ASSESSMENT OF
LETTER−SPEECH SOUND
LEARNING IN CHILDREN WITH
FAMILIAL RISK FOR DYSLEXIA IN
KINDERGARTEN
By Eline Teeuwen
Thesis for the bachelor Psychobiology
University of Amsterdam
dr. P.J.F. Snellings as supervisor
C.T. Verwimp, MSc as daily supervisor
Student number: 11711752
June 26, 2020
Acknowledgements
I’m extremely grateful to my daily supervisor C. Verwimp for her extraordinary support throughout my thesis, as well as Dr. J. Tijms for his support and insightful suggestions. I would also like to extend my gratitude to Dr. P. Snellings for the constructive advice. Moreover, I would like to thank S.
Bruinsma for the wonderful collaboration. It was always a pleasure coming to the UvA every day with such lovely and engaging people. Thanks again to all who helped me!!
Table of contents
Acknowledgements...2
Abstract...4
Introduction...5
Cognitive predictors of dyslexia...5
Educational approach...5
Familial risk for dyslexia...6
Detection of dyslexia...6
The current study...6
Method...9 Participants...9 Design...10 Measures...11 Procedure...11 Data analysis...12 Results...12 Descriptives...12
Reliability of assessment of dynamic training...12
Validity: letter−speech sound learning...13
Validity: influence of educational approach...15
Familial risk for dyslexia...16
Discussion...17
Reliability...17
Validity...17
Sensitivity...18
ASSESSMENT OF LETTER−SPEECH
SOUND LEARNING IN CHILDREN
WITH FAMILIAL RISK FOR DYSLEXIA
IN KINDERGARTEN
Abstract
Compared to traditional, static measures, dynamic assessment is a potential new measure aimed at earlier detection and intervention of children with dyslexia. During a game-based training, children had to learn to associate six Dutch speech sounds with novel artificial letters, followed by an
assessment of the letter−speech sound learning measuring the accuracy and reaction time. The main objective of this study was investigating whether the assessment of the dynamic training (ADT) was a sensitive measure for discriminating children with and without familial risk for dyslexia. Before assessing sensitivity of ADT, the reliability and validity were assessed. To do so, we selected a general sample of 248 kindergarten children, as well as a specific sample of 24 children at risk for dyslexia and 24 matched non-risk peers. Furthermore, phonemic awareness and letter knowledge were assessed. The results revealed that the ADT was reliable. Additionally, validity was indicated by a positive correlation between letter knowledge and letter−speech sound learning without the influence of the educational approach. Finally, ADT failed to discriminate children with and without familial risk for dyslexia based on the letter−speech sound learning performance. Although these results point towards the applicability of a dynamic assessment, more extended research is needed to improve validity of the task, by adjusting the ADT complexity or age group and by focusing on multiple deficits associated with dyslexia rather than merely focusing on letter−speech sound learning. Moreover, it might be useful to include a longitudinal design to assess which children with familial risk truly develop dyslexia at a later age.
Introduction
Reading, a cultural transmitted process guided by instructions, plays an important role in human culture and communication (Dehaene & Cohen, 2007; Univeristaria, 2001). Between 5% and 17% of children experience persistent reading problems due to developmental dyslexia (Habib & Giraud, 2013). An essential step for learning to read is learning to associate speech sounds with letters (Froyen, Bonte, Van Atteveldt, & Blomert, 2009). For the majority of learners it takes extensive practice before the alphabetic script code is automatically integrated (Blomert, 2011). Although improved, the integration of letter−speech sounds at the end of primary school is still not fully automated like it is in adults (Froyen et al., 2009). Moreover, children with developmental dyslexia seem to have even more enduring difficulties with this automatization, most likely due to a
letter−speech sound binding deficit resulting in problems with fluent reading (Gullick & Booth, 2020).
Cognitive predictors of dyslexia
Besides problems with letter−speech sound learning, poor phonemic awareness is strongly associated with reading disabilities (Melby-Lervåg, Lyster, & Hulme, 2012). Children with dyslexia and familial risk for dyslexia are detected with problems with phonemic awareness (Snowling, Gallagher, & Frith, 2013). Phonemic awareness (PA) is a widely investigated phenomenon and is defined as the abstract representation of speech sounds and the ability to identify and manipulate individual phonemes (Yeong & Rickard Liow, 2012). At first developmental dyslexia was claimed to be mainly caused by a phonemic awareness deficit (Vellutino, Fletcher, Snowling, & Scanlon, 2004). However, not all children with PA problems have dyslexia and not every dyslexic child has problems with PA (Wolf & Bowers, 1993). Previous research claimed that PA is necessary for letter−speech sound binding (Castles & Coltheart, 2004), but other studies found that one is able to learn letter−speech sound associations with limited PA (Castles, Coltheart, Wilson, Valpied, & Wedgwood, 2009).
Phonemic awareness strongly depends on the process of letter−speech sound acquisition (De Santos Loureiro et al., 2004). Only after literacy instruction PA improve, indicating PA to be the result of reading instruction (Morais, Bertelson, Cary, & Alegria, 1986). However, the skill of isolating phonemes, which requires PA, can be learned by children without knowledge of corresponding graphemes (Castles et al., 2009), which indicates that PA is not only caused by literacy training but has a reciprocal relationship with it. Furthermore, phonemic awareness and letter knowledge (LK) help with letter−speech sound learning (Cardoso-Martins, Mesquita, & Ehri, 2011; I. Karipidis et al., 2017; Share, 2004). Both difficulties with phonemic awareness and with letter knowledge seem to be early predictors of dyslexia (Carroll, Mundy, & Cunningham, 2014; Hulme, Nash, Gooch, Lervåg, & Snowling, 2015).
Educational approach
The educational approach for children has proven to influence the outcome of learning (McCandliss, 2010). The type of educational approach can vary among schools between a holistic, meaning−centred non-phonics approach and a phonics approach. The phonics approach contrary to the non-phonics instruction focusses on learning the letter−speech sound associations and explicitly explaining their applicability in reading and writing (Schaars & Segers, 2017). Combined with other reading instructions, the systematic phonics approach seems to be more effective in reading growth and reduction of reading problems than the non−phonics approach (Shriver & Development, 2000). In the Netherlands, formal reading instruction starts in first grade when children are around six years old. ‘Veilig leren lezen’ (Learning to Read Safely), an indirect phonics approach, is applied by roughly 75% of the Dutch schools (Garbe et al., 2016). However, the start and context of the pre-reading instruction in kindergarten varies among Dutch schools. In general, the pre-reading instruction focusses on the development of PA and letter identification (Netten & Verhoeven, 2012).
Familial risk for dyslexia
Since the reading disabilities can trigger a cascade of negative academic, economic and psychosocial consequences (Hudson, 2009), early detection of dyslexia is necessary. Early detection can result in improved interventions leading to prevention or reduction of the related problems (Höse et al., 2016). Due to a genetic predisposition the prevalence is higher when this developmental disorder occurs in the family. Children with familial risk for dyslexia have a 45% chance to develop a reading disability like dyslexia themselves (Snowling & Melby-Lervåg, 2016). Both dyslexic readers and normal readers are able to master basic knowledge of novel letter−speech sounds associations, but seem to differ in applying this knowledge in more complex tasks (Aravena, Snellings, Tijms, & van der Molen, 2013; Froyen et al., 2009). More specifically, despite equal knowledge of the associations, children with dyslexia performed slower and less accurate during letter−speech sound learning than normal readers under time-pressure (Aravena et al., 2013).
Detection of dyslexia
Dyslexia is typically not diagnosed until a child has persistently failed to learn to read, that is, often in grade 3 or later, whereas research suggests the optimal time-window for intervention is in kindergarten or first grade (Ozernov-palchik & Gaab, 2016). Traditionally dyslexia is measured with static measures such as assessments of letter knowledge and phonemic awareness. A new method is dynamic assessment, which is not widely investigated. Dynamic assessment, which is more focussed on the learning potential rather than on the learning outcome, predicts individual differences better than traditional, static measures (Aravena, Tijms, Snellings, & van der Molen, 2016). By employing dynamic assessment of letter−speech sound learning (L−SS learning), dyslexic learners can be
differentiated from normal learners (Aravena, Tijms, Snellings, & van der Molen, 2018). Furthermore, the results of dynamic assessment, especially the reaction time, seem to predict the future reading ability and the responsiveness to reading interventions (Aravena et al., 2016).
The current study
The goal of this study is to contribute to the emerging knowledge of letter−speech sound learning aiming to inform current reading interventions, especially for children with dyslexia or with familial risk for dyslexia.
To better understand the mechanisms involved in learning letter−speech sound associations, the current study investigates letter−speech sound learning in children in kindergarten. The main objective was investigating whether the assessment of the dynamic training (ADT) was a sensitive measure for discriminating children with and without familial risk for dyslexia. Before assessing the sensitivity of ADT, the reliability and the validity of the assessment of the dynamic training were investigated. First, was the ADT reliable? We hypothesize that the ADT was reliable. Second, the ADT validity was investigated by focussing on two parts. First of all, are either phonemic awareness or letter knowledge related to the ability to acquire letter−speech sound associations? We hypothesize that letter knowledge and phonemic awareness may be related to letter−speech sound learning but are not necessary to develop letter−speech sound associations. Second of all, does the educational approach influence the phonemic awareness, letter knowledge and the performance of letter−speech sound binding? We hypothesize that the educational approach has a stronger influence on phonemic awareness and letter knowledge than on letter−speech sound learnability. Besides focussing on letter−speech sound learning in general, we investigated whether the ADT was a sensitive measure for discriminating children with and without familial risk for dyslexia. So, do children with and without familial risk for dyslexia differ in their ability to learn new letter−speech sound associations? We hypothesize that at-risk children perform worse than no-risk children in the ability to learn letter−speech sound associations in kindergarten.
To investigate the difference in children with and without familial risk for dyslexia, we conducted a letter−speech sound training with children in kindergarten based on a prior study (Aravena et al., 2013). To do so, we selected a general school sample of 248 kindergarten children, as
well as a specific sample of 24 children at risk for dyslexia and 24 matched non-risk peers. The focus of the training was on letter−speech sound learning, because this seems to be key to dyslexia (Blomert, 2011). The training presented a game aimed at learning six novel letter−speech sound associations within an artificial orthography. By employing an artificial orthography, the initial learning of letter−speech sound associations was investigated, independent of age and any prior experience with letter−speech sound learning. The training was followed by assessment of dynamic training (ADT) recording the accuracy and reaction time of the responses as measures of the mastery of the learned correspondences. Furthermore, the letter knowledge and the phonemic awareness were assessed.
We expect the reliability of the ADT to be acceptable or above. Furthermore, as PA and LK help with letter−speech sound learning, both phonemic awareness and letter knowledge are expected to be linked to letter−speech sound learning in the school sample, as depicted in Figure 1.
Fig. 1. The predicted letter−speech sound learning in relation to the phonemic awareness (A) and the letter
knowledge (B) in the school sample.
Moreover, since the exposure to letter training and phonemic awareness can vary among schools, it is expected that in the school sample the educational approach leads to differences in phonemic awareness and letter knowledge which are both linked to the performance of
letter−speech sound binding. Figure 2 shows a simulation of the expected phonemic awareness and letter knowledge in relation to the schools.
Fig. 2. A simulation of the predicted phonemic awareness (A) and letter knowledge (B) in relation to
letter−speech sound learning between the different schools (coloured lines) in the school sample.
Since children with familial risk for dyslexia have a 45% chance to develop dyslexia (Snowling & Melby-Lervåg, 2016), prior findings concerning dyslexic readers are also expected to be similar in preliterate children with familial risk for dyslexia. Because dyslexics perform slower and less accurate when matching letter−speech sound correspondences than normal readers despite equal basic knowledge of the correspondences (Aravena et al., 2013), the prediction is that children with and without familial risk for dyslexia can be differentiated. More specifically, children with familial risk for dyslexia are expected to respond slower and to be more prone to errors than children without familial risk for dyslexia. Figure 3 shows the predicted reaction time and accuracy. Furthermore, children with familial risk for dyslexia are expected to have poorer phonemic awareness and less letter knowledge than no-risk children, because both phonemic awareness and letter knowledge seem early predictors of dyslexia (Carroll et al., 2014; Hulme et al., 2015). Figure 4 depicts the predicted phonemic awareness and letter knowledge.
Fig. 3. The predicted reaction time (A) and accuracy (B) of letter−speech sound learning between children with
Fig. 4. The predicted phonemic awareness (A) and letter knowledge (B) between children with (orange) and
without (blue) familial risk for dyslexia.
In summary, the objective was investigating whether the assessment of the dynamic training (ADT) was a sensitive measure for discriminating children with and without familial risk for dyslexia. First, reliability of ADT was investigated. Second, validity of ADT was assessed by investigating whether phonemic awareness or letter knowledge were related to the letter−speech sound learning
performance, and the influence of educational approach on this relation. For the sensitivity of ADT, we investigated whether pre-readers with and without familial risk for dyslexia differed in their ability to learn novel letter−speech sound associations. During a game-based training children learned to associate six speech sounds with novel artificial letters. After training, assessment of the learned associations measured the reaction time and accuracy. Furthermore, phonemic awareness and letter knowledge were assessed.
Method
Participants
Two samples were selected: the school sample and the RID sample. The total school sample consisted of 248 children (130 boys and 118 girls) with an age ranging from 4.08 to 6.81 years old (M = 5.27, SD = .52). The total RID sample consisted of 48 children, containing both children with and without familial risk for dyslexia. The ages ranged from 4.03 to 6.81 years old (M = 5.45, SD = .76). From the RID Institute, a Dutch organisation for dyslexia and other learning disabilities, we selected 24 children with an older brother or sister who was diagnosed with dyslexia, which defines them as children with familial risk for dyslexia (15 boys and 9 girls). From the school sample, 24 children without familial risk for dyslexia (15 boys and 9 girls) were selected as control group. The control group matched the at-risk children to control for type of school, school level, gender and age. All children were pupils in either the first or second year of kindergarten and spoke Dutch as their mother tongue. There were no exclusion criteria. Informed consent was given by the parents of each child. The ethics committee of the University of Amsterdam approved the study.
Design
Based on the training in prior research (Aravena et al., 2013), a training was developed with the aim of learning six novel letter−speech sound associations. The training consisted of a computer game in which children needed to match speech sounds to visual symbols, known as artificial letters.
Software-integrated instructions explained the specifics of the game at the start. In the game the children were told to save our sun. The underlying learning goal was not revealed to enhance implicit associative learning. Therefore, the learning process felt more natural and similar to real life learning. In the game, the artificial letters were displayed on balls and distributed across the screen. The child was repeatedly presented with a speech sound and was required to respond by tapping all of the corresponding grapheme balls. The non-corresponding balls acted as distractors. Feedback was given after every response; a green smiley for a correct response and a red sad face for an incorrect response. Correct responses resulted in success in the game and incorrect responses threatened this success. A new field was displayed when the child tapped on all of the correctly corresponding balls. The speech sound varied regularly, causing the tapping goal of the child to switch. The level of difficulty slowly increased during the game as the amount of distractors and time restrictions increased. Bonuses and time restrictions encouraged the child to play correctly and as fast as possible. Figure 5 displays screenshots of the training.
Fig. 5. Screenshots from the training. During training every letter is first introduced (A). After a speech sound is
given as goal, one or multiple artificial letters are presented (D). Feedback is given after every response; a green smiley for a correct response and a red sad face for an incorrect response (B&C).
The novel artificial letters were based on Asomtavruli, an ancient Georgian orthography. The script was novel in the sense that the artificial graphemes were unfamiliar to the children. This enabled the research of the initial learning of letter−speech sound associations. The artificial orthography
consisted of six graphemes, each randomly matched to a high-frequent Dutch phoneme (four consonants and two vowels). In Table 1 the letter−speech sound associations are displayed.
The artificial training was followed by three assessments; assessment of dynamic training, phonemic awareness and letter knowledge. Both the training and the assessment of dynamic training were executed on a tablet while the sounds were presented through headphones. The assessment of dynamic training functions as an indicator of the letter−speech sound integration.
Measures
Assessment of dynamic training. In the assessment of the dynamic training a phoneme (speech
sound) was presented through the headphones followed by a series of graphemes (artificial letters) displayed on the screen. One or more of these graphemes matched to the last heard phoneme and the remaining graphemes worked as a distraction. Each phoneme was presented randomly multiple times. The child was encouraged to respond as fast as possible by tapping the corresponding
grapheme balls. No feedback indicating whether the response was correct or incorrect was given. The software saved the responses including the reaction time automatically. The accuracy was
determined by the number of correct responses. The reaction time was expressed as the average reaction time of a correct response.
Phonemic awareness. The phonemic awareness (PA) was measured with the first sound awareness
task which is most suitable for children in kindergarten (Roos, Smit, Steege, & Wagenaar, 2010). For each item, the instructor presented four pictures and pronounced the monosyllabic high-frequent Dutch word belonging to the first picture. The child was required to point towards the one picture out of the remaining pictures with the same first speech sound as the first picture. The PA score was defined by the number of correct responses.
Letter knowledge. The letter knowledge was assessed with a list of 16 high-frequent letters in written
Dutch (Boets et al., 2010). These 16 Dutch letters are usually the first letters a child knows. In this task, the instructor pointed towards the letters one by one and the child named the corresponding phonemes out loud. The score was determined by the number of correct responses.
Procedure
As shown in Figure 6, an entire session consisted of the letter−speech sound training, the assessment of dynamic training, the phonemic awareness assessment and the letter knowledge assessment. In comparison with the Aravena et al. study (2013), the duration of the training and the assessment of dynamic training together were reduced to 15 minutes as this seemed a more suitable time frame. As a result, the total duration of the session was approximately 25 minutes. All participants completed the full session in a quiet classroom at the primary school of the child. Before a session started the instructor checked that the child was able to hear and see all stimuli of the training. For the artificial training the children were wearing headphones and used Lenovo Tab 2 A10-70F tablets in full-screen mode. The artificial training was programmed in Unity to operate on Android. The other assessments were conducted on a one-to-one basis with the instructor.
Fig. 6. Overview of a session. L−SS training: letter−speech sound training, ADT: assessment of dynamic training,
Data analysis
Data-analysis was executed in R-studio. To investigate the reliability of the assessment of the dynamic assessment (ADT), a correlation analysis was performed with time (first half and second half of ADT) as categorical predictor and accuracy and reaction time as dependent variables.
Next, validity of ADT was investigated in two parts. First, to determine the influence of phonemic awareness and letter knowledge on letter−speech sound learning, a multiple regression analysis was conducted with the school sample with PA and LK as predictor variables and L−SS learning as dependent outcome variable. Furthermore, to get more insight in the cause and consequence between PA and L−SS learning, we examined whether children with low PA were also able to reach high L−SS learning performance.
Second, to investigate the influence of the educational approach, a multilevel analysis with two levels was conducted. The first level consisted of PA and LK as predictors and L−SS learning as dependent variable with the primary school (9 schools) as second level predictor. New models with random intercept and/or random slope were compared to the original fixed intercept regression model, indicating whether a new model fitted significantly better.
Finally, we investigated the sensitivity of the ADT for discriminating children with and without familial risk for dyslexia with the RID sample. To determine the children differed based on
letter−speech sound performance, a MANOVA was conducted with condition (with or without familial risk for dyslexia) as independent predictor variable and accuracy and reaction time of L−SS learning as dependent outcome variable.
Results
Descriptives
Before running analyses, data were checked for missing values and outliers (i.e. scores that deviated from the Mahalanobis distance). Only those that fully completed the experiment were included in data-analysis. This resulted in 236 children in the school sample and 48 children in the RID sample. Table 2 shows the mean scores and standard deviations of the RID sample (2A) and of the school sample (2B).
Table 2. Means and standard deviations for letter−speech sound (L−SS) accuracy and reaction time during
assessment of dynamic training, the phonemic awareness and the letter knowledge of the RID sample (A) and of the school sample (B).
Reliability of assessment of dynamic training
To investigate the ADT reliability, two correlation analyses were performed with time (first half and second half of ADT) as categorical predictor and letter−speech sound learning as dependent variable, one with L−SS accuracy and one with L−SS reaction time. Since the assumptions of the Pearson correlation were not met, Spearman correlations were conducted. As showed in Figure 7, we found a significant positive association between the first and the second half of the assessment of dynamic training for accuracy, (Rs = .55, p < .001), and reaction time, (Rs = .62, p < .001). Since a split-half analysis only investigates the reliability of half a test, the Spearman-Brown Prediction formula was used to estimate the reliability for the full-length test. This resulted in acceptable reliabilities (e.g. George & Mallery, 2003) for accuracy (Rs = .71) and reaction time (Rs = .76).
Fig. 7. Scatter plot of the first and second half of the assessment of dynamic training for accuracy (A) and
reaction time (B) of the school sample.
Validity: letter−speech sound learning
To determine the influence of phonemic awareness and letter knowledge on letter−speech sound learning, two multiple linear regression analyses were conducted in the school sample. The correlations between PA, LK and L−SS learning are shown in Table 3.
Table 3. Pearson correlations among the test scores where * indicates a p-value below .05 and ** a p-value
below .001.
First, a multiple regression was calculated to predict accuracy based on PA and LK. Since the assumption of linearity and homoscedasticity were not met, a robust regression was executed. PA and LK were positively correlated with accuracy (F(2,236) = 42.75, p < .001), with an R2 of .333.
However, PA did not contribute significantly to predict L−SS accuracy after controlling for the variance explained by LK (F(2,236) = 2.06, p = .152). Therefore, LK was a significant predictor of L−SS accuracy (F(1,237) = 93.63, p < .001), with an R2 of .327. As showed in Figure 8, participants’ predicted accuracy
Fig. 8. Scatterplot of the letter knowledge in relation to the accuracy during ADT of the school sample. Second, a multiple regression was conducted to predict reaction time based on PA and LK. The Shapiro-Wilk test indicated that the assumption of normality on the residuals of the reaction time scores was not met. Therefore, the dependent variable (reaction time) was transformed with a natural logarithm (Tabachnick & Fidell, 2007). Transformed reaction time was therefore used
throughout this study instead of the original reaction time. PA and LK were positively correlated with reaction time (F(2,233) = 16.87, p < .001), with an R2 of .127. However, PA did not contribute
significantly to predict transformed reaction time after controlling for the variance explained by LK (p = .154). LK was a significant predictor of transformed reaction time (F(1,234) = 31.54, p < .001), with an R2 of .119. As shown in Figure 9, participants’ predicted logarithmic reaction time is equal to .508 +
.019 (LK), where LK was coded as number of correct responses.
Fig. 9. Scatterplot of the letter knowledge in relation to the natural logarithm of reaction time during ADT of the
school sample.
To investigate whether children with low PA were able to reach high L−SS learning
performance, we conducted two Chi-square tests of independence, one for L-SS accuracy and one for L-SS reaction time. For PA, children were divided in two groups: low PA or high PA. Low PA was defined as below chance level. This is, as the maximum score on the PA task was 16 and each item
consisted of three answer alternatives, a total score below 5.3. High PA was defined as a score above 5.3. The accuracy scores were split at the median, creating accuracy above the median and accuracy below the median. We followed the same procedure for reaction time, creating reaction time above and below the median. The assumptions of the Chi-square tests were met. Results of the Chi-square test indicated significant interaction between PA and accuracy (χ2 (1, N = 48) = 19.27, p < 0.01, r = .63)
and between PA and reaction time (χ2 (1, N = 48) = 9.91, p < 0.01, r = .45). This means that there is a
significant relationship between PA and L−SS learning. As depicted in Table 4, children with low PA often show low L−SS learning and are less likely to reach high L−SS performance compared to the children with high PA abilities. In contrast, children with high PA often show high L−SS learning. As some children with low PA were able to reach high L−SS learning, these results suggest that good PA abilities are not necessary but are helpful.
Table 4. The number of children with accuracy (A) and reaction time (B) below or above the median with low or
high PA.
Validity: influence of educational approach
To investigate the influence of the educational approach, two multilevel analyses with two levels were conducted, one for L−SS accuracy and one for L−SS reaction time. As described earlier, PA is no significant predictor of letter−speech sound learning after controlling for LK. Therefore, LK was the only first level predictor (fixed effect) with L−SS learning as dependent variable. The primary school (9 schools) was the second level predictor (random effect) resulting in clustering per school. Because the assumptions of linearity and homoscedasticity were not met, the results should be interpreted with care. For L−SS accuracy, the random intercept model fitted significantly better compared to the random intercept model without LK as fixed level-1 factor (p < .001). The random intercept and slope model did not fit better than the random intercept model (p = .796). So, the best fit for investigating accuracy based on LK with primary school as second level is the random intercept model (AIC = 1442.13, BIC = 1455.98, logLik = -717.06) (Figure 10). However, the fixed intercept model fitted significantly better than this random intercept model (p < .001). Therefore, the educational approach did not significantly influence the relation between LK and L−SS accuracy.
Fig. 10. Random intercept model of accuracy in relation to letter knowledge with the primary school as second
For L−SS reaction time, the random intercept model fitted significantly better compared to the random intercept model without LK as fixed level-1 factor (p < .001). The random intercept and slope model did not fit better than the random intercept model (p = .796). So, the best fit for investigating logarithmic reaction time based on LK with primary school as second level is the random intercept model (AIC = 79.31, BIC = 93.13, logLik = -35.65) (Figure 11). However, the fixed intercept model fitted significantly better than this random intercept model (p < .001). Therefore, the educational approach did not significantly influence the relation between LK and L−SS reaction time.
Fig. 11. Model of logarithmic reaction time in relation to letter knowledge with the primary school as random
effect. Each coloured line represents a different school.
Familial risk for dyslexia
To determine the influence of familial risk for dyslexia on letter−speech sound learning, we compared the scores of the letter−speech sound learning per condition. The assumptions were met. A
multivariate analysis of variance (MANOVA) was conducted with the RID sample with condition (with or without familial risk for dyslexia) as independent predictor variable and accuracy and reaction time during ADT as dependent outcome variables. As depicted in Figure 12, there was no statistically significant difference in letter−speech sound learning based on the child’s condition (F(1,46) = .503, p = .608, Pillai’s Trace = .022).
Fig. 12. The accuracy (A) and reaction time (B) between children without (No-risk) and with (At-risk) familial risk
A second analysis investigated the influence of familial risk for dyslexia on PA and LK. The assumptions were met. A multivariate analysis of variance (MANOVA) was conducted with the RID sample with condition (with or without familial risk for dyslexia) as independent predictor variable and PA and LK as dependent outcome variables. As depicted in Figure 13, there was a statistically significant difference based on the child’s condition, (F(1,46) = 5.421, p = .008, Pillai’s Trace = .194). This difference was significant for PA (F(1,46) = 9.263, p = .004), but not for LK (F(1,46) = 2.170, p = .148).
Fig. 13. The letter knowledge (A) and phonemic awareness (B) between children without (No−risk) and with
(At−risk) familial risk for dyslexia in the RID sample.
Discussion
Recently, a letter-speech sound binding deficit have been proposed to be associated with
developmental dyslexia. This study aimed to contribute to the emerging knowledge of letter−speech sound learning aiming to inform current reading interventions. We investigated whether the
assessment of the dynamic training (ADT) was a sensitive measure for discriminating children with and without familial risk for dyslexia. Before assessing the sensitivity of ADT, the reliability and the validity of the assessment of the dynamic training were investigated.
Reliability
The goal of the ADT was detecting children at risk for dyslexia. The results of a split-half reliability indicated that the assessment of the dynamic training had a reliability above .70. According to general conventions this is an acceptable reliability (Evers, Lucassen, Meijer, & Sijtsma, 2009), suggesting that all items contribute equally to the outcome.
Validity
In order to provide a window on the validity of the assessment of dynamic training, we addressed its construct validity. That is, we tested whether the task measured letter−speech sound learning like it is supposed to (Evers et al., 2009). For validity, we investigated whether phonemic awareness or letter knowledge were related to the letter−speech sound learning performance, and the influence of educational approach on this relation. First, the study confirmed a positive
correlation between LK, PA and L−SS learning. However, after controlling for the influence of LK, PA did not contribute significantly in predicting L−SS learning. As the correlation between PA and LK is high, an explanation might be that phonemic awareness and letter knowledge overlap. The phonemic awareness was measured with the first sound awareness task. Possibly knowledge of letters of the
first sounds helped with the phonemic awareness performance. Future research could involve younger children, as they have less letter knowledge resulting in less overlap between PA and LK.
Furthermore, letter knowledge was, as expected, positively correlated with accuracy, which is in line with prior research (Cardoso-Martins et al., 2011). Knowledge of the newly learned L−SS associations were correlated with the letter knowledge, indicating validity. Contrary to the hypothesised association, the correlation between letter knowledge and reaction time was also positive. A possible explanation could be a trade-off between reaction time and accuracy. Some children could have been highly focussed on responding accurately, and as a result responded slower. Another explanation could be that the delay in reaction time reflected a sense of mental control of the child. Some children with no clue which response was correct might have reacted faster. Future research could focus on making the training more understandable for the child.
Furthermore, a correlation between phonemic awareness and letter−speech sound learning was expected. Although a positive correlation between phonemic awareness and letter−speech sound learning was found, phonemic awareness did not contribute significantly to predict letter−speech sound accuracy after controlling for the variance explained by LK.
It has been postulated that PA is needed to learn letter−speech sound associations. However, recent studies have reported that PA may develop as a result of reading. To get more insight in the cause and consequence, we examined whether children with low PA were also able to reach high L−SS learning. We found that children with low PA often show low L−SS learning and are less likely to reach high L−SS performance. As some children with low PA were able to reach high L−SS learning, these results suggest that good PA abilities are not necessary but are helpful. Contrary to the classical theory where phonemic awareness is a necessity for learning to read (Castles & Coltheart, 2004), these results are in line with the recent theory. This recent theory states that letter−speech sound learning is central to learning to read, where phonemic awareness is a by-product of this process (Castles et al., 2009).
Besides investigating the relation between phonemic awareness and letter knowledge to letter−speech sound learning, the influence of the educational approach was investigated. Contrary to the hypothesis, the education approach did not influence the relation between letter knowledge and letter−speech sound learning, indicating validity of ADT. However, as children in kindergarten did not yet receive reading instructions, results might differ when older children are included. Future research could further investigate the validity of the ADT by taking the complexity and age group into account.
In summary, we found that LK and PA, two important indicators of dyslexia, were positively related to L−SS learning. Additionally, this relation was not moderated by the educational approach, suggesting that the ADT was valid and assessed what is what supposed to assess instead of school influence.
Sensitivity
In the next part of the study, we addressed the sensitivity of the ADT. Therefore, we investigated whether the ADT was able to discriminate children with familial risk for dyslexia from non-risk peers. Contrary to the hypothesis, the data demonstrated that children with and without familial risk for dyslexia did not differ in the letter−speech sound learning performance. However, the results confirm that children with and without familial risk for dyslexia can be differentiated based on letter
knowledge and phonemic awareness. This is in line with the theory that these two measures seem to be early predictors of dyslexia (Carroll et al., 2014; Hulme et al., 2015).
Letter knowledge and phonemic awareness are part of the traditional, static assessments for detecting dyslexia. However, traditional assessments are usually performed when a child is detected with a reading problem, although intervention at an earlier age is more effective (Ozernov-palchik & Gaab, 2016). Dynamic assessment can be executed at younger ages and thus before the start of reading instruction, making potential intervention more effective. Another advantage of dynamic
assessment is that it can predict individual differences better and that it focusses on the learning potential rather than the learning outcome. Therefore, dynamic assessment seems to be a fitting potential new measure for detecting dyslexia. Since dynamic assessment is a relatively new measure for detecting dyslexia, this study contributed to the emerging knowledge surrounding dynamic assessment aiming at improving early interventions.
Further research should take into account that only correlations can be suggested between letter knowledge and letter−speech sound learning from this study. No causal relationship can be deducted, since the precise development of this relationship is yet unclear.
Another limitation of this study is that this is conducted with 24 children with familial risk for dyslexia of who approximately 45% will develop dyslexia. This results in low power of the study. Also, it is unknown who does and does not develop dyslexia. Future research could conduct a longitudinal study investigating which children with familial risk for dyslexia developed dyslexia and how they differed from the other children.
Not being able to differentiate between children with and without familial risk could be due to multiple factors. For instance, the children were not triggered by the game to focus on the goal, as some children randomly tapped on the tablet. This could be due to lack of understanding or lack of motivation. Future research could focus on improving the level of the training and the motivation for the children. Furthermore, this study was focussed on the ADT which was assessed after the training. Besides investigating the knowledge after training, it would be interesting to investigate the learning growth during the training. Next to methodological changes, we could argue whether this study was too focussed on a singular deficit of dyslexia. According to a study of Pennington et al. (2012),
dyslexia is a disorder that could be caused by multiple deficits. Further research is needed to establish a method that can differentiate children based on multiple deficits.
In summary, the objective was investigating whether the assessment of the dynamic training (ADT) was a sensitive measure for discriminating children with and without familial risk for dyslexia. The ADT was reliable. Furthermore, the task was found to be valid as L-SS performance was related to LK and PA, the two most robust predictors of dyslexia. Also, the absence of influence of the
educational approach further indicated validity. However, ADT failed to differentiate children with and without familial risk for dyslexia based on letter−speech sound learning. Although these results point towards the applicability of a dynamic assessment, more extended research is needed to improve validity of the task, by adjusting the ADT complexity or age group and by focusing on multiple deficits associated with dyslexia rather than merely focusing on letter−speech sound learning. Moreover, it might be useful to include a longitudinal design to assess which children with familial risk truly develop dyslexia at a later age.
References
Aravena, S., Snellings, P., Tijms, J., & van der Molen, M. W. (2013). A lab-controlled simulation of a letter-speech sound binding deficit in dyslexia. Journal of Experimental Child Psychology, 115(4), 691–707. https://doi.org/10.1016/j.jecp.2013.03.009
Aravena, S., Tijms, J., Snellings, P., & van der Molen, M. W. (2016). Predicting responsiveness to intervention in dyslexia using dynamic assessment. Learning and Individual Differences, 49, 209– 215. https://doi.org/10.1016/j.lindif.2016.06.024
Aravena, S., Tijms, J., Snellings, P., & van der Molen, M. W. (2018). Predicting Individual Differences in Reading and Spelling Skill With Artificial Script–Based Letter–Speech Sound Training. Journal of
Learning Disabilities, 51(6), 552–564. https://doi.org/10.1177/0022219417715407
Blomert, L. (2011). The neural signature of orthographic-phonological binding in successful and failing reading development. NeuroImage, 57(3), 695–703.
https://doi.org/10.1016/j.neuroimage.2010.11.003
Boets, B., De Smedt, B., Cleuren, L., Vandewalle, E., Wouters, J., & Ghesquiére, P. (2010). Towards a further characterization of phonological and literacy problems in Dutch-speaking children with dyslexia. British Journal of Developmental Psychology, 28(1), 5–31.
https://doi.org/10.1348/026151010X485223
Cardoso-Martins, C., Mesquita, T. C. L., & Ehri, L. (2011). Letter names and phonological awareness help children to learn letter-sound relations. Journal of Experimental Child Psychology, 109(1), 25–38. https://doi.org/10.1016/j.jecp.2010.12.006
Carroll, J. M., Mundy, I. R., & Cunningham, A. J. (2014). The roles of family history of dyslexia, language, speech production and phonological processing in predicting literacy progress.
Developmental Science, 17(5), 727–742. https://doi.org/10.1111/desc.12153
Castles, A., & Coltheart, M. (2004). Is there a causal link from phonological awareness to success in learning to read? Cognition, 91(1), 77–111. https://doi.org/10.1016/S0010-0277(03)00164-1 Castles, A., Coltheart, M., Wilson, K., Valpied, J., & Wedgwood, J. (2009). The genesis of reading
ability: What helps children learn letter-sound correspondences? Journal of Experimental Child
Psychology. https://doi.org/10.1016/j.jecp.2008.12.003
De Santos Loureiro, C., Willadino Braga, L., Do Nascimento Souza, L., Nunes Filho, G., Queiroz, E., & Dellatolas, G. (2004). Degree of illiteracy and phonological and metaphonological skills in unschooled adults. Brain and Language. https://doi.org/10.1016/j.bandl.2003.12.008 Dehaene, S., & Cohen, L. (2007). Cultural recycling of cortical maps. Neuron, 56(2), 384–398.
https://doi.org/10.1016/j.neuron.2007.10.004
Evers, A., Lucassen, W., Meijer, R., & Sijtsma, K. (2009). COTAN Beoordelingssysteem voor de kwaliteit van tests, 1–46.
Froyen, D. J. W., Bonte, M. L., Van Atteveldt, N., & Blomert, L. (2009). The long road to automation: Neurocognitive development of letter-speech sound processing. Journal of Cognitive
Neuroscience, 21(3), 567–580. https://doi.org/10.1162/jocn.2009.21061
Garbe, C., Krosse, H. de, Lafontaine, D., Netten, A., Shiel, G., Strating, H., & Valtin, R. (2016). Literacy in The Netherlands, (March), 1–55.
George, D., & Mallery, P. (2003). SPSS for Windows step by step: A simple guide and reference. 11.0 update, 231. https://doi.org/10.1016/B978-0-444-88933-1.50023-4
Gullick, M., & Booth, J. R. (2020). Neurocognitive Bases of Crossmodal Integration Deficit in Dyslexia.
PsyArXiv, 1–40.
Habib, M., & Giraud, K. (2013). Dyslexia. Handbook of Clinical Neurology, 111, 229—235. https://doi.org/10.1016/b978-0-444-52891-9.00023-3
Höse, A., Wyschkon, A., Moraske, S., Eggeling, M., Quandte, S., Kohn, J., … Esser, G. (2016). Prävention von Lese-Rechtschreibstörungen: Kurz- und mittelfristige Effekte einer Förderung der
Vorschulalter. Zeitschrift Für Kinder- Und Jugendpsychiatrie Und Psychotherapie, 44(5), 377–393. https://doi.org/10.1024/1422-4917/a000456
Hulme, C., Nash, H. M., Gooch, D., Lervåg, A., & Snowling, M. J. (2015). The Foundations of Literacy Development in Children at Familial Risk of Dyslexia. Psychological Science, 26(12), 1877–1886. https://doi.org/10.1177/0956797615603702
I. Karipidis, I., Pleisch, G., Röthlisberger, M., Hofstetter, C., Dornbierer, D., Stämpfli, P., & Brem, S. (2017). Neural initialization of audiovisual integration in prereaders at varying risk for developmental dyslexia. Human Brain Mapping, 38(2), 1038–1055.
https://doi.org/10.1002/hbm.23437
McCandliss, B. D. (2010). Educational neuroscience: The early years. Proceedings of the National
Academy of Sciences of the United States of America, 107(18), 8049–8050.
https://doi.org/10.1073/pnas.1003431107
Melby-Lervåg, M., Lyster, S. A. H., & Hulme, C. (2012). Phonological skills and their role in learning to read: A meta-analytic review. Psychological Bulletin, 138(2), 322–352.
https://doi.org/10.1037/a0026744
Morais, J., Bertelson, P., Cary, L., & Alegria, J. (1986). Literacy training and speech segmentation.
Cognition. https://doi.org/10.1016/0010-0277(86)90004-1
Netten, A., & Verhoeven, L. (2012). PIRLS 2011 Encyclopedia. Educational policy and curriculum in
reading: The Netherlands (Vol. 2).
Ozernov-palchik, O., & Gaab, N. (2016). Tackling the “dyslexia paradox”: reading brain and behavior for early markers of developmental dyslexia. Wiley Interdisciplinary Reviews: Cognitive Science,
7(2), 156–176. https://doi.org/10.1002/wcs.1383.Tackling
Pennington, B. F., Santerre-lemmon, L., Rosenberg, J., Macdonald, B., Boada, R., Friend, A., … Byrne, B. (2012). Individual Prediction of Dyslexia by Single vs. Multiple Deficit Models. Journal of
Abnormal Psychology, 121(1), 212–224. https://doi.org/10.1037/a0025823.Individual
Roos, S. De, Smit, F., Steege, S., & Wagenaar, N. (2010). De invloed van fonemisch bewustzijn & omgevingsfactoren op letterklank leren.
Schaars, M. M. H., & Segers, E. (2017). Word decoding development in incremental phonics instruction in a transparent orthography. Reading and Writing, 30(7), 1529–1550. https://doi.org/10.1007/s11145-017-9735-3
Share, D. L. (2004). Knowing letter names and learning letter sounds: A causal connection. Journal of
Experimental Child Psychology, 88(3), 213–233. https://doi.org/10.1016/j.jecp.2004.03.005
Shriver, E. K., & Development, N. I. of C. H. and H. (2000). Report of the Nationial Reading Panel: Teaching Children to Read: An Evidence-Based Assessment of the Scientific Research Literature of Reading and Its Implications for Reading Instruction. U.S. Government Printing Office, 00– 4754. https://doi.org/10.1002/ppul.1950070418
Snowling, M. J., Gallagher, A., & Frith, U. (2013). Family Risk of Dyslexia Is Continuous : Individual Differences in the Precursors of Reading Skill Author ( s ): Margaret J . Snowling , Alison Gallagher and Uta Frith Published by : Wiley on behalf of the Society for Research in Child Development Stable U, 74(2), 358–373.
Snowling, M. J., & Melby-Lervåg, M. (2016). Oral language deficits in familial dyslexia: A meta-analysis and review. Psychological Bulletin, 142(5), 498–545. https://doi.org/10.1037/bul0000037 Tabachnick, B. G., & Fidell, L. S. (2007). Using Multivariate Statistics, 1–14.
Univeristaria, C. (2001). The importance of the act of reading. Florida Entomologist. https://doi.org/10.1057/rpm.2015.33
Vellutino, F. R., Fletcher, J. M., Snowling, M. J., & Scanlon, D. M. (2004). Specific reading disability (dyslexia): What have we learned in the past four decades? Journal of Child Psychology and
Psychiatry and Allied Disciplines, 45(1), 2–40.
https://doi.org/10.1046/j.0021-9630.2003.00305.x
Wolf, M., & Bowers, P. G. (1993). Theoretical links among naming speed , precise timing mechanisms and orthographic skill in dyslexia. Reading and Writing: An Interdisciplinary Journal, 5, 69–85. https://doi.org/10.1037/0022-0663.91.3.415
Yeong, S. H. M., & Rickard Liow, S. J. (2012). Development of phonological awareness in English-Mandarin bilinguals: A comparison of English-L1 and English-Mandarin-L1 kindergarten children.
