporting this relationship (Castles & Coltheart, 2004).

Phonological working memory is also a significant predictor of literacy difficulties (Baddeley, 2003; Gathercole & Baddeley, 1993). Deficits in phonological awareness and short-term memory can hinder the development of sublexical processes, which are crucial for literacy acquisition in alphabetic writing systems (Daucourt et al., 2018). It is widely recognized that children with dyslexia speaking different languages struggle with phonological awareness skills (Landerl et al., 2013). For example, Ziegler et al (2010) analyzed the predictive role of phonological awareness across different orthographic systems and found that its impact varies with orthographic depth. Thus, in transparent orthographies like Italian, the predictive role of phonological awareness is lower than in opaque orthographies like English. Moreover, its predictive role diminishes after the first year of schooling in transparent orthographies.

Landerl et al (2019) found that rapid automatized naming (RAN), rather than phonological awareness, consistently predicts reading fluency across different orthographies. Moreover, other studies have found that RAN is a predictor of not only reading speed but also reading accuracy development (Kirby et al., 2010; Swanson et al., 2003). Further, RAN has been shown to be related to reading comprehension (Arnell et al., 2009; Johnston & Kirby, 2006), indicating its broad role in reading (Araújo & Faísca, 2019; Huschka et al., 2021).

Domain-General Predictors

While the roles of phonological awareness and RAN are well documented (Manolitsis et al., 2009; Norton & Wolf, 2012; Schatschneider et al., 2004; Schmitterer & Schroeder, 2019; Torppa et al., 2007; Wolff, 2014), other cognitive skills also contribute significantly to early literacy (Burgoyne et al., 2019; Moll et al., 2014, 2016; Pasqualotto et al., 2024; Varvara et al., 2014). Efficient text decoding requires visual attention distribution on the page to select appropriate graphemes while inhibiting orthographic competitors (Grainger et al., 2016; Lallier et al., 2013; Vidyasagar, 2013). Thus, inhibition is crucial for selecting the correct phoneme for a grapheme and suppressing phonological competitors (Frisson et al., 2014; Ziegler et al., 2014). Cognitive flexibility is also important for building cross-modal connections and switching between various dimensions of written content such as phonology, morphology, semantics, and syntax (Blair & Razza, 2007; Colè et al., 2014; Georgiou & Das, 2016; Peng et al., 2019; St Clair-Thompson & Gathercole, 2006).

Longitudinal studies have shown that slow auditory attentional shifting in pre-readers at risk for dyslexia significantly predicts future reading impairments in primary school (Blockmans et al., 2023; Boets et al., 2008, 2011; Franceschini et al., 2012). For instance, Franceschini et al (2012) demonstrated that early attentional abilities, measured before reading acquisition, can predict later reading skills. Recent research further highlights the role of auditory attentional shifting, showing that it influences the learning trajectories of symbol-speech sound correspondence and modulates neural sound tracking in children with and without dyslexia (Guerra et al., 2024). Indeed, impaired attentional control in dyslexia has been linked to greater deficits in reading fluency and reduced activation in the typical left-hemisphere phonological and orthographic reading networks (Chyl et al., 2021).

Additionally, some domain-general deficits related to cognitive control are associated with developmental dyslexia (Pasqualotto & Venuti, 2020; Varvara et al., 2014), especially deficits in visual attention (for recent reviews, see Tang et al., 2023; Valdois, 2022). Such deficits can affect letter identification, letter position coding, processing of letter strings in parallel, and programming of eye movements, all of which are necessary for developing automatic word recognition and reading skills. Beyond visual attention deficits, weaknesses in cognitive control, such as poor working memory or deficient inhibitory processes, have also been identified as associated with dyslexia (Johann et al., 2020; Morris & Lonigan, 2022; Peng et al., 2019.

However, some meta-analyses and systematic reviews suggest that domain-general cognitive skills do not predict reading skills beyond foundational literacy skills such as phonological processing and comprehension. For example, Quinn and Wagner (2018) found that while there were large correlations between cognitive components (such as working memory and processing speed) and decoding and linguistic comprehension, a direct pathway from these cognitive skills to reading was not supported. A recent meta-analytic review by Peng et al (2022) demonstrated that children with developmental disabilities (DD) showed deficits across processing speed, inhibition, switching, working memory, and visuospatial skills. However, this review also found that domain-general skills may make unique contributions to reading difficulties beyond decoding and language comprehension, but their influence is limited when foundational literacy skills are accounted for.

Domain-Specific Predictors

Phonological awareness is a critical domain-specific predictor for spelling. Numerous studies have demonstrated a strong relationship between phonological awareness and spelling skills (Galuschka et al., 2020; Moll et al., 2014). For example, orthographic knowledge, involving understanding the conventions and rules of written language, significantly contributes to spelling accuracy (Apel, 2011; Ehri, 2014). Morphological awareness, the understanding of word structure and the ability to manipulate morphemes, is also essential for spelling skills (Grigorakis & Manolitsis, 2021).

Rapid automatized naming (RAN) has been extensively studied as a predictor of reading skills, but its role in spelling is also significant. RAN is an independent predictor of word spelling (Georgiou et al., 2012; Savage & Frederickson, 2006). Several studies indicate that RAN supports the development of orthographic representations of words, but its relationship with spelling varies across languages with different orthographic complexities (Georgiou et al., 2016; for a recent meta-analyses, see Chen et al., 2021). A study by Huschka et al (2021) on German children revealed that while RAN components like pause time and pause time consistency were significant predictors of reading fluency, they did not predict spelling performance, highlighting the complex nature of spelling development.

Domain-General Predictors

More domain-general cognitive functions, such as working memory, cognitive flexibility, and inhibitory control, are crucial predictors of spelling (Kim & Park, 2019). Working memory, in particular, is essential for holding and manipulating information during the spelling process (Swanson & Berninger, 1995; Vanderberg & Swanson, 2007). Studies have shown that working memory capacity is related to spelling performance (DeBono et al., 2012). Cognitive flexibility, the ability to switch between different tasks or mental processes, is important for spelling because it allows children to switch between different linguistic rules and apply them correctly (Berninger et al., 2008; Drijbooms et al., 2015; Quinlan et al., 2012). Inhibitory control, which is the ability to suppress irrelevant information and stay focused on a task, plays a significant role in filtering out inappropriate lexical representations at the word level, grammatical structures at the sentence level, and irrelevant ideas at the text level (Altemeier et al., 2008; Kellogg et al., 2013).

From this short review, it is evident that both domain-specific and domain-general cognitive skills play crucial roles in the development of reading and spelling abilities (Costa et al., 2018; Daucourt et al., 2018; Ruffini et al., 2024; Ziegler et al., 2020). However, the existing literature shows variability in how these predictors impact literacy outcomes across different languages and orthographies. Therefore, despite the wealth of research, there is still a need to explore these relationships in the context of the Italian language, which has been relatively understudied. The present study aimed to fill this cap and contribute to the literature on the relationship between reading and reading-related linguistic and cognitive skills, focusing on Italian, an understudied language. The study involved a sample of 239 Italian-speaking children from the Trentino area in northern Italy.

Participants

A total of 239 Italian first graders, aged from 5 years to 6 years and 11 months (mean age = 6 years and 3 months; SD = 5 months), of both sexes, who were enrolled in elementary public schools participated in the study. Children were recruited through three primary schools in Trentino's urban area. Parents and/or guardians of all the participants received a detailed explanation of the procedure of the study and provided a signed informed consent, in accordance with the Declaration of Helsinki.

All the children were native Italian speakers with normal or correct-to-normal vision and hearing abilities. The children's IQs were within the normal range (IQ > 85). Children with any psychological, neurological, or medical diagnoses, such as specific language impairment (SLI), attention-deficit/hyperactivity disorder (ADHD), autism spectrum disorder (ASD), epilepsy, or other relevant neurological and medical conditions were not included in the analysis. Notably, in Italy, the diagnosis of SLD can only be made starting at the end of second grade.

Exclusion criteria for participation in the study were the following: motor or cognitive impairment and children whose parents/guardians did not sign the consent form. As a result, 10 non-native Italian speakers, one student with intellectual disability, and three students whose parents/guardians did not sign the consent form were excluded from the final sample of 239 children.

Based on standardized scores obtained in the text reading task (Cornoldi & Colpo, 1998) and/or the writing task (Sartori et al., 1995) at T2, children were divided into two groups: “Poor” and “typically developing” (TD) (see Statistical Analysis below). The “Poor” group included children who scored lower than −1.5 SD from the expected value for their age and level of schooling in either or both tasks, resulting in 57 children in the “Poor” group. The “TD” group consisted of 172 children who did not meet this criterion.

Procedures

Children's reading-related predictors and the development of their reading skills were measured at two distinct time points (see Table 1). The first measurement (T1) occurred during the first half of the first grade, specifically in early January, after 3 months of formal schooling. During this period, all children were classified as emergent readers. The second measurement (T2) took place at the end of the academic year in June, by which time the children had transitioned to beginning readers. This allowed us to assess their reading abilities after receiving additional formal instruction and practice throughout the school year. Trained clinical psychologists administered the tests.

OPEN IN VIEWER

Table 1.

Study Design and Assessments.

Time point	Timing	Assessment focus	Tests administered
T1	January	Cognitive and reading predictors	CPM—Colored Progressive Matrices (Raven, 2003) TCM—Test delle Campanelle Modificato [Modified Bells Test] (Biancardi & Stoppa, 1997) VMI—Developmental Test of Visual-Motor Integration (Beery et al., 2010) PRCR-2—Prove di prerequisito per la Diagnosi delle Difficoltà di Lettura e di Scrittura [Tests of prerequisites for the diagnosis of difficulties in reading and writing] (Cornoldi et al., 2009)
T2	June	Literacy skills	Text Reading—Prove di lettura MT per la scuola elementare-2 [Reading tests for primary school] (Cornoldi & Colpo, 1998) Word Writing—DDE-2—Batteria per la valutazione della dislessia e disortografia evolutiva [Battery for the assessment of developmental reading and spelling disorders] (Sartori et al., 1995)

Most of the tests were administered individually, with the exception of the fluid intelligence, text comprehension, and writing (spelling) tasks, which were administered collectively. Tasks were administered in a pseudorandom manner (IQ, comprehension, and writing abilities were always assessed before the individual testing began). The individually administered tests were conducted in a 60-min session in a quiet room at the children's school, with breaks provided to avoid fatigue and ensure high-quality performance.

Tools

The assessment protocol included standardized cognitive-linguistic tests to evaluate IQ, literacy skills, visuo-spatial attention, and visual-motor integration skills (see Table 1 for an overview). Raw data from all tasks were converted to standardized scores (z-scores or standard scores) according to age and/or schooling, except for phonological awareness, which was expressed in percentiles.

A short description of the tests used at each time point is provided below.

T1

T2

At the second assessment time point, children’ literacy skills were evaluated using two different tasks:

Statistical Analysis

Statistical analysis was conducted using the SPSS (Statistical Package for Social Sciences; IBM), version 16.0. Based on the standardized scores obtained in the text reading task (Cornoldi & Colpo, 1998) and/or the writing task (Sartori et al., 1995) at T2 (i.e., at the end of the first year of primary school), the children were divided into two subgroups. The first group, named “Poor,” included children who scored lower than −1.5 SD from the expected value for their age and level of schooling in either or both tasks (n = 57). Children who did not meet this criterion were assigned to the “typically developing” (TD) group (n = 172).

Descriptive statistics of the participants’ background were presented in Table 2. The sample was adequately homogeneous in terms of six (112 males and 117 females) and age. Specifically, 25 females and 32 males were included in the “Poor” group, while 92 females and 80 males were included in the TD group. There were no significant differences between the two groups in terms of age (p = .89) and fluid intelligence (age: p = .89; IQ: p = .16). However, the literacy performance of the two groups differed significantly in both the text reading task and the writing test.

OPEN IN VIEWER

Table 2.

Descriptive Statistics of the Two Groups.

	“Poor”		TD
	M	SD	M	SD	p	g
Sex (F–M)	25–32		92–80		.32a	0.30
Age (years)	6.5	0.32	6.45	0.47	.89b	0.11
IQ (standard scores)	89.15	15.24	96.05	19.33	.16b	0.37
Text reading (z-scores)	−2.17	0.52	−0.04	0.18	.001b*	7.05
Word writing (z-scores)	−1.98	0.73	0.01	0.44	.02b*	3.77

Note. Means (SD) for the two groups (“Poor”: n = 57; TD: n = 172) and effect size (Hedges’ g) of the differences between groups. Children who obtained a score in the reading or writing tasks lower than −1.5 SD compared to the expected value for the same age and level of schooling were included in the “Poor” group. Children who did not meet the criterion were assigned to the TD group.

^a

χ²-score, ^b F-score.

*Significance at α = .05.

The following analyses were conducted to further explore the cognitive deficits at T1 in future poor readers or spellers and to predict reading abilities at T2 based on reading-related cognitive skills measured at T1. These analyses include t-tests for unpaired samples and hierarchical regression analyses, respectively, to identify key predictors and understand the development of literacy skills.

Cognitive Deficits at T1 in Future Poor Readers or Spellers

Using several t-tests for unpaired samples, differences in the performance on the tests administered at T1 (i.e., literacy predictors) between “Poor” and TD children were analyzed (see Table 3). The standardized scores (z-scores) were corrected according to Bonferroni's method for multiple comparisons (significance at <.05).

OPEN IN VIEWER

Table 3.

Performance at T1 for “Poor” and TD Children.

		“Poor”		TD
		M	SD	M	SD	p	g
PRCR-2	Semi-circlesa	−0.64	2.13	0.09	2.15	.31	0.34
	Letter recognitiona	−2.01	1.15	−0.87	1.95	.45	0.63
	Nonword repetitiona	−1.87	0.81	−0.01	0.93	.04*	2.06
	Phonetic segmentationa	−0.42	0.79	0.98	0.45	.04*	2.52
	Naming of intertwined objects—Accuracya	−1.04	0.55	−0.89	0.62	.58	0.25
	Naming of intertwined objects—Speeda	−5.15	1.58	−0.44	1.05	.03*	3.91
	Naming of pointed objects—Accuracya	−1.47	0.9	−0.05	0.45	.03*	2.39
	Naming of pointed objects—Speeda	−0.73	1.16	−0.25	1.12	.09	0.42
	Visual search of two lettersa	−0.15	1.34	0.05	2.01	.91	0.11
	Visual search of a sequence of lettersa	−1.04	1.07	1.01	1.62	.89	0.69
TCM	Fasta	−1.17	1.14	−0.01	1.25	.14	0.94
	Slowa	−1.88	1.52	−0.11	1.26	.07	1.33
VMI	VMIb	94.51	12.5	106.02	19.15	.97	0.64
	CMb	119	14.02	117.35	17.03	1.25	0.10
	PVb	104.1	13.52	118.02	18.97	1.14	0.78

Note. Performance obtained by “Poor” and TD children in the tests performed at T1. Performance is expressed as standardized scores based on normative data: ^a z-score (M = 0; SD = ±1); ^b standard score (M = 100; SD = ±15). Hedges’ g and Bonferroni corrected p values (*significance at <.05) between groups.

The results highlighted which tests played a predominant role in distinguishing between “Poor” and TD readers/spellers. Specifically, the “Poor” group performed significantly lower than the TD group in the following subtests of PRCR-2: Phonetic segmentation (“Poor” M = −0.42, SD = 0.79; TD = 0.98, SD = 0.45, p = .04); Nonword repetition (“Poor” M = −1.87, SD = 0.81; TD = −0.01, SD = 0.93; p = .04); the Speed score of the Naming of intertwined objects (“Poor” M = −5.15, SD = 1.58; TD = −0.44, SD = 1.05, p = .03); the Accuracy score of the Naming of pointed objects (“Poor” M = −1.47, SD = 0.9; TD = −0.05, SD = 0.45, p = .03). In addition, there was a trend in the score for the speed of this test (“Poor” M = −0.73, SD = 1.16; TD = −0.25, SD = 1.12, p = .09). Indeed, the nonword repetition and the naming tasks require higher attention load (i.e., attentional control) than the phonetic segmentation task as they encompass the processes involved in active memory (Nonword repetition) and shifting between selective and divided attention (Naming).

Furthermore, although there was no statistically significant difference between the two groups, the results revealed a trend toward significance in the slow score of the barrage task (TCM_accuracy) (“Poor” M = −1.88, SD = 0.52; TD = −0.11, SD = 1.26, p = .07).

Predicting Reading at T2 From Reading-Related Cognitive Skills at T1

Since we demonstrated that “Poor” children at T1 already showed deficits in attentional control and phonological awareness skills, we further explored the possible link between children's performance in the cognitive tasks administered at T1 (i.e., mid-first year) and the literacy skills at T2 (i.e., the end of the first grade) across our entire sample of children (N = 239), regardless of our a priori group classification of being at risk of a reading/spelling disorder.

Two separate hierarchical regression analyses were performed of each dependent variable (text reading and word writing) to investigate whether cognitive skills measured at T1 influenced later reading and spelling performances.

In the first step, age and IQ were included to control for possible confounding effects. In the second step, Nonword repetition and Phonetic segmentation were added. These two subtests of the PRCR-2 were introduced with the aim to detect the contribution to later literacy performance of the phonological awareness skills. The Nonword repetition task is also a short-term memory measure. In the third step, Naming of intertwined objects and Naming of pointed object were added to assess the effects of rapid lexical retrieval, processing speed, and the ability to separate target objects from distractors in challenging perceptual situations (i.e., overlapping images of objects).

The two hierarchical regression analyses, using the scores of the PRCR-2 as the independent variables and text reading and word writing as the dependent variables, led to the following results (see Table 4). For text reading, none of the demographic variables significantly predicted reading performance at T2, F(2, 236) = 2.011, p = .471, r²= .009. This indicates that controlling for gender and IQ did not alter the relationships between the variables of interest. Phonological awareness skills at T1 uniquely explained 9% of the variance in text reading, F(4, 234) = 4.754, p < .001, ΔR² = .101. The change in R² was significant, F(1, 234) = 15.86, p < .001. Specifically, performance on the Nonword repetition subtest made a greater contribution to the prediction (p = .001) than the Phonetic segmentation task (p = .038). Adding the scores on Naming of intertwined objects and Naming of pointed objects to the regression model explained an additional 25% of the variance, F(6, 232) = 9.573, p < .001, ΔR² = .149; this change in R² was significant, F(2, 232) = 12.471, p < .01. Specifically, both subtests significantly contributed to the model (p = .045 and p < .001, respectively).

OPEN IN VIEWER

Table 4.

Hierarchical Regression Analysis.

	Model 1		Model 2		Model 3
	β	p	β	p	β	p
Text reading
Age	−.166	.038	−.131	.074	−.111	.132
IQ	.008	.922	.012	.879	−.004	.952
Nonword repetition			.321	.000	.318	.001
Phonetic segmentation			.215	.011	.166	.038
Naming of intertwined objects					.178	.045
Naming of pointed objects					.344	.000
DF model	236		234		232
F-value ANOVA (vs. null model)	2.05		4.75*		9.57**
R2 (vs. null model)	.009		.123		.248
F-value ANOVA (vs. previous model)	2.28		15.86***		12.47***
ΔR2 (vs. previous model)	.008		.101		.149
Word reading
Age	−.055	.515	−.098	.145	−.080	.221
IQ	−.031	.716	−.022	.733	−.014	.830
Nonword repetition			.177	.008	.170	.010
Phonetic segmentation			.619	.000	.667	.006
Naming of intertwined objects					.109	.135
Naming of pointed objects					.121	.009
DF model	236		234		232
F-value ANOVA (vs. null model)	.66		6.57**		16.89***
R2 (vs. null model)	.013		.114		.163
F-value ANOVA (vs. previous model)	.66		17.57**		4.91**
ΔR2 (vs. previous model)	.013		.272		.035

Note. Summary of three steps, fixed entry, hierarchical regression analysis (on the entire sample N = 239) for variables predicting the performance in text reading and word writing at T2. Model 1 controls for participants’ age and IQ; Model 2 adds performance in tasks assessing phonological awareness at T1 (i.e., nonword repetition and phonetic segmentation); Model 3 adds performance in naming/speed of processing (i.e., naming of intertwined and naming of pointed objects).

*p < .05. **p < .01. ***p < .001.

For word writing, the regression analyses showed that IQ and age, introduced as control variables in the first step, accounted for a nonsignificant proportion of the explained variance, F(2, 236) = 0.661, p = .851, r²= .009. Phonological awareness skills at T1 explained 11% of variance, F(4, 234) = 6.572, p < .001, ΔR² = .272, and this model provided a better fit of the data, F(1, 234) = 17.574, p < .001. In particular, both scores contributed equally to the prediction (Phonetic segmentation: p = .006; Nonword repetition: p = .010). Finally, the addition to the regression model of the scores in the tests assessing naming and attentional capacities explained an additional 16% of variance, F(6, 232) = 16.892, p < .001, ΔR² = .035. This change in R² was significant, F(2, 232) = 4.92, p < .01. Important, only performance on the Naming of pointed objects test made a significant contribution to the prediction (p = .009), whereas performance on Naming of intertwined objects did not (p = .135).

These results indicate that in novice readers, the ability to read and write efficiently depends on the ability to manipulate sounds of the language, rapidly access the lexicon, and flexibly allocate attention (i.e., processing speed) that the child exhibited one year earlier.

Limitations

The following limitations of the current study bear mentioning. First, the assessment battery administered at T2 was limited to two tasks (i.e., text reading and word writing). A more in-depth characterization of the reading and spelling profile of the participants would require other standardized tests. However, our assumption was that text reading should be considered the most ecologically valid task, since it not only requires word-level decoding and lexical strategies but also semantic and syntactic processes.

Second, a major limitation of the study was that cognitive predictors were assessed only once at study onset. This design choice precluded the identification of potential interactive relations between cognitive skills and reading development over time. Longitudinal data would provide a more dynamic understanding of how these predictors influence literacy acquisition.

Finally, the generalizability of the results is subject to certain limitations since our sample comprised only Italian-speaking first graders. Therefore, our findings can be generalized specifically to the language under investigation and, possibly, to other languages with similar consistent orthography. Although we hypothesize that our findings may be generalized to text reading in different orthographies, the influence of each predictor would likely vary with the transparency of the alphabetic system.