Faculty of Social and Behavioural Sciences
Graduate School of Child Development and Education
The Nature of Visual Attention Span and its
Utility as a Clinical Marker for Dyslexia
Research Master Child Development and Education Research Master Thesis
Student: Marieke Majoor (0470848)
Supervisors: M. van den Boer and E.H. de Bree External reviewers: P.F. de Jong and J. Tijms July 2017
Preface
This thesis consists of two studies. Regarding the first study, I recruited and tested the participants with dyslexia. The participants of the control group were recruited by my supervisors and tested by student assistants. Regarding the second part of the study, both the recruitment and testing of the participants was done by me and by child psychologists working at IWAL. The analyses of both studies were performed by me.
I would like to thank IWAL, the student assistants, and most of all my supervisors for their contribution to this thesis.
Abstract
Visual attention span (VAS) has been found to be a core skill determining reading and spelling performance, independent from phonological awareness and rapid automatized naming. However, it is not yet clear whether the VAS task measures purely visual skills or also involves verbal decoding skills, and whether VAS contributes to the diagnostic process of dyslexia. In Study 1, the nature of VAS was investigated. Participants were 22 children with dyslexia and 18 control group children. Three conditions of the VAS task with different levels of reliance on phonological decoding were administered. It was examined whether the children scored differently on the conditions and whether the scores differed between the groups in relation to the three tasks. Results showed that the participants scored differently on the tasks, providing evidence that VAS does not reflect merely visual skills, but verbal
decoding skills as well. In Study 2 the utility of VAS as a clinical marker of dyslexia was examined, as the existing cognitive measures are unable to explain all underlying causes of dyslexia. In 60 children with persistent reading and spelling problems, the relation of VAS with literacy and phonological awareness, rapid automatized naming and grapheme-phoneme connections were studied. Results indicated that the contribution of VAS to dyslexia is relatively equal to the contribution of these cognitive skills, highlighting the need for additional research into the underlying causes of dyslexia.
Keywords: visual attention span, developmental dyslexia, multiple deficit model,
phonological awareness, rapid automatized naming, grapheme-phoneme connections
The nature of visual attention span and its utility as a clinical marker for dyslexia
Developmental dyslexia is a learning disorder that is characterised by severe and
persistent difficulties in reading and/or spelling at the word level (Snowling, 2001). Research into development dyslexia has long focused on finding one single cognitive deficit that could explain all behavioural symptoms of the disorder, resulting in numerous theories regarding the causes of dyslexia. For several decades, the consensus has been that an underlying cognitive cause of dyslexia is a deficit in phonological processing (Di Filippo, Zoccolotti, & Ziegler, 2008; Frith, 1997; Peterson & Pennington, 2015; Snowling, 2001; Vellutino,
Fletcher, Snowling, & Scanlon, 2004; Ziegler & Goswami, 2005). According to the
phonological theory of dyslexia, problems with identifying and manipulating speech sounds result in poorer letter-sound correspondences, which in turn result in difficulties with
decoding unknown words and the development of specific orthographic knowledge (Peterson & Pennington, 2015; Vellutino et al., 2004).
Although theorists supporting the phonological theory agree on the core role of
phonology, they have different perspectives about the nature of the phonological deficit. As a result, several cognitive skills have been studied as part of the phonological theory, such as phonological awareness (PA), rapid automatized naming (RAN) and grapheme-phoneme connections. PA refers to the ability to identify and manipulate phonemes (speech sounds) within words, while RAN refers to the ability to rapidly name familiar objects such as letters, numbers, colours or pictures (Denckla & Rudel, 1976, Wolf, Bowers, & Biddle, 2000). Both PA and RAN have been found to be predictors of reading performance in both typically developing children as children with dyslexia in many languages (e.g., Bradley & Bryant, 1983; de Jong & van der Leij, 1999; Kirby, Georgiou, Martinussen, & Parrila, 2010; Landerl & Wimmer, 2008; Schatschneider, Fletcher, Francis, Carlson, & Foorman, 2004; van den Boer, van Bergen, & de Jong, 2015; Wolf & Bowers, 1999; Ziegler et al., 2010). In addition, PA has been found to be related to spelling performance (e.g. Landerl & Wimmer, 2008; van
den Boer et al., 2015; Verhagen, Aarnoutse, & van Leeuwe, 2010). Findings regarding RAN as a predictor of spelling performance are less straightforward, as some studies have found an effect of RAN on spelling (Verhagen et al., 2010), while in other studies, no effect was found (Landerl & Wimmer, 2008; van den Boer et al., 2015). Nevertheless, the contribution of PA and RAN in dyslexia is relatively uncontested.
The role of grapheme-phoneme connections is more debated. In order to develop reading skills, the association between letters and sounds need to be automatized and
integrated into audio-visual units, in such a way that an individual is able to instantly connect a grapheme to a phoneme (Blau et al., 2010; Ehri, 2005). Evidence for the importance of strong grapheme-phoneme connections in reading has mostly been found in brain studies, which have shown that dyslexics do have weaker grapheme-phoneme connections than average readers (Blau et al, 2010; Blau, van Atteveldt, Ekkebus, Goebel, & Blomert, 2009). Finding from behavioural research is less conclusive. For instance, Blomert & Vaessen (2009) found that, although performance on the letter-sound integration task (a task used to measure both accuracy and speed of grapheme-phoneme connections) was found to be related to reading and spelling ability in children from 8 to 11 years old, grapheme-phoneme
connections did not seem to explain variance in reading ability after controlling for PA and RAN. In addition, no differences in accuracy were found between typically developing children and children with dyslexia after grade 2 (Blomert & Vaessen, 2009). However, differences in speed between the two groups remained. In another study, grapheme-phoneme connections were found to predict reading performance in beginning readers and spelling performance in both beginning and more experienced readers. In all, the findings show that the role of grapheme-phoneme connections in predicting literacy skills is not yet clear, so further research is necessary to determine its role in dyslexia.
Although the phonological theory of dyslexia is widely acknowledged, the notion of a phonological deficit as the single underlying cause of dyslexia has been refuted. Studies have
found PA and RAN to independently contribute to reading ability, indicating that children with dyslexia can suffer from either a deficit in PA or RAN (Morris et al., 1998; Wolf & Bowers, 1999). Hence, PA and RAN are two independent risk factors for dyslexia. Moreover, there are several shortcoming to a single deficit model of dyslexia in general. First, on the one hand, several studies have shown that not all individuals with poor phonological abilities have dyslexia (e.g. Bishop, McDonald, Bird, & Hayiou-Thomas, 2009; Snowling, 2008; Snowling et al., 2003), while on the other hand, some individuals with dyslexia have well developed phonological abilities (e.g. Pennington et al., 2012; Valdois et al., 2011). Second, the different behavioural symptoms observed in individuals with dyslexia could not be explained by one cognitive factor (Ramus & Ahissar, 2012). Third, the phenomenon of comorbidity cannot be explained by a single deficit model, as some disorders co-occur more often than expected by chance (see also Pennington, 2006). This had led researchers to adopt a multifactorial view of dyslexia. According to the multiple deficit model proposed by Pennington (2006),
developmental dyslexia is a result of the interaction of multiple genetic and environmental risk factors. These aetiological factors influence the development of neurological functions and cognitive processes, which in turn result in the behavioural symptoms of developmental dyslexia. Thus, the model implies that dyslexia cannot be caused by a single aetiological or cognitive risk factor.
Due to the complex nature of dyslexia, the precise cause of the disorder is yet
unknown. Despite their link with dyslexia, PA, RAN and the ability to associate graphemes and phonemes are insufficient to explain all cases of dyslexia. The three cognitive skills combined only explain approximately 40% of the variance in reading and spelling (Vaessen & Blomert, 2013), indicating that other factors influence reading and spelling performance as well. Besides phonological theories, visual theories about the underlying causes of dyslexia have been postulated (e.g. Livingstone, Rosen, Drislane, & Galaburda, 1991; Stein & Walsh, 1997). A visual theory that has gained interest over the last decade is the visual attentional
span (VAS) deficit hypothesis (Bosse et al., 2007; Bosse & Valdois, 2009; Valdois et al., 2004). The VAS refers to the number of distinct visual elements (letters or numbers) that can be processed simultaneously in one glance (Valdois, Lassus-Sangosse & Lobier, 2012) and is assessed using a task in which 5-letter consonant strings are presented for a short duration. Children are asked to name as many letters as possible. The outcome measure is the number of letters that can be reproduced.
According to the VAS hypothesis, the poorer performance of people with dyslexia on the VAS task is due to a smaller visual attentional window, referring to the amount of
information extracted from the orthographic input. The multiple trace memory model (Ans, Carbonnel, & Valdois, 1998; Valdois et al., 2004), in which the VAS hypothesis is grounded, states that reading relies on two types of reading procedures. In the global procedure, the visual attention window extends over all letters in a word, enabling the processing of words as a whole. In the analytic procedure, visual attention narrows down to process smaller
orthographic units, such as letters, letter clusters or syllables. It is hypothesized that, due to a smaller visual attentional window, people with dyslexia are unable to spread their attention across letter strings, which makes it impossible to process words in the global mode. Therefore, they more often have to rely on the analytic procedure, which is slower than the global procedure.
In line with the multiple deficit theory of dyslexia, the VAS deficit hypothesis does not suggest VAS to be a replacement for phonological deficits, but it emphasizes the
contribution of a visual attentional deficit to literacy problems in at least a part of the dyslexic population (Valdois et al., 2003). This is supported by the study of Bosse, Tainturier, and Valdois (2007), who showed that PA and VAS deficits contributed independently to developmental dyslexia, as a majority of English and French speaking dyslexic children exhibited either a selective phonological or VAS cognitive deficit. Moreover, VAS is found to be a predictor of individual differences in reading ability independent of the influence of PA
and RAN in both normally developing children and children with dyslexia (Bosse et al., 2007; Bosse & Valdois, 2009; van den Boer et al., 2015).
Despite these findings supporting the VAS hypothesis, the hypothesis is criticized as well, as it is yet unclear what the VAS task measures precisely. Some researchers have argued against the notion of the VAS task measuring purely visual skills. Hawelka and Wimmer (2008) found that readers with dyslexia did not perform differently than non-impaired readers when involvement of verbal processes was avoided. Children were presented strings with five letters or pseudoletters, but instead of naming as many letters as possible, they were asked to indicate whether a predefined target was shown. As a result, they argued that reading
problems in individuals with dyslexia could not be traced to inadequate visual processing of letter strings. Further support against the notion of a purely visual nature of VAS was found by Ziegler et al. (2010). Not only did they avoid involvement of oral naming, they also administered three different versions of the VAS task, in which they presented either five letters, digits, or symbols. They argued that that letters and digits map onto phonological codes, but symbols do not. Results showed that children with dyslexia performed significantly worse than normally developing children on a two-alternative forced choice task with letter and digit strings, but not on a similar task composed of symbol strings, indicating more important deficits for verbal material than for nonverbal material if oral naming is avoided. Alternatively, Ziegler et al. (2010) suggest that the elements in the VAS task are verbally decoded and that poorer performance on the task may therefore be due to impaired symbol-sound mapping. Thus, it might be that the VAS task measures the ability to generate verbal codes rather than purely visual processing.
However, the debate still persists, as Lobier, Zoubrinetzky, and Valdois (2012) found evidence for a visual nature of VAS. In 109 typically developing children and 14 dyslexic children, they administered several version of the VAS task: the original VAS task and five categorization tasks, which were both verbal (letters and digits) and non-verbal (shapes,
pseudoletters and Japanese characters). In the categorization tasks, children were shortly presented multiple elements of a target category and a distracter category, and were asked to count the number of elements belonging to the target category. The results showed that children with dyslexia performed worse than the typically developing children on all VAS tasks. As children with dyslexia were unable to process multielement strings regardless of element type, Lobier et al. (2012) argued that children with dyslexia are impaired in visual processing and not phonological processing.
The issue of whether VAS has a purely visual nature might have implications for its utility as a clinical marker for dyslexia. If VAS involves purely visual skills, the VAS task might tap a completely different underlying cognitive skill than PA, RAN and grapheme-phoneme connections, indicating that it might be of additional value to these skills. However, if a VAS deficit reflects impaired symbol-sound mappings, the explanation as to why children with dyslexia score lower on a VAS task seems more in line with the phonological theory of dyslexia. This would indicate that VAS might measure similar skills as PA, RAN and grapheme-phoneme connections, suggesting that it might be redundant or that it might be a substitute of one of the other cognitive measures of dyslexia. Thus, in order to expand the knowledge about the underlying causes of dyslexia, the relation of VAS with PA, RAN and grapheme-phoneme connections needs to be studied.
Several studies have investigated the relationship between VAS and other cognitive skills. In general, no significant correlations or weak correlations were found between VAS and PA (Bosse et al., 2007; van den Boer et al., 2015), indicating that VAS and PA are independent abilities. The independence between VAS and PA was also demonstrated by the finding that VAS and PA made unique contributions to reading and spelling performance, as mentioned earlier. A similar pattern is found for the relationship between VAS and RAN. VAS and RAN seem to correlate only weakly, while both skills independently predict reading performance (van den Boer et al., 2015). These findings suggest that VAS might be a valuable
addition to these task in diagnosing dyslexia. However, Saksida et al. (2016) did not find VAS to explain additional variance after phonological skills were controlled for, indicating that VAS does not have additional value to the current cognitive skills in predicting reading problems. The relationship between the VAS task and the letter-sound integration task has, to the best of our knowledge, not yet been studied.
The purpose of the current study is twofold. As the issue concerning the nature of the VAS has not been resolved, the first goal is to assess whether the VAS task measures purely visual skills or if phonological skills are involved as well. In addition, since PA, RAN and grapheme-phoneme connections are unable to explain all variance in reading and spelling performance (Vaessen & Blomert, 2013), the second goal is to investigate the utility of VAS as a clinical marker for dyslexia in comparison to these established cognitive skills related to developmental dyslexia.
Study 1
In the first study we aimed to provide more insight in the nature of the VAS task by assessing whether performance of the task reflects purely visual processing skills or phonological skills as well. To answer the research question, two new versions of the task were created beside the original VAS task. As in the original VAS task, stimuli consisted of five-letter strings, but unlike the original VAS task, one of the letters was presented in a deviating font. In the first new task, children were asked to name which letter in the sequence was presented in a deviating font (VASletter task), while in the second new task, children were asked to name the position of the letter in a deviating font (VASposition task). Thus, contrary to previous studies (Hawelka & Wimmer, 2008; Lobier et al., 2012; Ziegler et al., 2010), oral report was not avoided. In addition, all three VAS tasks comprised letters instead of non-verbal elements.
Importantly, the three VAS tasks differed in the extent to which they called upon phonological decoding skills or visual processing skills. The original VAS task could be
considered the most difficult task in terms of phonology, as children were asked to name multiple letters. Conversely, the VASposition task could be considered the easiest task, as children only had to name the position of a letter, not the letter itself. If the VAS task measures purely visual skills, children would be expected to have similar scores on all three conditions of the task. However, if phonology plays a role in the VAS task, children would be expected to score differently on the tasks, with lower scores on the more difficult task in terms of phonology and higher scores on the easiest task. As the original VAS task is most likely to rely heavily on phonological skills, scores on this task would be expected to be the lowest. The highest scores would be expected to be found on the VASposition task, as phonological skills seem to play the slightest role in this task.
In addition, it was investigated how the three conditions of the VAS task relate to reading. A stronger relationship between reading and a condition of the VAS would indicate the use of similar cognitive processes, while a weaker relationship would indicate the use of different cognitive processes. Phonological decoding skills play an important part in reading, as children have to map visual or orthographic codes onto phonological codes. The three VAS tasks require different degrees of phonological processing, with strong involvement of
phonology in the original VAS and little involvement in the VASposition task. Therefore, a stronger relationship between reading and the original VAS compared to the VASposition would imply that phonological decoding is important in the VAS task. However, if the VAS task measures purely visual processing skills, the strength of the relationship between reading and the three VAS tasks would not be expected to be different.
Last, the performance on the three conditions of the VAS task were compared between children with dyslexia and typically developing children. As previous studies have shown that the VAS task can distinguish between children with dyslexia and typically developing
children, it was expected that children with dyslexia perform worse than typically developing children on the three conditions of the task. Furthermore, if VAS reflects purely visual
processing skills, it would be expected that the differences in scores between the two groups are equal for all three tasks. However, if VAS also involves phonological skills, the difference in scores between the children with dyslexia and the typically developing children would be expected to be larger on the more difficult original VAS and smaller for the easier
VASposition task.
Method Participants
A total of 40 children (25 boys (62.5%)) from third and fourth grade participated in this part of the study. Twenty-two children had an official diagnosis of dyslexia (12 boys) and 18 children (13 boys) were part of the control group. The two groups did not differ with regard to gender, χ² (1) = 1.32 p = .251. The mean age of both the dyslexic and the control group was 9 years and 7 months (SD = 6.92 months, range 105 – 128 months; SD = 6.55 months, range 103 – 127 months respectively). The mean age of the children in the control group was equivalent to that of the children with dyslexia, t(38) = .019, p = .985. With regard to word reading fluency, which was measured as the number of words children could
accurately read in one minute, the children with dyslexia (M = 35.45, SD = 9.66) scored significantly lower than the children of the control group (M = 60.83, SD = 14.72), t(38) = 6.55 p < .001.
Measures
Visual attention span: original VAS. The participants were administered a whole report VAS
task as designed by Valdois et al. (2003). The task was displayed on a computer screen using Microsoft PowerPoint. First, a plus sign was presented, after which five letters were shown simultaneously for 200 ms in 28-point Arial font (e.g. M H S D B). The letter strings were created from 10 consonants (B, D, F, H, L, M, P, R, S, T). Children were asked to name as many letters of the string as possible in the correct order. As two other versions of VAS were also presented (see below), the task was a reduced version of the VAS task designed by
Valdois et al. (2003). Originally, the task comprised 20 letter strings, but now only the first 10 letter strings were presented, each of the consonants was presented once in each position. To assess whether reducing the number of letter strings by half would influence the results, the scores of 10 participants of the second part of the study were analyzed. No significant difference was found between the scores on the first 10 letter strings (M = 29.80, SD = 7.55) and the last 10 letter strings (M = 29.90, SD = 7.70); t(9) = .080, p = .938. Scores comprised the number of correctly repeated letters, with a maximum score of 50. Both letter identity and order were taken into account. Thus, contrary to the original scoring of Valdois et al. (2003) in which the scores were based on letter identity only, in this study children also had to name the letters in the correct order.
Visual attention span: name the letter (VASletter). The VAS was also assessed with an
alternative VAS-task. Children were presented with 50 letters strings consisting of five letters, similar to the original VAS-task. The letter strings were constructed from the same 10
consonants used in the original task. However, one of the letters was presented in a different font (30-point Bodoni MT) (see Appendix). Each letter was presented in the deviating font once in each position. Children were asked to name only the deviating letter. Scores were based on the number of deviating letters named correctly (max = 50).
Visual attention span: name the position (VASposition). Another alternative task to assess the
VAS was the VASposition. The VASposition had a design similar to VASletter, as the same consonants were used and each letter was presented in a deviating font in each position once (see Appendix). However, the letters were presented in a different order. Children were asked to name the position of the deviating letter (position 1 through 5) instead of the letter itself. The score consisted of the number of times the children correctly named the target position (max = 50).
Word reading. Word reading skills were measured with the One-Minute Test (Eén Minuut
difficulty. Children were instructed to read the words as accurately and quickly as possible in one minute. The score consisted of the number of words read correctly. Test-retest reliability was good, with a range between r = .82 and r = .92.
Procedure
Children with a diagnosis of dyslexia were recruited from centers for learning disabilities where they were following treatment, while the participants of the control group were recruited from primary schools in the Netherlands. From March to July 2016, all children were tested individually in a session of about 30 minutes. As the data was part of a larger study, children were also administered two word reading tasks and a silent and oral text reading task. To assure that the order in which the tasks were presented did not influence the results, the tasks were administered in different orders. Children in the control group first made the original VAS-task, after which the other two VAS-tasks were presented in a counterbalanced order. For the dyslexic children, all three VAS-task where presented in a random order. Parents gave informed consent for their child to participate in the study. The study was approved by the ethics committee of the Faculty of Social and behavioral Sciences of the University of Amsterdam (id: 2016-CDE-6725. Title: Thesis 2 Visual Attention Span).
Results Data cleaning
Before running the analyses, data was checked for missing values and outliers. None of the variables had missing values. Three values were found to be extreme outliers (M±3sd) (Tabachnick & Fidell, 2013). To prevent data loss, the extreme values were replaced by the lowest score (within the normal range) of the corresponding variable. After replacing the values, both assumption of homogeneity of variance-covariance matrices and the assumption of sphericity were met.
Table 1 shows the mean scores of both the control group and the group of children with dyslexia on the three conditions of the VAS task. A repeated measures ANOVA was conducted with the variables group (dyslexia vs. no dyslexia) and VAS (original VAS, VASletter, VASposition) to assess whether children with dyslexia score differently than typically developing children on the three conditions of the VAS task. A significant main effect was found for group, F(1, 38) = 11.187, p = .002, partial ŋ² = .227. As expected, the dyslexic children performed worse than the control group, regardless of condition. The results also showed a main effect for VAS condition, F(2, 76) = 9.472, p < .001, partial ŋ² = .200, indicating that the participants scored differently on the three conditions. A Bonferroni post-hoc test showed a significant difference between the original VAS task and the VASposition task (p = .001). Children scored significantly higher on the VASposition task. In addition, a significant interaction effect between group and VAS task was found, F(2, 76) = 9.716, p < .001, partial ŋ² = .204 indicating that for the children with dyslexia, the scores on the three conditions of the VAS task showed a different pattern than for the control group (see Figure 1). As presented in Table 1, the largest difference between the two groups was found on the original VAS task, t(38) = -4.727, p <.001. The difference between the two groups on the VASletter task was also significant, t(38) = -2.109, p < .001. The difference in performance on the VASposition task was non-significant. Taken together, these results indicate that children with dyslexia performed worse than the typically developing children on the tasks that called upon phonological recoding skills, while their performances matched on the task in which phonology played a negligible role.
[Table 1]
Given the interaction between group and VAS condition, separate repeated measures ANOVA’s were conducted per group. Findings showed that, as expected, the control group children did not perform differently on the three conditions of the VAS tasks. However, a significant difference in scores on the three tasks was found for the children with dyslexia, F(1, 21) = 768.438, p < .001, partial ŋ² = .973. Results of the Bonferroni post hoc test indicated that the children with dyslexia scored significantly lower on the original VAS task than the VASletter task (p = .006) and the VASposition task (p < .001). Also, scores on the VASposition task were significantly higher than scores on the VASletter task (p =.048). Thus, scores on the conditions of the VAS task differed significantly for the children with dyslexia, with the lowest scores found on the task in which phonological skills played a predominant role, and the highest scores on the task with minor influence of phonological skills.
It should be noted that, although a comparison between the conditions of the VAS could be made because all three tasks had a maximum score of 50, the original VAS differed from the other two tasks in the way these scores were composed. The original VAS comprised 10 strings, while the VASletter and VASposition tasks consisted of 50 strings. Thus, as the study does not provide a perfectly fair test for comparison of the original VAS task to the other VAS tasks, results need to be interpreted with caution.
In the next step to determine whether performance on the VAS task relies more heavily upon phonological or visual skills, a correlation analysis was conducted on measures of word reading fluency and the three conditions of the VAS task for all participants (Table 2). Significant correlations were found between the three conditions of the VAS task. In addition, all three VAS tasks correlated significantly with word reading fluency. However, the strength of the correlations differed: The correlation between word reading fluency and the original VAS task was strong, moderate for VASletter and weak for VASposition.
[Table 2]
Discussion
The aim of the present study was to gain insight in the nature of the VAS task by assessing whether the VAS task involves purely visual skills or phonological decoding skills as well. To answer the research question, performances of children with dyslexia and
typically developing children on three conditions of the VAS task were compared. The three conditions of the VAS task differed in the extent to which they called upon phonological decoding skills. The outcomes revealed that the children with dyslexia scored lower than the typically developing children on the three conditions of the VAS task, indicating that the VAS task can distinguish between children with dyslexia and typically developing children. This is in line with previous findings by Bosse et al. (2007).
Regarding the three conditions of the VAS task, the weakest performance of the children with dyslexia was found on the original VAS task. As this task called most heavily upon phonological skills, these results support the notion that performance on the VAS task does not just reflect purely visual processing deficits, but also reflects phonological deficits. In addition, the strongest performance was found on the VASposition task, in which the influence of phonological skills was minimized. The finding that the VAS task not just have a purely visual nature was supported by the correlation analysis, in which the strongest
correlation was found between word reading skills and the original VAS task, while the weakest correlation was found between word reading skills and the VASposition task. Thus, the results of this study indicate that performance on the original VAS task at least partly reflects phonological skills instead of purely visual processing skills. This supports the findings of previous studies reporting the involvement of phonological processing skills in VAS (Hawelka & Wimmer, 2008; Ziegler et al., 2010).
According to the multiple deficit model of dyslexia, several cognitive risk factors can cause reading and spelling problems. Three current risk factors for dyslexia, PA, RAN, and grapheme-phoneme connections, are unable to explain all variance in reading and spelling performance (Vaessen & Blomert, 2013). Previous studies have indicated that VAS might be implicated in developmental dyslexia. Therefore, the main goal of the present study is to investigate the utility of VAS as a clinical marker for dyslexia. Knowledge about the possible role of VAS as an independent risk factor for developmental dyslexia contributes to the knowledge about underlying cognitive deficits of the disorder. The results, in turn, could influence the diagnostic procedure of developmental dyslexia.
As explained earlier, findings of the first part of the study might have implications for its use as a clinical marker. The first study showed that the VAS task cannot be considered a purely visual task, since phonological decoding skills were also found to be involved.
Phonology plays an important role in PA, RAN and grapheme-phoneme connections as well, indicating that VAS might be mutually related to these underlying cognitive skills. Earlier research has shown that the VAS task cannot be considered either a PA or a RAN task, as VAS contributes to reading performance independent of PA and RAN (Bosse & Valdois, 2009; van den Boer et al., 2015). The relationship with grapheme-phoneme connections, another cognitive skill related to reading performance, has, to the best of our knowledge, not yet been studied. In the letter-sound integration task, which is used to measure the quality of grapheme-phoneme connections, children are required to quickly and accurately associate graphemes to phonemes. Therefore, poor performance on the task gives an indication of poor symbol to sound mappings. A similar interpretation of VAS was given by Ziegler et al. (2010). The findings of our first study supported this notion of VAS reflecting impairments in symbol to sound mappings. It could thus be argued that the VAS task and the letter-sound integration task both measure an identical underlying skill.
The VAS task and the letter-sound integration task differ in important aspects as well. First, the VAS task measures the ability to process multiple elements, while in the letter-sound integration task only single elements are presented. Furthermore, the letter strings of the VAS task are only presented for a short amount of time, while the single letters of the letter-sound integration task remain available. There are, however, arguments that these differences could also be in favor of the hypothesis that the VAS task and the letter-sound integration task measure the same underlying cognitive skill. Since the letters in the VAS task are only presented for a short amount of time and children are asked to name more than just one letter, it could also be argued that strong grapheme-phoneme connections are essential for good performance (see also van den Boer et al., 2015). Therefore, an important aim of this study is to assess whether the visual attention span task may be a valuable addition to the existing tasks or whether it measures skills equivalent to the letter-sound integration task. If so, the visual attention span task might be redundant, or it might be an improvement to the letter-sound integration task.
However, the utility of VAS is not only measured by its value relative to that of PA, RAN and grapheme phoneme connections. In order of a cognitive skill to be useful in
diagnostics, the task measuring the underlying cognitive skill must be able to identify children with a specific disorder. Therefore, we will also assess the diagnostic accuracy of VAS in detecting children with word reading problems. This has, to the best of knowledge, not yet been investigated.
Method Participants
The sample comprised 60 participants (36 boys) with persistent literacy difficulties. The mean age of the participants was 9 years and 7 months (SD = 16.67 months, range 90 – 160 months). All participants had Dutch as their native language. All participants were referred to diagnostic and treatment centers for dyslexia. Full-scale IQ of the children ranged
from 74 to 135 (M = 100.67, SD = 13.02). All children attended mainstream primary education. The participants were from families with a relatively high socio-economic status (SES). Information on parental education level served as a measure for SES. Three groups were distinguished based on the highest educational level obtained by one of the parents. Parents who completed secondary education or lower were classified as having a low educational level (5%); those who completed intermediate vocational education an average educational level (23.3%), and those who completed tertiary education a high educational level (58.3%). No information was available on the parents’ educational level for the remaining 13.3% of students.
For the sensitivity and specificity analysis, the sample was extended with data from a control group. The control group comprised 638 students (313 boys (49.1%)) from second to fifth grade. The mean age of the control group was 9 years and 6 months (SD = 12.96 months, range 85 – 153 months).
Measures
The 3DM battery of tests (Blomert & Vaessen, 2009) was used to measure literacy (reading and spelling) and literacy-related abilities (PA, RAN, grapheme-phoneme
connections). In this individually administered test, a computer is used to register both accuracy and reaction time. In addition, the VAS and IQ were assessed.
Word reading. The task comprised three subtasks containing high-frequency words,
low-frequency words and pseudowords. There were 75 words for each subtask, displayed on five sheets with 15 items each. For each subtask, the difficulty of the words increased from monosyllabic words without consonant clusters to 3 or 4 syllabic words with consonant clusters. The children were asked to read the words aloud as quickly and accurately as possible for 30 seconds per level. The score for speed consisted of the total number of words read correctly on the three subtasks. For the accuracy measure, a proportion of correctly read words was calculated by dividing the number of words read correctly by the total number of
words read within the time limit. Test-retest reliability is r = .73 for accuracy and r = .95 for speed (Blomert & Vaessen, 2009). In the control group, word reading was assessed with the One Minute test as described in Study 1.
Spelling. In this task, a word was presented aurally and partly visually. The visual
presentation missed a letter or letter combination. For instance, auditorily presented stimulus ’boom’ (‘tree’) and visual stimulus ‘b__m’. The child was asked to choose the missing letter (combination) out of four options visually presented on screen by striking a key on a response box as fast as possible (e.g. ‘o’ ‘aa’ ‘a’ ‘oo’). The task consisted of 54 words which were either spelled phonetically (18 items) or contained Dutch spelling rules (36 words). Two scores were obtained: Accuracy (percentage of correct responses) and mean response time (seconds per item). Internal consistency was good: r = .80 for accuracy and r = .94 for response time (Blomert & Vaessen, 2009).
Phonological awareness. PA was assessed with a phoneme deletion task. Twenty-three
nonwords (e.g. ‘teuk’) were presented auditorily. The participants were asked to delete a phoneme that was within a consonant cluster or at the beginning or the end of a nonword (e.g. ‘teuk’ minus ‘k’). After deleting a phoneme, the nonwords never turned into a real word. The task comprised 23 items. Both accuracy (percentage of correct responses) and mean response time in seconds were measured. Internal consistency was good, r = .85 (Blomert & Vaessen, 2009).
Rapid naming. The rapid naming task consisted of two subtasks: letters and digits. The
participants were asked to name the items presented on the computer screen as fast and accurately as possible. Each subtask was administered twice and consisted of three rows of five stimuli. In each sheet, the position of the stimuli varied. As children rarely name the items incorrectly, incorrect responses are registered, but are not incorporated in the norms of the task. Thus, the score per subtask consisted only of the mean response time in seconds of
the two screens. Split-half reliability is r = .80 for letters and r = .83 for digits (Blomert & Vaessen, 2009).
Grapheme-phoneme connections. Both a letter-sound discrimination task (90 items) and a
letter-sound identification task (45 items) were used to measure integration of grapheme-phoneme connections. In the discrimination task, children were asked to match a grapheme-phoneme to one of four presented letters or letter combinations by striking the corresponding key on a response box (e.g. aurally presented ‘d’ and visually presented ‘b’ ‘d’ ‘p’ ‘t’). In the identification task, children were presented with both a phoneme and a grapheme and were asked to determine whether the sound and letter were congruent or incongruent (e.g. aurally presented ‘eu’ and visually presented ‘ui’). Again, answers were given by striking the corresponding key. Both accuracy (percentage of correct responses) and mean response time (seconds per item) were taken into account. The scores on discrimination and identification tasks were combined in a score reflecting the mean accuracy (percentage of correct responses) and a score reflecting the mean response time. Internal consistency was sufficient: letter-sound identification: r = .72 for accuracy and r = .90 for response time; letter-letter-sound
discrimination: r = .82 for accuracy and r = .96 for response time (Blomert & Vaessen, 2009).
In addition to the tasks of the 3DM, the following tasks tapping cognitive skills were administered.
Visual attention span. VAS was assessed with the VAS-task as described in Study 1. Whereas
in Study 1 the number of items was reduced by half, in Study 2 all 20 letter strings were presented, so all letters were presented twice in each letter position. Scores consisted of the number of letters the participants were able to repeat (max. = 100). Both the letter identity and the order in which the letters were repeated were taken into account.
IQ. The Dutch version of the third edition of the Wechsler Intelligence Scale for Children
(WISC-III-NL) was used to measure general cognitive ability in children from 6 to 16 years old (Wechsler, 2005). In this study, we used the Full-scale IQ. The IQ-score was based on 10 subtests, namely Information, Similarities, Vocabulary, Comprehension, Picture Completion, Picture Arrangement, Block design and Object Assembly. The reliability, construct validity and quality of the (Dutch) norms of the test were found to be satisfactory (Evers, van Vliet-Mulder, & Groot, 2000).
Procedure
Children were tested in two dyslexia centers at several different locations. They visited these centers twice in the period from March to September 2016. Each session took around two and a half hours. The VAS task was added to the original test battery used in a diagnostic assessment for dyslexia. The tasks were administered by a child psychologist working at the dyslexia centers. Parents gave passive consent for their child to participate in the study. The study was approved by the ethics committee of the Faculty of Social and behavioral Sciences of the University of Amsterdam (id: 2016-CDE-6725. Title: Thesis 2 Visual Attention Span).
Results Data cleaning
The initial dataset was checked for outliers. Values that were more than three standard deviations from the mean were examined (Tabachnick & Fidell, 2013). Six extreme outliers were found, all belonging to different tasks. The outliers were replaced by the closest values within the normal range of the corresponding variables. Furthermore, some participants had missing values on one or more of the variables. On most of the variables, less than 5% of the cases were missing. However, on the variable PA speed, 11 cases were missing. When a child gives incorrect answers on six or more items of the PA task, the average response time cannot be measured reliably. Therefore, no data was available on PA speed for children who did not
meet this requirement. As the data was not missing at random, missing values could not be imputed, χ² (29) = 73.434, p < .001.
Descriptive statistics
Table 3 shows the descriptive statistics of word reading (speed and accuracy), spelling recognition (speed and accuracy), grapheme-phoneme connections (speed and accuracy), PA (speed and accuracy), RAN (letters and digits) and VAS. As shown by the minimum and maximum scores, one or more children failed to give a correct response on PA accuracy, indicating that there might be a floor effect. However, the score distribution on PA accuracy was approximately normal. Furthermore, a possible ceiling effect was found on the accuracy measure of grapheme-phoneme connections, as one or more children gave a correct response on all items. The distributional properties, however, showed that grapheme-phoneme
accuracy was only slightly skewed.
[Table 3]
Identification of children with dyslexia using VAS
The first step in investigating the possible role of VAS as a clinical marker for dyslexia was to examine the diagnostic accuracy of the VAS task in detecting children with word reading problems. Estimates of sensitivity and specificity were calculated. Regarding VAS, no norms were available to differentiate between children with or without a VAS deficit. Therefore, threshold scores that could be used as a cutoff were calculated from data of 638 primary school children. A cut-off point of below the 10th percentile was of clinical interest, with children scoring at or below the cut-off being categorized as “having a deficit” and children scoring above the cut-off being defined as “not having a deficit”. The age range of the children was relatively wide, as the data comprised children from grade 2 to grade 5. Therefore, different threshold points were calculated for each grade. For word reading, the same cut-off point was selected.
Table 6 shows the sensitivity, specificity and overall accuracy of the VAS task in detecting word reading problems. The sensitivity of the VAS task was 43.4%, which means that the task correctly identified 43.4% of the children with word reading problems. Of the children with sufficient word reading skills, 91.3% were identified correctly. The VAS task failed to identify 56.6% of the children with word reading problems, while it falsely identified 8.7% of the children who did not score within the lowest 10% on word reading.
To determine the significance of these values, sensitivity and specificity were also calculated for PA, RAN and grapheme-phoneme connections. For each cognitive measure, two categories were formed based on the cut-off point of 10%. Children scoring at or below the cutoff point were classified as “having a deficit”, while children scoring above the cut-off point were defined as “not having a deficit”. Since the 3DM is used in the diagnostic
procedure of dyslexia, the published norms of this test were used to assign children to the groups (Blomert & Vaessen, 2009).
It should be noted that these analyses could only be performed within the group of children referred to diagnostic and treatment centers for dyslexia, as no data was available from a control group of typically developing children. Therefore, these results must be interpreted with some caution. The sensitivity of the original cognitive measures ranged from 31.0% (PA speed) to 60.0% (RAN digits), with an average sensitivity of 42.5%. This
carefully suggests that the sensitivity of the VAS task is relatively similar to the sensitivity of the other cognitive tasks.
[Table 4]
Relation of VAS with PA, RAN and grapheme-phoneme connections
In the next step, the relationship of VAS with literacy skills and PA, RAN and grapheme-phoneme connections were assessed. Due to the selective nature of the sample of
children with dyslexia, the variance of the dependent and independent variables was limited, while the age range of the participants was relatively wide. Therefore, age was controlled for in all analyses. Results show that VAS correlated significantly with word reading accuracy, word reading speed and spelling speed (see Table 4). With regard to the established
phonological measures, significant moderate and weak correlations were found between VAS and PA accuracy, and VAS and RAN digits, respectively. These correlations were in the expected directions. Contrary to expectations, VAS did not correlate significantly with measures of grapheme-phoneme connections, carefully suggesting that these tasks involved different cognitive processes.
With regard to PA, RAN and grapheme phoneme connections, significant correlations were found between the two measurements of each task. In general, most of the
measurements did not correlate significantly with each other. However, if significant correlations were found, they were weak to moderate and in the expected directions. Regarding the relationship between literacy skills and PA, RAN and grapheme-phoneme connections, only PA accuracy correlated weakly with all four literacy measures. Conversely, grapheme-phoneme accuracy did not correlate significantly with any of the literacy measures. As expected, cognitive measurement reflecting speed correlated predominantly with speed measures of literacy. Again, all correlations were in the expected directions.
[Table 5]
Predictors of reading and spelling
Hierarchical regression analyses were performed to assess the contribution of PA, RAN, grapheme-phoneme connections and VAS to reading and spelling skills (N = 57). To reduce the number of predictors, one measure of each of the established cognitive skills was selected. Since many values of the speed measure of PA were missing, it was decided to use
the accuracy measure. This was supported by the correlations with the dependent variables, as PA accuracy correlated significantly with all four dependent variables. Based on the
correlations between the two measures of grapheme-phoneme connections and reading and spelling skills, grapheme-phoneme speed was chosen as predictor variable. None of the dependent variables correlated significantly with grapheme-phoneme accuracy, while word reading speed and spelling speed were found to be significantly correlated with grapheme-phoneme speed. Regarding RAN, a new variable was created, reflecting the mean response time of RAN letters and RAN digits.
Due to the relatively wide age range of the sample, age was entered as a control variable in the first step of each analysis. PA accuracy, the combined score of RAN, speed of grapheme-phoneme connections and VAS were added in the second step. Results are
presented in Table 5. In all analyses, age accounted for a substantial amount of variance. PA, RAN, grapheme-phoneme connections and VAS accounted for additional explained variance in word reading accuracy (11.6%), word reading speed (18.1%) and spelling speed (18.7%). Contrary to expectations, the cognitive measures did not explain variance in spelling
accuracy, although the increase in explained variance almost reached significance (p = .05) Standardized beta coefficients show that, with regard to the cognitive measures, PA appeared to have the strongest effect on word reading accuracy. Furthermore, although the cognitive measures did not explain addition variance in spelling accuracy, an effect of PA on spelling accuracy was found (Table 5). None of the other cognitive measures were significant predictors of these outcome variables. RAN was found to be a significant predictor of word reading speed, while grapheme-phoneme connections was a significant predictor of spelling speed. Again, none of the other measures appeared to have a significant effect on word reading or spelling speed, although the effect of VAS on word reading speed almost reached significance (p = .05). Most of the effects were in the expected directions. However, the effect of RAN on word reading accuracy and the effect VAS on spelling accuracy were in the
opposite directions than they were expected to based on their correlation with the dependent variables. This indicates that a net suppression effect is present, which means that a variable increases the predictive power of another variable (Tabachnick & Fidell, 2013). The
suppressor variable for VAS was found to be PA accuracy, as the magnitude of the effect of VAS on spelling accuracy became much smaller after leaving PA accuracy out of the regression analysis. For RAN, the suppressor variable could not be identified. However, as the effect of VAS on spelling accuracy and the effect of RAN on word reading accuracy were not significant, the suppressor effects did not influence the results.
Thus, the analyses revealed that all three established cognitive measurements significantly contributed to either reading or spelling skills. While PA was found to be a predictor of the accuracy measures of both word reading and spelling, RAN and grapheme-phoneme were predictors of either word reading speed or spelling speed. The role of VAS in predicting reading and spelling performance was not significant, although the effect on word reading speed almost reached significance.
[Table 6]
Descriptive statistics of children scoring low on VAS
Last, we assessed how a VAS deficit relates to deficits in the PA, RAN and grapheme-phoneme connections by taking a closer look at the children scoring within the lowest 10% on each of the cognitive tasks. As shown in Table 7, the number of children scoring within the lowest 10% was highest on the VAS task. On most tasks, 40% to 50% of the children got unsatisfactory scores. This percentage was lower for grapheme-phoneme accuracy, where 33.3% of the children scored within the bottom 10%.
Next, we focused on the 30 children scoring low on VAS. Table 8 shows the number of established cognitive tasks on which these children score within the bottom 10%. The
majority of the children who scored low on VAS also scored within the lowest 10% on two of the original cognitive tasks related to literacy. Four children had satisfactory scores on all components, while also four children failed on all six components. No patterns can be
observed by looking at the tasks and combination of tasks on which the children scored within the lowest 10%. This is also found when looking at the total number of children with a VAS deficit who scored unsatisfactory on the established cognitive tasks. Regarding PA, 10
children (33.3%) scored low on accuracy and 12 children (40.0%) scored low on speed. Also, 12 children (40.0%) scoring low on VAS had unsatisfactory scores on RAN letters, while 18 children (60%) had unsatisfactory scores on RAN digits. Last, 14 children (46.7%) scored within the bottom 10% on grapheme-phoneme accuracy and 13 children (43.3%) scored within the lowest 10% on grapheme-phoneme speed.
[Table 7] [Table 8]
Discussion
In the second part of the study, we examined the utility of VAS as a clinical marker of dyslexia by investigating the diagnostic accuracy of VAS and its relation with PA, RAN and grapheme-phoneme connections. We were especially interested in the relationship between VAS and grapheme-phoneme connections, due to the seemingly similar processes required to successfully complete the tasks measuring these cognitive skills.
An important finding was that performance on the VAS task was not related to
performance on the letter-sound integration task, indicating that VAS and grapheme-phoneme connections reflect different underlying processes. In addition, in line with earlier research (Bosse & Valdois, 2009; van den Boer et al., 2015) we found no significant correlations or only weak to moderate correlations with PA and RAN, supporting the claim that the VAS task
also cannot be considered a PA or RAN task. The finding that a VAS deficit can occur independently from a deficit in PA, RAN or grapheme-phoneme connections was also supported by the finding that a VAS deficit often co-occurred with deficits in PA, RAN and grapheme phoneme connections, but that no pattern could be observed. Thus, deficits in VAS did not seem to coincide more often with deficits in either PA, RAN and grapheme-phoneme connections. If VAS would have significant overlap with one of the other cognitive measures, we would have expected to predominantly observe one specific combination of deficits. Taken together, these results show that VAS could not be a considered a replacement for one of the established cognitive measures, as all tasks seem to reflect different underlying
cognitive skills.
Two possibilities remained open, that of VAS being either redundant or VAS being a valuable addition to PA, RAN and grapheme-phoneme connections. In all, the results suggest that VAS can be considered a risk factor for dyslexia, but that the contribution of VAS to dyslexia is relatively equal to the contribution of the current cognitive skills. As expected, we found an association between VAS and word reading and spelling performance, especially word reading and spelling speed. However, VAS was not found to be a significant
explanatory factor for reading and spelling performance in an analysis that included PA, RAN and grapheme-phoneme connections. These results were in line with the findings of Saksida et al. (2016), but contradicted studies that showed that VAS explained additional variance on top of PA and RAN (Bosse et al., 2007, van den Boer et al., 2015). However, this differences could have been due to the selective nature of the sample.
Despite the nonsignificant contribution of VAS to reading and spelling performance, we found the diagnostic accuracy of VAS to be in line with the accuracy of PA, RAN and grapheme-phoneme connections. Since this has not yet been studied, this finding adds to the knowledge of VAS as a clinical marker for dyslexia. Furthermore, the number of children with a VAS deficit was relatively equal to the number of children with deficits in either one of
the current cognitive skills, indicating that a VAS deficit occurs as often as the other cognitive deficits. In all, these results contribute to the research on the causes of dyslexia by providing insight into the role of VAS as a possible risk factor for dyslexia in comparison to the original risk factors.
General discussion
The present study aimed to gain more insight in VAS in two ways. First, we investigated the nature of the VAS task by comparing the performance of children with dyslexia and controls on three different VAS tasks. The three VAS tasks varied in the extent to which they called upon phonological decoding skills or visual processing skills. Second, we studied the role of VAS as a possible clinical marker of dyslexia by examining its relation to literacy skills and its relationship with three current clinical markers of dyslexia: PA, RAN, and grapheme-phoneme connections.
We found evidence that phonological decoding plays a role in the VAS task, indicating that the VAS task does not seem to reflect merely visual processing skills. The finding that the VAS task can at least partially be considered a phonological task contradicts the VAS deficit hypothesis and other visual hypotheses stating that VAS has a merely visual nature. The VAS deficit hypothesis posits that individuals with dyslexia suffer from a
limitation in the number of elements that can be processed simultaneously due to a narrowed attentional window (Valdois et al., 2004). Because of the smaller attentional window, children with dyslexia are unable to spread their attention across letter strings. From the VAS deficit theory, children would be expected to have similar scores on VAS tasks that involve different degrees of phonological processing. However, as children are asked to verbally decode letters, it is suggested that phonological skills influence children’s performance on the task (Ziegler et al., 2010). Our findings disconfirmed the prediction by the VAS hypothesis, as we found children to score lowest on a task that relied most heavily on phonological processing, and highest on task in which verbal decoding was minimized. These results seem to indicate that
poor performance on the VAS task also reflects the inability to verbally decode letters, instead of a visual problem with parallel processing of visual elements.
Regarding the ongoing debate about the nature of the VAS task, our findings support earlier studies claiming that the VAS task cannot be considered a purely visual task, but that poor performance on the task reflects impaired symbol to sound mappings (Hawelka & Wimmer, 2008; Ziegler et al., 2010). Hawelka and Wimmer (2008) found adults with
dyslexia not to perform differently than non-impaired individuals when involvement of verbal decoding was avoided. In addition, Ziegler et al. (2010) showed that children with dyslexia perform worse than typically developing children on VAS tasks presenting verbal material (letters and digits), but not on VAS tasks presenting nonverbal material (symbols) that do not map to phonological codes. Our study adds to these findings in showing that the performance of children on a VAS task is influenced by the degree of phonological decoding that is required. As the letter report task is still the standard way of testing VAS, a strong aspect of our study is that all conditions of the task comprised five-letter strings, instead of other stimuli. Thus, support for the involvement of verbal decoding has been found in both adults and children with dyslexia, and in several conditions.
Despite the fact that phonological decoding skills are involved in VAS, we did not find VAS to measure similar underlying skills as PA, RAN and grapheme-phoneme
connections, in which phonology also plays an important role. In line with earlier research, we found that VAS cannot be considered a PA or a RAN task (Bosse & Valdois, 2009; van den Boer et al., 2015), as we found relatively small correlations between these tasks in children with persistent reading and spelling difficulties. More importantly, the absence of a correlation between VAS on the one hand and grapheme-phoneme accuracy and speed on the other hand indicated that poorer performance on the VAS task does not seem to be related to weaker grapheme-phoneme connections. Thus, VAS also cannot be considered a letter-sound integration task, as both tasks seem to involve different underlying processes.
An important difference between the VAS task and the letter-sound integration task is the number of elements that are presented. While the VAS task comprises five-letter strings, only one letter is presented in the letter-sound integration task. Moreover, the stimuli in the letter-sound integration task remain available, while the letters in the VAS task are only presented for a short amount of time. Thus, children do not only have less time to process the orthographic units, but they also have to process multiple elements within that limited time frame. Therefore, an alternative interpretation of VAS might be that it reflects the ability to activate verbal codes in parallel.
Several findings support this alternative interpretation of VAS. First, Bosse et al. (2007) did not find differences between children with dyslexia and control group children in their ability to identify briefly presented single letters. In addition, Lassus-Sangosse,
N’guyen-Morel, and Valdois (2008) showed that children with dyslexia did not perform worse than typically developing children on a task in which five letters were presented
sequentially for a short amount of time. These findings suggest that children with dyslexia are able to verbally decode orthographic units when they are presented briefly one after another. Moreover, the same children did perform worse than their typically developing counterparts when they were asked to name simultaneously presented letters. Thus, despite their ability to process single letters, they were unable to process orthographic units simultaneously. Both Bosse et al. (2007) and Lassus-Sangosse et al. (2008) interpreted these findings in favour of the VAS hypothesis, claiming that the different findings on the two tasks stemmed from a reduction in the number of visual elements that can be visually processed in parallel.
However, in light of the findings of this study that VAS cannot be considered a purely visual skill, these results support the notion that a VAS deficit reflects the inability to decode letters in parallel.
Another goal of this study was to determine the utility of VAS as a clinical marker for dyslexia. After showing that VAS cannot be considered a PA, RAN or letter-sound
integration task, two possibilities remained open: that of VAS being either redundant or VAS being a valuable addition to PA, RAN and grapheme-phoneme connections. We found that VAS does seem to contribute to developmental dyslexia, but that its contribution is rather equal to the contribution of PA, RAN and grapheme-phoneme connections. Several findings supported this claim. First, the number of children with a deficit in VAS was relatively equal to the number of children with a deficit in either one or more of the other cognitive measures, indicating that the prevalence of a VAS deficit is similar to the prevalence of a deficit in PA, RAN and grapheme-phoneme connections in children with persistent reading and spelling problems. In our study, we found 50% of the children to be hampered by a VAS deficit. These number are in line with numbers found by Bosse et al. (2007), who reported that 59% of the French children and 41.5% of the English children with dyslexia exhibited a VAS deficit or a combined phonological and VAS deficit. Contrary to these findings, Saksida et al. (2016) found that deficits in VAS affected only 28.1% of the children with dyslexia. However, as explained by Saksida et al., (2016), the relatively low percentage of children with a VAS deficit found in their study might be due to their sample of severely impaired readers, which resulted in a high prevalence of the phonological deficit. In all, these findings suggest that a VAS deficit is present in a substantial number of children, supporting its role in
developmental dyslexia.
Furthermore, the diagnostic accuracy of VAS in detecting children with word reading problems seemed to be in line with the accuracy of PA, RAN and grapheme-phoneme
connections. However, these results need to be interpreted with caution, as these values were obtained within our sample of children with persistent literacy problems. Furthermore,
although the number of children with word reading problems who were correctly identified by the VAS task might seem relatively low (43.1%), it is important to note that, as explained by the multiple deficit theory, several cognitive risk factors can cause literacy problems.
impossible to obtain a perfect predictive accuracy. As the diagnostic accuracy of VAS has not been studied before, it is difficult to determine the significance of this value. However, the predictive value of 43.1% balances between the prevalence of a VAS deficit found in our study and the study of Bosse et al. (2007). The finding that the VAS task was successful in differentiating between children with dyslexia and typically developing children further supports the notion that the VAS task is able to detect children with reading problems.
However, despite these findings supporting the role of VAS in dyslexia, we did not find VAS to be a significant explanatory factor for reading and spelling performance in an analysis that included PA, RAN and grapheme-phoneme connections. These results are in line with those of Saksida et al. (2016), who found that VAS did not explain additional variance on top of phonological skills. They do, however, contrast with the findings of other studies, in which the contribution of VAS to reading and spelling performance was shown to be
independent of PA (Bosse & Valdois, 2009; Bosse et al., 2007; van den Boer et al., 2015). These different findings might be due to selective nature of our sample. Nevertheless, our findings do indicate that VAS does not add to PA, RAN and grapheme-phoneme connections, as these three factors all contributed to at least one measure of reading and spelling
performance, whereas VAS did not.
Our findings underline the complexity of developmental dyslexia as described by Pennington (2006). In line with the multiple deficit model of dyslexia, we found VAS deficits to often co-occur with deficits in PA, RAN and grapheme-phoneme connections. This
indicates that all four cognitive factors can contribute to dyslexia, highlighting the complex nature of dyslexia. However, as VAS did not stand out in any of the analyses comparing all cognitive measures and did not predict reading and spelling performance, we could not conclude that VAS is a valuable addition to PA, RAN and grapheme-phoneme connections. Thus, despite many positive findings concerning the role of VAS in dyslexia, we argue that adding the VAS task to established tasks used in the diagnostic procedure will not
significantly influence the number of children who will be diagnosed with developmental dyslexia.
As VAS did not turn out to be of additional value to the existing cognitive risk factors, the problem remains that a large amount of variance in reading and spelling is unexplained. Despite years of research, not all causes of dyslexia are fully understood. Consequently, the existing tasks are only able to identify a subset of the children hampered by the disorder. Children with reading and spelling problems not explained by established cognitive risk factors will be overlooked. The literacy problems of these children will not be remediated, which might have severe consequences. Studies have shown that the presence of dyslexia has socio-emotional effects, as children with dyslexia suffer from higher levels of stress and often show emotional manifestations such as fear, shyness and loneliness (Alexander-Passe, 2008). Thus, as dyslexia not only has severe implications for academic performance of children, but also for their socio-emotional development, additional research into the causes of dyslexia is necessary to not only identify children with dyslexia, but to also develop interventions that match the specific needs of children hampered by this heterogeneous disorder.
Furthermore, we did find some interesting results regarding VAS that need to be addressed as well. Although we did not find VAS to be a predictor of reading and spelling skills, we did find VAS to be related to word reading and spelling. The strongest relationship was found with word reading and spelling speed. The finding that VAS and reading speed are related is in line with the literature (Bosse & Valdois, 2009; Lobier, Dubois, & Valdois, 2013). As we have refuted the notion of the VAS hypothesis that VAS measures purely visual skills, the relation of VAS with reading speed cannot be explained by the fact that dyslexics have a smaller visual attentional window. Therefore, we would like to propose an explanation in line with the dual-route cascaded model (Coltheart, Rastle, Perry, Langdon, & Ziegler, 2001). Like the multiple-trace memory model by Ans et al. (2004), the dual-route cascaded model differentiates between two routes. In the lexical route, all letters are simultaneously
