Journal of Experimental Psychology:
Learning, Memory, and Cognition
No Solid Empirical Evidence for the SOLID (Serial Order
Learning Impairment) Hypothesis of Dyslexia
Eva Staels and Wim Van den Broeck
Online First Publication, October 13, 2014. http://dx.doi.org/10.1037/xlm0000054
CITATION
Staels, E., & Van den Broeck, W. (2014, October 13). No Solid Empirical Evidence for the
SOLID (Serial Order Learning Impairment) Hypothesis of Dyslexia. Journal of Experimental
Psychology: Learning, Memory, and Cognition. Advance online publication.
http://dx.doi.org/10.1037/xlm0000054
Journal of Experimental Psychology:
Learning, Memory, and Cognition
2014, Vol. 40, No. 6, 000
© 2014 American Psychological Association
0278-7393/14/$12.00 http://dx.doi.org/10.1037/xlm0000054
No Solid Empirical Evidence for the SOLID (Serial Order Learning
Impairment) Hypothesis of Dyslexia
Eva Staels and Wim Van den Broeck
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
Vrije Universiteit Brussel
This article reports on 2 studies that attempted to replicate the findings of a study by Szmalec, Loncke,
Page, and Duyck (2011) on Hebb repetition learning in dyslexic individuals, from which these authors
concluded that dyslexics suffer from a deficit in long-term learning of serial order information. In 2
experiments, 1 on adolescents (N ⫽ 59) and 1 on children (N ⫽ 57), no empirical evidence was obtained
for impaired Hebb learning in dyslexics, whether the same data-analytical procedure as Szmalec et al.
was used or whether some methodological improvements were applied (e.g., using a more sensitive index
of Hebb learning, and equating groups on filler performance with state trace analysis). In an additional
state trace analysis, aggregating data over participants, it was shown that performance on the repeated
Hebb sequences was almost perfectly predictable from performance on the nonrepeated sequences
(fillers). The implications of these findings are outlined for the current discussion on the mechanisms for
encoding immediate serial recall and long-term sequence learning and for computational models
attempting to simulate these mechanisms.
Keywords: dyslexia, serial order learning, Hebb learning, sequence learning, short-term memory
seek an underlying cause of dyslexia in the domain of general
cognitive processes. Before discussing the literature about implicit
sequence learning in dyslexia, we first take a brief look at these
diverse explanations of dyslexia.
The most dominant theory, supported by a majority of reading
researchers, attributes the specific problems associated with dyslexia to a phonological processing deficit (for reviews, see Stanovich & Siegel, 1994; Vellutino, Fletcher, Snowling, & Scanlon,
2004; Ziegler & Goswami, 2005). These phonological problems
cause difficulties in learning grapheme–phoneme associations and
in mappings between how words appear in print and how they
sound (Bradley & Bryant, 1983; Ramus et al., 2003; Snowling,
2001). However, this phonological deficit is not presumed to be
entirely specific to the reading domain. Indeed, a fundamental
problem with the phonological deficit hypothesis is why the phonological processing deficit fails to manifest itself or is much less
evident outside the domain of the reading process. Indeed, one
would expect the extremely complex phonological processes in
speech perception and production to be even more vulnerable to a
phonological deficit than the phonological processes involved in
reading relatively simple words at the start of reading development. Recently, several studies have casted doubt about the precise
nature of the presumed dyslexic phonological deficit. It seems that
the phonological representations of persons with dyslexia are
normal, but the phonological problems they experience can only be
demonstrated when task requirements enhance the processing load
in terms of short-term memory (STM), conscious awareness, or
time constraints (Ramus & Szenkovits, 2008). Hence, the basic
problem of individuals with dyslexia is not in the nature of the
phonological representations but rather in accessing these representations (see also Boets et al., 2013).
Even though the very hallmark of dyslexia is a specific failure
to learn to read, dyslexic persons show impairments on a wider
variety of cognitive tasks. Cognitive differences between reading-
It is not uncommon for research into specific memory deficits in
certain disabled groups to offer important insights into general
memory processes overall (e.g., implicit memory in amnesia; see
Schacter, 1987). The current study investigates implicit sequence
learning in primary school children and higher education students
with dyslexia. As learning to read words and texts can be understood as the acquisition of grapheme and phoneme sequences,
studying sequence learning in dyslexics can potentially improve
our insights into the mechanisms and impact of sequence learning
in general. Developmental dyslexia is commonly defined as a
disability characterized by low reading achievement and deficiencies in learning to spell and write (Snowling, 2012). Since the
beginning of the research into dyslexia, a number of causal hypotheses have been formulated. The literature is divided into two
types of causal hypotheses. On the one hand, most researchers
believe in a rather specific deficit. On the other, some researchers
Eva Staels and Wim Van den Broeck, Department of Clinical and Life
Span Psychology, Faculty of Psychology and Educational Sciences, Vrije
Universiteit Brussel.
Both authors contributed equally to the study. The results of this study
were presented at the Nineteenth Annual Meeting of the Society for the
Scientific Study of Reading in July 2012. It is our pleasure to thank those
who made this study possible. In particular, we would like to thank
the children who participated in this study, their parents, and their teachers.
We also thank the following research assistants for collecting of the data:
Sandra Dammans, Elodie Truyens and Dascha Nowak. Finally, we thank
Arnaud Szmalec for kindly sharing his data.
Correspondence concerning this article should be addressed to Wim Van
den Broeck, Vrije Universiteit Brussel, Faculty of Psychology & Educational Sciences, Department of Clinical & Life Span Psychology, Pleinlaan
2, B-1050 Brussels, Belgium. E-mail: [email protected]
1
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
2
STAELS AND VAN DEN BROECK
disabled and nondisabled persons are found virtually everywhere.
It has been demonstrated that poor readers score lower than normal
readers on syntactic awareness tasks (Bentin, Deutsch, & Liberman, 1990; Chung, Ho, Chan, Tsang, & Lee, 2013), on measures
of general linguistic awareness (Johnson, 1993; Siegel & Ryan,
1984), on general rule learning tasks (Fletcher & Prior, 1990;
Manis et al., 1987; Morrison, 1984), on short term memory and
processing strategy tasks (Boets, De Smedt, Cleuren, Vandewalle,
& Ghesquiere, 2010; Brady, 1991; Share, 1994; Siegel, 1994;
Torgesen, 1978), and on measures of metacognitive strategies
(Baker, 1982; Wong, 1991). These results are consistent with the
one-half standard deviation IQ deficit displayed by poor readers
(Stanovich, 1986). Based on the observation that dyslexic persons
often present a range of impairments, some researchers believe that
the underlying cause of dyslexia should be situated in a more
general cognitive process.
One example of a deficit in a general cognitive process that
might have a negative impact on the acquisition of reading and
writing skills, because learning to read involves both explicit and
implicit processes, is a deficit in implicit learning. Initially, children learn grapheme–phoneme correspondences explicitly and often under supervision. Afterward, they apply and continue to learn
how phonology is mapped into its written representation implicitly
and unsupervised (Gombert, 2003; Petersson & Reis, 2006;
Ziegler & Goswami, 2005). Yet the literature on implicit learning
in dyslexia is relatively new and has produced mixed results (for
reviews, see Folia et al., 2008; Orban, Lungu, & Doyon, 2008).
Implicit learning in dyslexic persons has often been investigated
by using different variants of the serial reaction time (SRT) task
(Nissen & Bullemer, 1987). Although no task is process pure (e.g.,
Destrebecqz & Cleeremans, 2001), learning in an SRT task is
called implicit because the acquisition of the knowledge about the
presented sequence occurs without explicit learning instructions
and is also observed if the presented sequence is not explicitly
discovered by the participant. In this paradigm, participants have
to respond as quickly and as accurately as possible to stimuli that
follow a sequential pattern. The participants are unaware of the
existence of the repeating sequence. Learning is then demonstrated
by a decrease in reaction time and number of errors in response to
the sequential pattern compared to random stimulus presentation.
Several studies have reported that dyslexic children are impaired in
implicit sequence learning (Menghini, Hagberg, Caltagirone,
Petrosini, & Vicari, 2006; Stoodley, Harrison, & Stein, 2006;
Stoodley, Ray, Jack, & Stein, 2008; Vicari et al., 2005; Vicari,
Marotta, Menghini, Molinari, & Petrosini, 2003). However, other
studies failed to find a specific dyslexic deficit in implicit sequence
learning (Deroost et al., 2010; Kelly, Griffiths, & Frith, 2002;
Roodenrys & Dunn, 2008; Waber et al., 2003).
The focus of the presented study is yet another paradigm that
can be used to investigate whether dyslexic students suffer from a
sequential learning deficit: the Hebb repetition learning paradigm.
Hebb (1961) asked his participants to perform an immediate serial
recall task in which one specific sequence of digits was repeated
every third trial. Participants performed significantly better in the
recall of repeating sequences compared with nonrepeating sequences. This phenomenon was called the Hebb repetition effect.
In a recent study, Szmalec, Loncke, Page, and Duyck (2011)
demonstrated that the Hebb repetition effect is affected in dyslexic
individuals, even for nonverbal modalities. Their interpretation of
the results is that dyslexics do not suffer from an overall implicit
sequence learning deficit but rather experience difficulties with the
consolidation or transfer of serial-order information, initially
stored in STM memory, into a stable long-term memory (LTM)
trace. Consolidation in LTM is seen as a partly independent or
separate modality-nonspecific process (Szmalec et al., 2011, p.
1275). The SRT task differs from the Hebb repetition paradigm in
such a way that the Hebb effect is not really a measure of pure
implicit learning, because participants are mostly aware of the
repeating sequence. However, awareness appears to have no impact on the degree of learning in this paradigm (e.g., Stadler, 1993;
but see Weitz, O’Shea, Zook, & Needham, 2011). In other words,
Hebb learning seems to be implicit in some participants but explicit in others.
Importantly, Szmalec et al. (2011) have made a persuasive
argument for the crucial role of processing serial order information
in language learning. As learning serial order information is commonly believed to play a crucial role in cognition because many
aspects of daily human behavior are sequential in nature, this
ability is probably very essential in language learning and language processing as well (e.g., Conway & Christiansen, 2001;
Page & Norris, 2008, 2009). The precise nature of the relation
between serial-order learning and language learning has long been
debated (e.g., Conway & Pisoni, 2008). The idea that the Hebb
repetition effect is a laboratory analogue of novel word learning
has further been elaborated by Szmalec, Duyck, Vandierendonck,
Mata, and Page (2009). In their study, sequences of nonsense
syllables (e.g., zi-lo-ka-ho-fi-se-be-ru-mo) were used in a Hebb
learning procedure. Afterward, an auditory lexical decision task
was administered. The nonwords in this task were constructed
from the syllables used in the repeating Hebb sequences (ziloka,
hofise, berumo). Response times on the Hebb-based nonwords
were significantly longer than on the control nonwords—that is,
they were more slowly identified as nonwords. Apparently, the
repeated sequences of syllables learned in the Hebb procedure had
established stable, long-term lexical representations. The researchers concluded that this procedure is similar to the way children
implicitly learn words from sequence regularities in the phonological and orthographic input from their environment (see also
Mosse & Jarrold, 2008, for converging correlational evidence).
These research findings provide the rationale for the hypothesis
that the various reading and spelling problems experienced by
dyslexic individuals arise from impairments in learning serial
order information in LTM. Bogaerts, Szmalec, Hachmann, Page,
and Duyck (2013) present this SOLID (serial-order learning impairment in dyslexia) hypothesis as a novel integrative account of
dyslexia. The SOLID account of dyslexia can be linked to the
double deficit theory of dyslexia (Wolf & Bowers, 1999). According to this theory, processes underlying the ability to rapidly name
letters presented serially are a second source of reading problems,
apart from a phonological deficit. Interestingly, the serial format of
rapid naming (RAN) was found to correlate stronger with reading
than the discrete format (Bowers, 1995). The rapid integration of
visual and orthographic processes, tapped by RAN tasks, would be
crucially required in learning to read (Bowers & Ishaik, 2003). In
two experiments we attempted to replicate Szmalec et al.’s (2011)
finding of a Hebb learning deficit in dyslexics, including some
methodological improvements. Before presenting the experiments,
we discuss these methodological issues.
NO EVIDENCE FOR THE SOLID HYPOTHESIS OF DYSLEXIA
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
Methodological Issues and Improvements
The first issue concerns the well-established comorbidity of
developmental dyslexia and attention deficit disorders (ADD;
Araujo, 2012; Boada, Willcutt, & Pennington, 2012). Because the
Hebb learning task is very demanding on sustained and focused
attention, a differential Hebb effect is not necessarily the result of
a deficit in serial learning but may be attributed to the possible
comorbid attention problems of persons with dyslexia (see also
Wimmer’s analogous critique on the automatization deficit hypothesis of Nicolson & Fawcett, 1990; Wimmer, Mayringer, &
Raberger, 1999). Because this comorbidity does not imply that
each individual with dyslexia suffers from (minor) attentional
problems—indeed, many dyslexics do not show any attentional
problems at all—it is essential to avoid this confound. The most
sensitive way to control for this confounding factor without losing
statistical power is to analyze the complete data once including and
once without including attentional functioning as a covariate.
Another even more important methodological issue concerns the
statistical analysis performed by Szmalec et al. (2011). The slopes
of the regression lines based on performances on successive Hebb
and filler sequences for each individual participant were used as
the dependent variable in an analysis of variance (ANOVA). One
serious problem with this method is that slopes and intercepts are
often negatively correlated, especially when maximum performance has an upper bound. If the intercept (initial performance) is
already high, there is little room for improvement. This means that
the slope of a person who starts his sequence with a very good
score tends to be lower than the slope of a person who starts his
sequence with a very low score. In fact, the mean correlation
between the intercepts and slopes for the three Hebb conditions in
the Szmalec et al. (2011) study was substantial (r ⫽ –.54, p ⬍ .01),
as was also the case in our replication study (r ⫽ –.52, p ⬍ .01).
This means that slopes are poor measures and should be avoided,
because they inevitably miss a substantial portion of crucial variance in Hebb learning. In order to adequately capture the gradualness of Hebb learning over consecutive trials, we constructed a
new index based on a weighted sum of correct responses.
A third methodological improvement involves the use of state
trace analysis as a matching technique (see Van den Broeck &
Geudens, 2012). This method allows a direct comparison of Hebb
trial performance when performance on filler trials is equated
between the groups. Therefore, this method is more sensitive to
detect a specific Hebb learning deficit because by matching dyslexic and control subjects on filler performance, both groups are
matched on sequential learning without the crucial phase of consolidation in LTM. This method of equating performances on filler
trials is also necessary bearing in mind the often reported verbal
STM problems of poor readers. Indeed, if the aim of the study is
to compare performances on Hebb sequences with performances
on filler sequences, it is mandatory that performances on filler
trials be equated between groups.
A final methodological issue pertains to the scoring method that
Szmalec et al. (2011) used to evaluate Hebb learning. In this
method an item was scored as correct if and only if it was recalled
in its correct position. Hence, items that were recalled in a correct
serial order but in the wrong position were not counted. This
means, for example, that if a participant recalled all items of a
sequence in the correct order except the first one, his or her score
3
would be 0. So in fact, only the position of the items is considered,
not the serial order. We address this problem by using the McKelvie scoring method (McKelvie, 1987). In this method, both the
position and serial order of items are taken into account.
Experiment 1
In this first experiment Hebb learning was examined in adolescents by replicating the study by Szmalec et al. (2011). The results
were analyzed once using exactly the same method and dataanalytical procedures as applied in the Szmalec et al. (2011) study,
and once taking into account some methodological improvements.
Potential methodological enhancements were the use of a statistical control on attentional processes, the procedure of weighted sum
scores to detect Hebb learning, the method of state trace analysis
as an improved matching technique, and McKelvie’s scoring
method (McKelvie, 1987). Except for state trace analysis, which
has already been proven to offer a better match on baseline
performance, each of these adaptations was tested empirically
before they were adopted in the data analysis.
Method
Participants. A total of 59 higher education students participated in this study. Twenty-six students (17 females, nine males)
had an official diagnosis of dyslexia, and 33 students (25 females,
eight males) were IQ-matched control students without any reading problems. Dyslexic participants were diagnosed either by an
individual speech therapist or by a specialized center. The diagnoses were all based on three criteria that are used by the Stichting
Dyslexie Nederland (Netherlands Dyslexia Foundation; 2008): (a)
reading and/or spelling abilities significantly below the level of
performance expected for their age (i.e., below the 10th percentile); (b) resistance to instruction despite effective teaching; (c)
impairment that cannot be explained by extraneous factors, such as
sensory deficits. For further validation, two Dutch reading tests
that are diagnostic for dyslexia were administered. The first was
the One Minute Test (OMT; Brus & Voeten, 1973), a word reading
test in which participants are instructed to correctly read aloud as
many words as possible. The raw score is the number of correctly
read words after 1 min of reading. The second was the Klepel (Van
den Bos, Lutje Spelberg, Scheepsma, & de Vries, 1994), a nonword reading test in which participants are instructed to correctly
read aloud as many nonwords as possible within 2 min.
To match the dyslexic and the control group on IQ, a short-form
IQ measure was used (Turner 1997), including the Similarities,
Comprehension, Block Design and Picture Completion subtests of
the Wechsler Adult Intelligence Scale (Dutch version; 3rd ed.;
Wechsler, 1998). Our experimental procedure also included a test
of sustained attention and two questionnaires concerning attentional functioning. The test of sustained attention was the
Bourdon-Wiersma (van der Ven & Smit, 1989), a dot cancelation
test. The test consisted of a scoring form with 50 lines, each line
containing 25 groups of dots (each group consists of three, four, or
five dots in random order). Participants were instructed to cross out
all groups of four dots (targets). Each line contained eight targets
in a random order. The experimenter used a digital stopwatch to
measure the time participants required for each line. Afterward, the
number of correctly marked groups of four dots was determined
STAELS AND VAN DEN BROECK
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
4
the other in the verbal–auditory Hebb task. The use of both lists in
one of the two presentation modalities was counterbalanced across
participants. Thirty sequences of nine nonsense syllables were
presented serially in the middle of a computer screen to the
participants for immediate serial recall. The 30 sequences consisted of 20 nonrepeated sequences (filler sequences) and one
Hebb sequence that was repeated 10 times. Participants were not
informed that one sequence (the Hebb sequence) recurred every
third trial. At recall, all nine syllables and one question mark were
distributed randomly on the computer screen. Participants were
instructed to repeat the nine syllables in the correct serial order.
Participants were instructed to say “blank” for every omitted item.
The spoken output of the participants was written down by the
experimenter. According to our general strategy to maximize the
probability for Hebb learning to take place, we considered a
potential stimulus-specific order-effect as less important than reducing random error variance. Therefore, the same Hebb sequence
and filler sequences were used in the same order for all participants
receiving the same list of nonsense syllables (half of the subjects).
Verbal–auditory Hebb condition. In the verbal–auditory Hebb
condition, 30 sequences of nonsense syllables were presented
auditorily to the participants for immediate serial recall. The syllables were digitally recorded by a female speaker and presented
auditorily through speakers at a rate of one per second. Thirty
sequences consisted of 20 nonrepeated sequences (filler sequences) and one Hebb sequence that was repeated 10 times.
Participants were not informed that one sequence (the Hebb sequence) recurred every third trial. As in the previous condition, for
all participants receiving the same list of nonsense syllables, the
same Hebb sequence and filler sequences were presented in
the same order. At recall, the participants were instructed to repeat
the nine syllables in the correct serial order. Participants were
instructed to say “blank” for every omitted syllable. The spoken
output of the participants was written down by the experimenter.
Visuospatial Hebb condition. In the visuospatial Hebb condition, we used the dots task, a Corsi-like (Corsi, 1972) visuospatial
immediate serial recall task. We used sequences of nine black dots
randomly presented on a white background on a computer screen.
The dots were presented serially at a rate of one per second, with
each dot on a different location on the screen. As Hebb learning in
the Szmalec et al. (2011) study was particularly weak in the
visuospatial condition, we chose a relatively easy dot configuration
and the number of correct marked groups of four dots per second
was calculated. Two versions of a Dutch self-report questionnaire
(ZVAH; Baeyens, Van Dyck, Broothaerts, Danckaerts, & Kooij,
2012) to detect attention problems were used. Both versions contained the same questions. The first one consisted of questions
regarding the participants’ childhood; the second asked about
current attentional functioning. As can be seen in Table 1, the
group of dyslexic students and the control group differed significantly only on the two measures that are diagnostic for dyslexia, on
the test of sustained attention and on the self-report questionnaire
concerning attentional problems during childhood. These measures
clearly indicate that the clinically diagnosed dyslexic group also
suffered from (mild) attentional problems.
Design, procedure, and materials. Testing took place on an
individual basis in a quiet room at the participant’s school or home.
The experimental procedure consisted of two test phases. Each test
phase lasted approximately 1 hr. To prevent experimenter bias, test
assistants were extensively instructed to consider all collected data
as potentially interesting and to assume an empirical attitude that
sets each theoretical position on an equal footing. This is a standard procedure at our lab.
First phase. During the first testing phase the two reading
tests, the four IQ subtests of the WAIS–III, the attention test, and
the attention questionnaires were administered.
Second phase: Hebb tasks. In the second phase, the Hebb
procedure was carried out (Hebb, 1961). In this procedure, participants are asked to perform an immediate serial-recall task in
which one particular sequence is repeated every third trial. The
Hebb procedure we used was similar to that of Szmalec et al.
(2011) and was programmed in E-Prime (Version 1.2; Schneider,
Eschman, & Zuccolotto, 2002). The order of the three Hebb tasks
was counterbalanced across participants. We assessed awareness
of the repetition of Hebb sequences in a postexperimental interview (after the three Hebb tasks) that consisted of one question:
“What did you notice about the different sequences presented in
the three tasks?” For each task the answer of the participant was
scored as “yes” or “no” according to their awareness of the
sequences.
Verbal–visual Hebb condition. We adopted the two lists of
nine nonsense syllables constructed by Szmalec et al. (2011):
da-fi-ke-mo-pu-sa-ti-vo-zu and ba-du-ki-le-mu-so-to-vi-za. For
each individual, one list was used in the verbal–visual Hebb task,
Table 1
Characteristics of the Experimental and Control Group
Control
(n ⫽ 33)
Dyslexic
(n ⫽ 26)
Group
difference
Characteristics
M
SD
M
SD
t
p
Age in years
Word reading test
Nonword reading test
WAIS-III Picture Completion
WAIS-III Similarities
WAIS-III Block Design
WAIS-III Comprehension
Bourdon-Wiersma no. correct items/second
Self-report questionnaire: attention childhood
Self-report questionnaire: attention present
21.64
88.36
97.24
11.21
13.00
10.55
13.15
0.648
12.00
10.91
1.83
13.82
12.25
2.16
1.95
2.53
2.21
0.08
2.92
3.12
20.69
70.73
70.96
11.92
13.23
10.77
13.42
0.581
13.54
12.27
2.45
11.93
15.58
2.04
2.01
2.18
2.06
0.11
2.67
2.93
1.69
5.16
7.26
1.29
0.45
0.36
0.48
2.65
2.09
1.71
.10
.00
.00
.20
.66
.72
.63
.01
.04
.09
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
NO EVIDENCE FOR THE SOLID HYPOTHESIS OF DYSLEXIA
to enhance the probability to observe Hebb learning. Dot locations
and sequences were the same for all participants. As in the two
previous conditions, 20 filler sequences were nonrepeated sequences, and one Hebb sequence was repeated at every third trial.
After each sequence, all dots were shown on the screen, and the
participant used the mouse to indicate the correct order in which
the dots had occurred.
Scoring methods and empirical tests of sensitivity. Two scoring methods were adopted. First, we used the same scoring method
as Szmalec et al. (2011). In this scoring method, for every recalled
sequence a score was defined by counting every recalled item in
the same position as that in which it was initially presented. This
means that items that were recalled in a correct serial order but in
the wrong position were not counted. To address this problem, we
also used McKelvie’s scoring method. In this method, position and
serial order of items are taken into account (for details, see
McKelvie, 1987).
After adopting one of the two scoring methods, a measure of
Hebb learning over the 10 trials was required. As already explained, using the slopes of the learning curves is a dubious
method to detect Hebb learning. For that reason we constructed a
new index of Hebb learning that avoids the disadvantages of using
slopes. Of course, one could simply take the mean or summed
Hebb performance over the 10 trials, but by thus weighting each
trial equally, this method seems to miss the fact that learning
improves gradually over trials. To capture this idea of gradualness
and to take into account the empirical learning curves, we calculated a weighted sum for each individual by weighting the Hebb
score for each of the 10 Hebb trials, based on the predicted overall
difference between the filler score of a given series and the Hebb
score of the corresponding series. More specifically, for each
condition, we took the mean Hebb performance over participants
for each of the 10 Hebb trials and the mean filler performance of
each of the two filler trials preceding a Hebb trial, and then
calculated a linear or quadratic regression equation (whichever
fitted the data better) for the filler trials and for the Hebb trials.
Then, the differences between the predicted filler scores and Hebb
scores served as the weights for calculating the weighted sum.1
To determine which scoring method and index of Hebb learning
were to be used in our improved data analysis, we tested these
methods empirically on their criterion validity by checking how
well they correlated with the reading scores and with being dyslexic or not. From Table 2 it emerges that the method of using
slopes is far inferior to the method of the weighted sums, because
all correlations with the reading measures and with the diagnostic
group variable are considerably larger for the latter. As the correlations of the method of weighted sums using the McKelvie
scoring system were very similar and certainly not larger than
those of the classic scoring method, we proceed in our improved
data analysis with the method of weighted sums based on the
Szmalec scoring method. It is noteworthy that the substantial
correlations between the improved Hebb learning index and diagnostic category cannot be taken as evidence for a specific Hebb
learning deficit in dyslexia, as no attempt is made here to control
for filler performance.
In a final preliminary analysis we tested whether attentional
functioning could potentially be a confounding factor by inspecting the correlations of the three tests on attentional functioning
with the three measures of Hebb learning (the weighted sums for
5
Table 2
Criterion Validity of the Scoring Methods and Indices of Hebb
Learning: Correlations With Reading Measures and
Diagnostic Category
Scoring method
Szmalec
Slope (VV)
Weighted (VV)
Slope (VA)
Weighted (VA)
Slope (VS)
Weighted (VS)
McKelvie
Weighted (VV)
Weighted (VA)
Weighted (VS)
Diagnostic
category (dyslexic
Real word
Pseudoword
or control group) reading (OMT) reading (Klepel)
⫺.054
⫺.318ⴱ
⫺.304ⴱ
⫺.543ⴱⴱ
⫺.051
⫺.296ⴱ
⫺.058
.139
.196
.419ⴱⴱ
⫺.009
.230
.053
.138
.288ⴱ
.525ⴱⴱ
.007
.347ⴱⴱ
⫺.312ⴱ
⫺.525ⴱⴱ
⫺.252ⴱ
.134
.449ⴱⴱ
.148
.157
.536ⴱⴱ
.273ⴱ
Note. OMT ⫽ One Minute Test; VV ⫽ verbal-visual condition; VA ⫽
verbal-auditory condition; VS ⫽ visuospatial condition.
ⴱ
p ⬍ .05. ⴱⴱ p ⬍ .01.
the three conditions). Three out of nine correlations were statistically significant (ranging between .25 and .35). Although attentional functioning does not seem to play a major role in Hebb
learning, given that attentional functioning is also correlated with
diagnostic category (see Table 1), it seems warranted to analyze
the data once with and once without a statistical control for
attentional functioning.
Results
First, we present the mean filler and Hebb scores for the typical
and for the dyslexic readers over the three conditions, and compare
them with those of the Szmalec et al. (2011) study (see Table 3).2
We observed a clear Hebb learning effect in all three conditions for
control subjects as well as for the dyslexic participants. The
magnitude of this Hebb effect seems to be similar to the Szmalec
study, at least for the verbal–visual and for the verbal–auditory
conditions. For the visuospatial condition, however, the picture is
clearly different. In this case the Hebb effect in the Szmalec study
is clearly smaller for the control subjects than in our study, and
even nonexistent for the dyslexic subjects. Another notable observation is that although our dyslexic participants scored quite similar on the filler trials compared to their compeers in the Szmalec
study, they performed significantly inferior on these trials compared to the normal readers in our study. The difference between
the normal and dyslexic readers in the Szmalec study on the filler
trials was much smaller and even nonexistent in the visuospatial
1
For instance, the regression equation for filler trials in the verbal–
auditory condition was F-score ⫽ 3.64 ⫹ .08Trial, and for the Hebb trials
H-score ⫽ 4.26 ⫹ .31Trial. By taking the difference of both equations, the
weights-equation resulted: W ⫽ .62 ⫹ .23Trial. Because Hebb learning
cannot have taken place on the first Hebb trial, this weight was fixed to
zero. To obtain the weighted sum for each individual, the Hebb score on
each trial was multiplied by the appropriate weight and then summed.
2
Mean Hebb scores proved to be as sensitive to detect Hebb effects
compared to the weighted sum scores, as was shown by the correlations
with diagnostic category and the reading scores.
STAELS AND VAN DEN BROECK
6
Table 3
Means and Standard Errors for Mean Filler and Hebb Scores for Normal and Dyslexic Readers
Over Three Conditions
Current study
Fillers
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
Group and condition
Normal readers
Verbal-visual
Verbal-auditory
Visuospatial
Dyslexic readers
Verbal-visual
Verbal-auditory
Visuospatial
Effect sizes between groups (d=)
Verbal-visual
Verbal-auditory
Visuospatial
Szmalec et al. 2011
Hebb
Fillers
Hebb
M
SE
M
SE
M
SE
M
SE
4.44
4.35
3.93
.24
.26
.15
6.21
6.47
6.58
.28
.25
.25
3.79
2.86
3.43
.24
.23
.25
5.74
5.07
4.42
.34
.50
.26
3.38
2.74
3.63
.24
.20
.15
5.16
4.44
5.66
.27
.36
.30
3.40
2.50
3.55
.21
.17
.22
4.76
3.54
3.57
.28
.40
.25
0.82
1.26
0.37
0.69
1.23
0.63
0.43
0.44
⫺0.13
0.80
0.85
0.83
Note. For comparability the presented mean filler and Hebb scores are simple arithmetic means and not
weighted sum scores.
condition. Figure 1 shows the mean number of correctly recalled
Hebb and filler items as a function of trial number, group and
presentation modality.
In a first analysis we tested the hypothesis that dyslexic students
show impaired Hebb sequence learning compared to age- and
IQ-matched control students. Here, we analyze our data exactly as
Szmalec et al. (2011) did in their study. In this analysis, scores are
calculated using the classic scoring method. An item was scored as
correct if it was recalled in its correct serial position. To measure
the Hebb repetition effect, for each participant the gradient of the
regression line through points representing the performance on
successive Hebb repetitions was compared with the gradient of the
regression line for corresponding filler items (e.g., Page, Cumming, Norris, Hitch, & McNeil, 2006; Szmalec et al., 2011). The
gradient values were entered into a multivariate analysis of variance (MANOVA) with group (control vs. dyslexic) as a between
subjects independent variable, and sequence type (filler vs. Hebb)
and modality (verbal–visual vs. verbal–auditory vs. visuospatial)
as within-subject independent variables. We observed significant
main effects of sequence type, F(1, 57) ⫽ 126.13, ␩p2 ⫽ .689, p ⬍
.000; of modality, F(2, 56) ⫽ 4.586, ␩p2 ⫽ .141, p ⫽ .014; and of
group, F(1, 57) ⫽ 4.869, ␩p2 ⫽ .079, p ⫽ .031. Unlike Szmalec et
al. (2011), however, the crucial Group ⫻ Sequence type interaction effect was not significant, F(1, 57) ⫽ .128, ␩p2 ⫽ .002, p ⫽
.722, indicating a similar Hebb effect for the control and the
dyslexic group. Planned comparisons of the Group ⫻ Sequence
type interaction effect for the three modalities confirmed the
absence of a differential Hebb effect (for the verbal–visual condition: F(1, 57) ⫽ 0.143, ␩p2 ⫽ .003, p ⫽ .707; for the verbal–
auditory condition: F(1, 57) ⫽ 1.355, ␩p2 ⫽ .023, p ⫽ .249; for the
visuospatial condition: F(1, 57) ⫽ 0.0003, ␩p2 ⫽ .000, p ⫽ .987).
In our second analysis, the previously explained methodological
weaknesses are addressed. First, a weighted sum was used as an
improved index of Hebb learning and state trace analysis was
adopted to be able to compare directly the performance on Hebb
trials to the performance on filler trials. Second, the analyses were
performed once without statistically controlling for attentional
functioning, and once with this control.
In state trace analysis (STA), performance on Hebb sequences is
regressed separately on performance on filler sequences for the
dyslexic group and the control group. We tested whether a single
line is suitable to explain the data (the null model not including
reading group) or whether two different lines (one for dyslexic
students and one for control students) are needed to describe the
relation between Hebb sequence performance and filler sequence
performance (the full model). If a single line fitted the data, this
would imply that the relation between Hebb sequence performance
and filler sequence performance is not affected by dyslexia. If, on
the other hand, two lines fitted our data better, and the one for
dyslexic students were situated lower than the one for control
students, this would be direct evidence for a specific Hebb learning
deficit in dyslexic students.
In this analysis, the sum of the filler scores and the weighted
sum of the Hebb scores were plotted against each other for each
participant (see Figure 2). At each performance level on filler trials
both groups are equal and can be compared directly on Hebb
performance. First, we removed the bivariate outliers based on the
Mahalanobis distances by excluding the data-points with distances
(D2) above a cutoff value corresponding to the 97.5% quantile of
the chi-square distribution (Rousseeuw & Van Zomeren, 1990).
This procedure resulted in the removal of one outlier in the
verbal–visual condition (a subject of the control group), one outlier
in the verbal–auditory condition (one subject of the control group),
and two outliers in the visuospatial condition (one dyslexic subject
and one of the control group). Importantly, in order to validly
compare Hebb performance between groups, STA requires the
data points to overlap on the filler scores (Prince, Brown, &
Heathcote, 2012; Van den Broeck, Geudens, & van den Bos,
2010); otherwise participants in one group are compared with
statistically extrapolated but nonexistent participants in another
group, resulting in meaningless conclusions. For example, when
no dyslexic readers are found in the upper tail of the distribution of
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
NO EVIDENCE FOR THE SOLID HYPOTHESIS OF DYSLEXIA
7
Figure 1. Mean number of correctly recalled filler and Hebb items in the different conditions, for both typical
and dyslexic readers. Accuracy values for filler trials represent the average of the two filler trials preceding each
Hebb repetition.
filler scores, the normal readers falling in that region are compared
with statistically predicted scores based on the performance of
dyslexic readers with lower filler performance. To avoid this
problem, the statistical analyses are exclusively based on the
overlapping part of the filler distributions. This procedure resulted
in the removal of nine additional subjects in the verbal–visual
condition (three dyslexic and six control subjects), 13 subjects in
the verbal–auditory condition (one dyslexic and 12 control subjects), and 11 subjects in the visuospatial condition (four dyslexics
and seven control subjects). Although no mechanical procedure
exists to determine the overlapping distributions, in the case of the
verbal–auditory and visuospatial conditions the overlapping regions on the filler scores were clearly discernable. Only for the
verbal–visual condition it was less clear which data points to retain
and which to remove. However, as we report forthwith, the statistical conclusions for this condition were quite similar if we analyzed the data without removing any subject.
Then, we tested in a hierarchical regression analysis whether
group contributed significantly to Hebb performance after including filler performance in the regression equation. This analysis
Figure 2. A: State-trace plot for the verbal–visual condition for typical readers and dyslexic readers. B: State-trace plot for the verbal–auditory condition
for typical readers and dyslexic readers. C: State-trace plot for the visuospatial condition for typical readers and dyslexic readers.
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
8
STAELS AND VAN DEN BROECK
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
NO EVIDENCE FOR THE SOLID HYPOTHESIS OF DYSLEXIA
showed that adding group as a predictor did not significantly
improve fit in the verbal–visual task (R2 null model ⫽ .340; ⌬R2 ⫽
.001; F change [1,46] ⫽ .103; p ⫽ .750) and in the visuospatial
task (R2 null model ⫽ .121; ⌬R2 ⫽ .002; F change [1,43] ⫽ .113;
p ⫽ .738). In the verbal–auditory task a significant group effect
was found (R2 null model ⫽ .464; ⌬R2 ⫽ .050; F change [1,42] ⫽
4.280; p ⫽ .045), Hence, the null hypothesis—that the state trace
curves for the typical and for the disabled readers do not differ—
could not be rejected for two out of three conditions. This means
that if there was in reality no difference between the curves of the
two groups (if H0 is true), the probability of finding a difference as
large as or even larger than in our sample is .75 for the verbal–
visual task, .04 for the verbal–auditory task, and .74 for the
visuospatial task. As STA entails that the null hypothesis is in fact
a substantive hypothesis, these numbers do not seem really convincing. Note that a nonsignificant result due to a lack of power
must not be confused with support for the null hypothesis. What
we really want to know is the probability that the null hypothesis
or the alternative hypothesis is true given the observed data (the
inverse probability). To this end, a Bayesian analysis was performed in which the probability of the null model was compared to
the probability of the full model given the data and given the
assumption that no model was preferred above the other. For this
comparison the Bayesian information criterion (BIC) was calculated for both models. The BIC has been proposed by Raftery
(1996) as an index to assess the overall fit of a model and allows
a comparison of models (see also Long, 1997). Given that BIC
assesses whether the model fits the data sufficiently well to justify
the number of parameters that are used, the model with the lowest
BIC is the best fitting, yet parsimonious model. The BIC values
indicated that the null model fitted the data best for two tasks
(BICnull ⫽ 468.15 and BICfull ⫽ 471.93 for the verbal–visual task,
BICnull ⫽ 434.77 and BICfull ⫽ 434.21 for the verbal–auditory
task, BICnull ⫽ 448.82 and BICfull ⫽ 452.53 for the visuospatial
task). Based on the differences between these BIC values, the
Bayesian factors could be calculated (Kass & Raftery, 1995). The
Bayesian factor is equal to the posterior odds in favor of the most
likely hypothesis. The posterior odds of the null model (MN)
relative to the full model (MF) equal
Pr(M N ⁄ Observed Data)
Pr(M F ⁄ Observed Data)
.
The Bayesian factors favoring the null model were 3.78 and
3.71 for the verbal–visual and the visuospatial tasks, respectively,
whereas the Bayesian factor (.56) favored the full model for the
verbal–auditory task (the Bayesian factor for the verbal–visual task
when all data points were included, including the nonoverlapping,
was 2.88 favoring the null model). According to the criteria
proposed by Kass and Raftery (1995), the data provided substantial
evidence for the null hypothesis for the verbal–visual task and for
the visuospatial task, whereas the data for the verbal–auditory task
gave barely more support for the full hypothesis than for the null
hypothesis. Next we tested whether the previous conclusions were
affected by including the variables indicative of attentional functioning into the first step of the hierarchical regression analysis. As
the statistical tests showed the same pattern of results—the visualverbal and visuospatial conditions showing no significant groupeffect while the verbal–auditory condition did show a significant
9
between-group difference—we only present here the Bayesian
factors. The Bayesian factors favoring the null model were 3.71
and 3.18 for the verbal–visual and the visuospatial tasks, respectively. Although the Bayesian factor favoring the full model for the
verbal–auditory task was somewhat higher than without controlling for attentional functioning (1.64), this support is considered
weak according to Kass and Raftery (1995).
As Szmalec et al.’s (2011) hypothesis is explicitly formulated as
modality nonspecific, a more powerful analysis of their hypothesis
is possible by integrating all data from the three tasks into one
multilevel analysis. As a result of the nested nature of the design
(within-person conditions are nested under the between-persons
level), within-person variance and between-persons variance
should be modeled appropriately. Multilevel modeling (Raudenbush & Bryk, 2002) was used to obtain state trace functions for
each individual, to obtain average parameters (and their variances)
across participants, and to compare the parameters from these
functions in the reading disabled and typical reading samples. This
analytic approach first required the specification of a conditionlevel model (within subject), which represents the performance for
each Hebb condition as a function of the performance from the
corresponding filler condition for each individual. Second, personlevel models (between subjects) were specified representing each
regression parameter (intercept and slope) from the condition-level
model as a dependent variable. In the null model these parameters
were written as a function of overall means and individual unique
effects. Additionally, group effects on the regression parameters
were included in the full model.
Filler scores and Hebb scores were first transformed to z scores
to make them comparable over conditions. Next, we removed the
bivariate outliers based on the Mahalanobis distances by excluding
the data points with distances (D2) above a cutoff value corresponding to the 97.5% quantile of the chi-square distribution
(Rousseeuw & Van Zomeren, 1990). This procedure resulted in
the removal of five data points. Additionally, 25 data points, lying
outside the overlapping region, were removed. As a consequence,
147 out of 177 (3 ⫻ 59) data points were included in the analysis.
Fitting the full model using MPlus 7 (Muthén & Muthén, 1998/
2012) produced the following parameter estimates: For the average
intercept of the typical readers, ␥00 ⫽ .108 (SE ⫽ 0.093, p ⫽ .244);
for the average slope of the typical readers, ␥10 ⫽ .631 (SE ⫽
0.123, p ⬍ .000); for the difference in intercepts between the
reading groups, ␥01 ⫽ –.176 (SE ⫽ 0.142, p ⫽ .214); and for the
difference in slopes between the reading groups, ␥11 ⫽ .092 (SE ⫽
0.175, p ⫽ .601). These results show that disabled readers demonstrated a slight but nonsignificant Hebb learning deficit (a lower
intercept). However, the null hypothesis—that the state trace
curves for the typical and for the disabled readers do not differ—
could not be rejected. As evidenced by a likelihood ratio test, the
full model was not significantly different from the null model,
␹2(2) ⫽ 2.152, p ⫽ .341. The BIC value for the null model was
351.10 and for the full model 358.93, resulting in a Bayesian factor
favoring the null model of 7.83. This implies that the null model
was about 8 times as likely as the full model, a number that can be
interpreted as substantial evidence (Kass & Raftery, 1995). The
parameter estimate for the average intercept of this null model was
␥00 ⫽ .021 (SE ⫽ 0.069, p ⫽ .768), and for the average slope,
␥10 ⫽ .704 (SE ⫽ 0.087, p ⬍ .000). Thus, for all participants an
increase in performance of one filler item, expressed in z scores,
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
10
STAELS AND VAN DEN BROECK
was associated with an increase in performance of .71 Hebb items.
In conclusion, these results show that no different Hebb learning
was observed between the two groups across the three modalities.
Nearly 80% of the participants indicated that they were aware of
the repeating Hebb sequence (79.7% in the verbal–visual condition, 78.0% in the verbal–auditory condition, and 79.9% in the
visuospatial condition). However, participants of the control group
detected the Hebb sequences more frequently than the dyslexics
(about 89% vs. 65%). Importantly, Hebb performance was better
when the sequence was detected, as indicated by the positive
biserial correlations between mean Hebb score performance and
awareness: r ⫽ .26, p ⫽ .045 for the verbal–visual condition, r ⫽
.30, p ⫽ .022 for the verbal–auditory condition, and r ⫽ .52, p ⬍
.000 for the visuospatial condition. Nevertheless, the Hebb learning effect was still present for the participants who were unable to
detect the sequences (the main effect of sequence type was statistically significant for each modality, all ps ⬍ .02). Finally, as an
alternative test of a differential Hebb effect, we tested if reading
scores, as a continuous measure of reading ability, were correlated
with Hebb scores after the influence of the filler-scores was
partialed out from the Hebb-scores: All ps were nonsignificant.
Discussion
The purpose of Experiment 1 was to try to replicate a differential Hebb-learning effect in disabled versus normal adolescent
readers, as reported by Szmalec et al. (2011). First, as our results
showed, we observed a clear Hebb-learning effect in all three
conditions for control subjects as well as for the dyslexic participants, implying that Hebb learning was confined neither to specific
stimulus material nor to specific order effects. Thus, it appears that
our experimental procedures were more sensitive in detecting
Hebb learning than those in the Szmalec study, since in the latter
the Hebb effect in the visuospatial condition was smaller for the
normal readers and even nonexistent in the dyslexic readers (see
Table 3). This last observation is atypical, since filler performance in their visuospatial condition was even somewhat better
for the dyslexics than for the normal readers. Nevertheless, we
could not identify a specific Hebb learning deficit in dyslexics
in any of the three modalities (verbal–visual, verbal–auditory, and
visuospatial), when applying the data-analytical procedure used by
Szmalec et al. However, we demonstrated empirically that the use
of slopes is a poor technique to detect Hebb learning and might be
the reason why we couldn’t replicate a dyslexic Hebb learning
deficit. Therefore, we developed a more sensitive method, the
weighted sums, to measure Hebb learning. After adopting this
methodological improvement and after equating both groups more
precisely on filler performance with STA, we found a significant
group effect for the verbal–auditory condition, but not for the two
other conditions. Moreover, a Bayesian analysis revealed that even
this effect in the verbal–auditory condition could not be interpreted
as providing more support for a dyslexic serial order learning
impairment than for the hypothesis that no such deficit exists. In
the two other conditions the evidence clearly supported the null
hypothesis. This conclusion was further strengthened by a multilevel analysis, generating more statistical power, in which the three
conditions are appropriately nested under the participant level.
Although our methodological perfections are arguably important (especially STA and the use of a weighted sum as an index of
Hebb learning), the reason for failing to replicate the Hebb learning deficit seems not to be related to the method adopted. Moreover, a reanalysis of the data of Szmalec et al. (2011), using STA
with mean filler and mean Hebb scores, revealed marginal
between-groups effects for the verbal–visual and the verbal–
auditory conditions (p values of the F-change statistics were .065
and .057, respectively), whereas the visuospatial condition showed
a statistically significant effect (p ⫽ .017). Thus, also in Szmalec’s
study, the main culprit seems not to be the method adopted but the
data that are genuinely different. We can only speculate on the
reasons for these differing results. One plausible possibility is that
adolescent or adult dyslexics are not naive subjects; they are
reported to sometimes react in a sophisticated or strategic manner.
More specifically, Harrison, Edwards, and Parker (2008) provided
evidence that postsecondary-level students may be motivated to
exaggerate or magnify their symptoms. One could speculate that
students in the Szmalec study perceived a direct link between their
performance in the experiment and the justification to get compensatory facilities, as these facilities directly depended on the
diagnosis given by the diagnostic center involved. In our study,
there was no such link at all. One way to solve this problem is to
distinguish gross achievement measures from more subtle effects
that are little known to the participants (Lindstrom, Coleman,
Thomassin, Southall, & Lindstrom, 2011). Another, more obvious
way to avoid these complications and probably other indirect
effects of carrying a history of reading problems is to replicate the
study in more naive subjects (i.e., children). An additional reason
for a replication study in children is that, from a theoretical
viewpoint, exactly the same predictions should be made for this
population. As a matter of fact, such a study can be viewed as a
more direct test of the SOLID hypothesis stating that the Hebb
repetition effect as a laboratory analogue of novel word learning is
disturbed in dyslexic individuals. Although Szmalec et al. (2011)
indicated that they were the first to study the Hebb repetition effect
in relation to dyslexia, Gould and Glencross (1990) used the Hebb
paradigm in children. In their study, the group of disabled readers
scored significantly lower on the repeated digits task, but on the
repeated blocks task the disabled group performed as well as
normal readers. The authors concluded that their data do not
support a general deficit in serial organization.
Another possible explanation for the divergent findings is that
our dyslexic subjects might be less severely impaired than those in
Szmalec et al. (2011) and are merely garden-variety poor readers
(Gough & Tunmer, 1986; Stanovich, 1988). This possibility is
clearly refuted by the data, as our dyslexics’ intellectual functioning was above average (see Table 1) and their mean word reading
score was even lower than that of the dyslexic group in the
Szmalec study (70.73 vs. 75.25; the mean nonword reading score
was 70.96 in our study vs. 65.44 in theirs). In the same vein,
the filler scores of our dyslexic participants were very similar to
those of the dyslexics in the Szmalec study. Remarkably however,
the filler scores of our dyslexics were clearly inferior to those
of the normal readers, whereas in the Szmalec study the difference
on filler trials between the normal and dyslexic readers was much
smaller and even nonexistent in the visuospatial condition. The
pattern in our results could be seen as a confirmation of the
alternative hypothesis that dyslexics suffer from a STM serial
order learning deficit, rather than a deficit of the transfer from
NO EVIDENCE FOR THE SOLID HYPOTHESIS OF DYSLEXIA
evaluated sustained attention. In the Score! subtest, children have
to count the number of “scoring” sounds they hear on a tape as if
they were keeping the score on a computer game across several
trials. In the Score DT subtest, the same attention task of counting
scoring sounds is combined with another task involving listening
for an animal name that occurs at some stage during a spoken news
report. A Dutch ADHD questionnaire (AVL; Scholte & Van der
Ploeg, 2005) was used to asses attentional functioning. The questionnaire was completed by the parents and the teacher of the
participant. Table 4 shows that the group of dyslexic children and
the control group differed on the two measures that are diagnostic
for dyslexia and on all attention tests and questionnaires.
Design, procedure and materials. Testing took place on an
individual basis in a quiet classroom at the participant’s school.
The experimental procedure consisted of two test phases. Each test
phase lasted approximately 1 hr.
First phase. During the first testing phase the two reading
tests, the four IQ subtests of the WISC–III, and the attention tests
were administered. The two attention questionnaires were completed by the teacher and the parents of each participant.
Second phase: Hebb tasks. In the second phase, the Hebb
procedure was implemented (Hebb, 1961). The Hebb procedure of
Szmalec et al. (2011) was adjusted to our younger participants. To
minimize verbal processes, we administered two nonverbal tasks,
a visuospatial Corsi blocks task and a visual forms sequences task.
As in Gould and Glencross (1990), a digit sequences task was also
run. The three Hebb tasks were administered in the same order for
all participants. The sequences we used in this study were sequences of only seven items. Similar to our first study, we presented 30 sequences; 20 of them were nonrepeated sequences
(filler sequences) and one Hebb sequence was repeated 10 times.
For all three Hebb tasks, all participants received the same list of
sequences, to reduce random error variance. Participants were not
informed that one sequence (the Hebb sequence) recurred every
third trial. The children’s awareness of the repetition of Hebb
sequences was not assessed. Although the issue of awareness is
certainly relevant, it was felt that self-report would not be very
reliable in this sample.
Digit sequences. For the first Hebb task we used a verbal–
auditory task with digit sequences. Thirty sequences of seven
digits were presented serially to the participants for immediate
STM to LTM (see Hachmann et al., 2014; Perez, Majerus, Mahot,
& Poncelet, 2012; but see Staels & Van den Broeck, 2013).
The most plausible explanation for the conflicting findings,
however, is that the data of both studies are valid and dependable
but that the results of the Szmalec et al. (2011) study on the
verbal–visual and verbal–auditory conditions are not in contradiction with the null hypothesis, whereas their visuospatial condition
was not sensitive enough to detect Hebb learning in all subjects.
We return to this possibility in the final discussion.
Experiment 2
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
11
In the second experiment, Hebb learning was examined in
children by adapting the Szmalec procedure (Szmalec et al., 2011).
As in the first experiment, results are analyzed once using exactly
the same method and data-analytical procedures as applied in the
Szmalec et al. (2011) study, and once taking into account some
methodological improvements.
Method
Participants. Fifty-seven fifth and sixth grade children participated in this study. Their age ranged from 9 years 0 months to
12 years 0 months, with a mean age of 10 years 7 months.
Twenty-five children (seven girls, 18 boys) were officially diagnosed with dyslexia; the other 32 IQ-matched children (25 girls,
seven boys) never showed any reading problems. As in Experiment 1, dyslexic participants were diagnosed by an individual
speech therapist. The diagnoses were all based on three criteria
which are used by the Stichting Dyslexie Nederland (Netherland
Dyslexia Foundation; 2008). For further validation two Dutch
reading tests that are diagnostic for dyslexia were administered:
the One Minute Test (OMT; Brus & Voeten, 1973) and the Klepel
(Van den Bos et al., 1994). To match the dyslexic and the control
group on IQ, a short-form IQ measure was used (Turner 1997),
including the Similarities, Comprehension, Block Design and Picture Completion subtests of the Wechsler Intelligence Scale for
Children III (Dutch version; Wechsler, 2005).
We included two tests and one questionnaire to examine attentional functioning. The tests were Score! and Score DT, two
subtests of the Test of Everyday Attention for Children (TEA–Ch;
Manly, Robertson, Anderson, & Nimmo-Smith, 1999). Both tests
Table 4
Characteristics of the Experimental and Control Groups
Controls
(n ⫽ 32)
Dyslexics
(n ⫽ 25)
Group
difference
Characteristics
M
SD
M
SD
t
p
Age (years)
Word reading test
Nonword reading test
WISC-III Picture Completion
WISC-III Similarities
WISC-III Block Design
WISC-III Comprehension
TEA-Ch Score!
TEA-Ch Score DT
AVL parents (ADHD questionnaire)
AVL teacher (ADHD questionnaire)
10.41
66.63
63.34
8.38
10.84
8.84
9.53
8.69
10.59
3.44
2.78
0.62
13.72
16.71
2.51
3.18
2.42
3.55
2.22
3.05
2.12
2.03
10.76
47.48
38.52
8.68
10.84
8.84
9.44
7.44
8.56
4.92
5.04
0.72
9.76
9.38
2.63
2.94
2.82
3.31
2.16
3.92
2.86
2.89
1.99
5.90
6.65
0.45
0.00
0.01
0.10
2.13
2.21
2.25
3.46
.05
.00
.00
.66
1.00
1.00
.92
.04
.03
.03
.00
STAELS AND VAN DEN BROECK
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
12
serial recall. The digits were digitally recorded by a female speaker
and presented auditorily through speakers at a rate of one per
second. At recall, the participants were instructed to repeat the
seven digits in the correct serial order. Participants were instructed
to say “blank” for every omitted item. The spoken output of the
participants was written down by the experimenter. The task was
programmed in Microsoft Office PowerPoint 2007.
Corsi blocks sequences. The second task was a Corsi block
tapping task, a visuospatial task (Corsi, 1972). For this task, seven
blocks (3 cm ⫻ 3 cm ⫻ 3 cm) were randomly glued to a solid
piece of cardboard (30 cm ⫻ 42 cm; see Figure 3). Participants
were instructed to repeat 30 sequences of seven taps on a set of
seven blocks that were presented to them. Sequences of blocks
were tapped at approximately one block/second. Participants were
instructed to say “blank” for every omitted item. The sequence
pointed out by the participants was written down by the experimenter.
Visual form sequences. The third task was also a visual task.
We used sequences of visual forms without any meaning to prevent participants from verbalizing the items (see Figure 4). The
visual forms were selected from the Microsoft Word 2007 symbol
database. Thirty sequences of seven visual forms were presented
serially in the middle of a computer screen to the participants for
immediate serial recall. At recall, all seven visual forms were
presented randomly on the computer screen. To avoid spatial
learning of the response, on each trial another random arrangement
of the seven forms was presented. Participants were instructed to
point out the seven forms in the correct serial order. Participants
were instructed to say “blank” for every omitted visual form. The
sequence pointed out by the participants was checked by the
experimenter on a sheet of paper with all forms listed. The task
was programmed in Microsoft Office PowerPoint 2007.
Scoring methods and empirical tests of their sensitivity. As
in Experiment 1, the scoring method and index of Hebb learning
were empirically tested on their criterion validity by checking how
well they correlated with the reading scores and with being dyslexic or not. Although the correlations of Hebb learning with the
diagnostic category were smaller in children than in adolescents, as
in Experiment 1 the method of using slopes is far inferior to the
method of the weighted sums because all correlations with the
reading measures and with the diagnostic group variable are considerably larger for the weighted sums (some correlations of the
slopes are even in the wrong direction; see Table 5). As the
Figure 4. Visual forms sequences task.
correlations of the method of weighted sums using the McKelvie
scoring system were very similar and certainly not larger than
those of the classic scoring method, we proceed in our improved
data analysis with the method of weighted sums based on the
Szmalec scoring method.
We tested whether attentional functioning could potentially be a
confounding factor by inspecting the correlations of the three tests
on attentional functioning with the three measures of Hebb learning (the weighted sums for the three conditions). Three out of nine
correlations were statistically significant (ranging between .27 and
.36). Although attentional functioning does not seem to play a
major role in Hebb learning, given that attentional functioning is
also correlated with diagnostic category (see Table 4), it seems
warranted to analyze the data once with and once without statistically controlling for attentional functioning.
Results
First, we present the mean filler and Hebb scores for the typical
and for the dyslexic readers over the three conditions (see Table 6).
Although the three tasks differed in degree of difficulty (the visual
forms sequences task was clearly the most difficult), Hebb learning
appears to have occurred in all three tasks (for the visual forms
sequences task a one-tailed paired t test resulted in p ⫽ .051).
Figure 5 shows the mean number of correctly recalled Hebb and
filler items as a function of group and presentation modality.
In the first analysis we tested the hypothesis that dyslexic children
show impaired Hebb sequence learning compared to age- and IQmatched children. As in our first experiment, the data are analyzed
precisely as in Szmalec et al. (2011). In this analysis, scores are
calculated using the first scoring method and slopes are used as an
Table 5
Criterion Validity of the Scoring Methods and Indices of Hebb
Learning: Correlations With Reading Measures and
Diagnostic Category
Scoring method
Szmalec
Slope (D)
Weighted (D)
Slope (B)
Weighted (B)
Slope (VF)
Weighted (VF)
McKelvie
Weighted (D)
Weighted (B)
Weighted (VF)
Figure 3.
Display of Corsi blocks task.
Diagnostic
category (dyslexic
or control group)
Real word
reading (OMT)
Pseudoword
reading
(Klepel)
⫺.051
⫺.161
⫺.015
⫺.057
⫺.055
⫺.199
⫺.228
.124
⫺.074
.112
.118
.342ⴱⴱ
⫺.127
.273ⴱ
.014
.255ⴱ
.200
.428ⴱⴱ
⫺.139
⫺.080
⫺.137
.143
.129
.346ⴱⴱ
.075
.076
.391ⴱⴱ
Note. OMT ⫽ One Minute Test; D ⫽ digit sequences condition; B ⫽
Corsi blocks condition; VF ⫽ visual forms condition.
ⴱ
p ⬍ .05. ⴱⴱ p ⬍ .01.
NO EVIDENCE FOR THE SOLID HYPOTHESIS OF DYSLEXIA
Table 6
Means and Standard Errors for Mean Filler and Hebb Scores
for Normal and Dyslexic Readers Over Three Conditions
Fillers
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
Groups and conditions
Normal readers
Digits
Blocks
Visual forms
Dyslexic readers
Digits
Blocks
Visual forms
Effect sizes between groups (d=)
Digits
Blocks
Visual forms
Hebb
M
SE
M
SE
2.98
3.86
2.13
.18
.18
.12
3.64
5.03
2.31
.20
.20
.18
2.61
3.81
1.78
.17
.17
.09
3.28
4.94
1.88
.23
.23
.14
0.40
0.05
0.62
0.32
0.09
0.51
index of Hebb learning. The slopes were entered into a MANOVA
with group (control vs. dyslexic), sequence type (filler vs. Hebb), and
modality (digit sequences, Corsi block sequences, and visual form
sequences) as the independent variables. We observed a significant
main effect of sequence type, F(1, 55) ⫽ 45.431, ␩p2 ⫽ .452, p ⬍ .000,
but we observed no significant main effect of group, F(1, 55) ⫽ .904,
␩p2 ⫽ .016, p ⫽ .346, or modality, F(2, 54) ⫽ 2.502, ␩p2 ⫽ .085, p ⫽
.091. The crucial Group ⫻ Sequence type interaction effect was also
not significant, F(1, 55) ⫽ .087, ␩p2 ⫽ .002, p ⫽ .769, indicating a
similar Hebb effect for the control and the dyslexic group. Planned
comparisons of the Group ⫻ Sequence type interaction effect for the
three modalities confirmed the absence of a differential Hebb effect
(for digit sequences: F(1, 55) ⫽ 0.117, ␩p2 ⫽ .002, p ⫽ .733; for Corsi
block sequences: F(1, 55) ⫽ 0.025, ␩p2 ⫽ .000, p ⫽ .874; for visual
form sequences: F(1, 55) ⫽ 0.537, ␩p2 ⫽ .010, p ⫽ .467).
In our second analysis, the methodological improvements were
included. First, a weighted sum was used as an improved index of
Hebb learning and state trace analysis was adopted to be able to
compare directly the performance on Hebb trials to the performance on filler trials. Second, the analyses were performed once
without statistically controlling for attentional functioning, and
once with this control. In these analyses, for each participant, the
sum of the filler scores and the weighted sum of the Hebb scores
were plotted against each other (see Figure 6).
First, we removed the bivariate outliers based on the Mahalanobis distances by excluding the data-points with distances (D2)
above a cutoff value corresponding to the 97.5% quantile of the
chi-square distribution. This procedure resulted in the removal of
two outliers in the digits condition (both in the control group), two
outliers in the Corsi blocks condition (both in the dyslexic group),
and two outliers in the visual forms condition (both in the control
group). As in Experiment 1, the statistical analyses are exclusively
based on the overlapping part of the filler distributions. This
procedure resulted in the additional removal of two subjects in the
visual forms condition (two control subjects) and none in the other
two conditions.
Then, we tested in a hierarchical regression analysis whether group
contributed significantly to Hebb performance after including filler
performance in the regression equation. This analysis showed that
13
adding group as a predictor did not significantly improve fit in the
digit sequences task (R2 null model ⫽ .465; ⌬R2 ⫽ .000; F change
[1,52] ⫽ .003; p ⫽ .955), in the Corsi block sequences task (R2 null
model ⫽ .388; ⌬R2 ⫽ .000; F change [1,52] ⫽ .025; p ⫽ .875), or in
the visual form sequences task (R2 null model ⫽ .291; ⌬R2 ⫽ .001;
F change [1,52] ⫽ .038; p ⫽ .845). As can be inferred from these
results, no different Hebb learning was observed between the two
groups in all three modalities. As in Experiment 1, a Bayesian analysis
was performed in which the probability of the null model was compared to the probability of the full model given the data and given the
assumption that no model was preferred above the other. For this
comparison, BICs were calculated for both models. The BIC values
indicated that the null model fitted the data best for all tasks (BICnull ⫽ 363.73 and BICfull ⫽ 367.73 for the digit task, BICnull ⫽
374.26 and BICfull ⫽ 378.17 for the Corsi blocks
task, BICnull ⫽ 405.52 and BICfull ⫽ 409.50 for the visual form task).
The Bayesian factors all favored the null model: 4.00, 3.91, and 3.98
for the digit sequences, the Corsi block sequences and the visual form
sequences tasks, respectively. According to Kass and Raftery’s (1995)
criteria, the data provided substantial evidence for the null hypothesis
for the three tasks.
Next we tested whether the previous conclusions were altered by
including the variables indicative of attentional functioning into the
first step of the hierarchical regression analysis. As the statistical tests
showed the same pattern of results, we only present here the Bayesian
factors. The Bayesian factors favoring the null model were 3.93, 3.95,
and 3.82 for the digits task, the Corsi blocks task, and the visual forms
task, respectively. Clearly, these results were virtually the same as
without controlling for attentional functioning.
As in Experiment 1, we integrated all data in one multilevel
analysis to obtain more power. filler scores and Hebb scores were first
transformed to z scores to make them comparable over conditions.
Next, we removed the bivariate outliers based on the Mahalanobis
distances by excluding the data points with distances (D2) above a
cutoff value corresponding to the 97.5% quantile of the chi-square
distribution. This procedure resulted in the removal of six data points.
This time, the data points of both groups nicely overlapped for the
filler scores. As a consequence, 165 out of 171 (3 ⫻ 57) data points
were included in the analysis. Fitting a full model, including group as
a predictor variable in the person level model, MPlus 7 produced the
following parameter estimates: for the average intercept of the typical
readers, ␥00 ⫽ .011 (SE ⫽ 0.078, p ⫽ .885); for the average slope of
the typical readers, ␥10 ⫽ .590 (SE ⫽ 0.081, p ⫽ .000); for the
difference in intercepts between the reading groups, ␥01 ⫽ –.012
(SE ⫽ 0.117, p ⫽ .92); and for the difference in slopes between the
reading groups, ␥11 ⫽ .022 (SE ⫽ 0.137, p ⫽ .873). These results
show that the null hypothesis—that the state trace curves for the
typical and for the disabled readers do not differ— could not be
rejected. As evidenced by a likelihood ratio test, the full model was
not significantly different from the null model, ␹2(2) ⫽ 0.04, p ⫽
.980. The BIC value for the null model was 389.14 and for the full
model 399.31, resulting in a Bayesian factor favoring the null model
of 10.17. This implies that the null model was about 10 times as likely
as the full model, a number that can be interpreted as substantial
evidence (Kass & Raftery, 1995). The parameter estimate for the
average intercept of this null model was ␥00 ⫽ .005 (SE ⫽ 0.058, p ⫽
.932), and for the average slope, ␥10 ⫽ .599 (SE ⫽ 0.065, p ⬍ .000).
Thus, for all participants an increase in performance of one filler item,
expressed in z scores, was associated with an increase in performance
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
14
STAELS AND VAN DEN BROECK
Figure 5. Mean number of correctly recalled filler and Hebb items in the different conditions, for both typical
and dyslexic readers. Accuracy values for filler trials represent the average of the two filler trials preceding each
Hebb repetition.
of .60 Hebb items. Summarized, these results show that no different
Hebb learning was observed between the two groups across the three
modalities. Before discussing these results, we present an aggregated
STA that bears on the relationship between the learning of filler
sequences and Hebb sequences and their underlying memory mechanisms.
Underlying dimensionality of the memory processes
involved. Although we used STA as an equivalence or matching
technique (see Van den Broeck & Geudens, 2012) to detect differential Hebb-learning in Experiments 1 and 2, STA as originally
developed by Bamber (1979) has recently become an important
instrument in detecting underlying causality in cognitive and memory research through the seminal work of Geoffrey Loftus (Loftus,
Oberg, & Dillon, 2004; see also Dunn & James, 2003; Newell &
Dunn, 2008; Prince et al., 2012). The fundamental question STA
seeks to answer is whether the relationship between a set of causal
variables and a set of indicator variables is mediated by one or
more latent variables (Prince et al., 2012). Szmalec et al. (2011)
explicitly stated that “learning in Page and Norris’s model depends
both on the quality of the short-term representation of a to-berecalled list and on an independent weight-change process governed by a variable learning rate” (Szmalec et al., 2011, p. 1275).
Hence, consolidation in LTM underlying Hebb performance is
partly based on an independent modality-nonspecific process that
is different from the STM process needed to remember the filler
sequences. The central question is whether the relationship be-
Figure 6. A: State-trace plot for the digit sequences condition for typical readers and dyslexic readers. B: State-trace plot for the Corsi blocks sequences
condition for typical readers and dyslexic readers. C: State-trace plot for the visual form sequences condition for typical readers and dyslexic readers.
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
NO EVIDENCE FOR THE SOLID HYPOTHESIS OF DYSLEXIA
15
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
16
STAELS AND VAN DEN BROECK
tween a set of indicator variables (filler performance and Hebb
performance) and a set of causal variables (sequence type and
reading group) is determined by the same underlying latent processing mechanism or by a partly independent learning factor over
and above serial learning mechanisms in STM (see Figure 7). Note
that the figure optionally has a path (dotted arrow) from diagnostic
category to STM to capture the possibility that individuals with
dyslexia would already have a problem with serial learning in
STM. This does not alter the prediction of an additional deficit of
order learning in LTM.
The only requirement in STA is the broad assumption that
latent variables have a monotonic effect on the indicator variables. Hence, when proficiency at the underlying memory process increases, performance on filler and Hebb sequences
should also increase. The specific form this monotonically
increasing (or decreasing) relationship takes for each indicator
is irrelevant. From this assumption, a crucial prediction can be
derived (Bamber, 1979): If only one latent variable is required
to explain the relationship between indicator variables and
causal variables, the plot of one indicator against the other must
also increase (or decrease) monotonically. The outcome of an
STA is a state trace plot. The axes of this plot, representing the
indicator variables, define a state space for the examined processes (see Figure 8).
The trace factor (modality) in combination with the other
causal variables induces variation among the points in the plot,
which can be considered “as providing a ‘trace’ of the behavior
of the system in the corresponding state space” (Prince et al.,
2012, p. 80). If only one latent process underlies the behavior of
that system, variation of the latent factor creates a curve that
increases (or decreases) monotonically. When we translate this
condition to our Hebb learning experiments and hypothesize
that the same memory processes are responsible for both Hebb
learning in dyslexic and typical readers, all data points should
be situated on one monotonically increasing curve. Alternatively, if both reading groups’ underlying memory mechanisms
differ, each group will have its own curve. As we have detailed
in the analyses of Experiments 1 and 2, this prediction was
incorrect. However, by stating that the underlying memory
processes are indistinguishable in dyslexic and typical readers,
Figure 7. Latent dimensionality underlying the relationship between a set
of causal variables and a set of indicator variables as predicted by Szmalec
et al. (2011).
there is still uncertainty about which memory processes are
involved. At this point of reasoning, it is possible that for both
groups STM processes underlie filler performance, and an additional consolidation process into LTM underlies Hebb performance. From Figure 8, however, it becomes apparent that when
the results are averaged out over the participants to reduce
unreliability due to individual error variance, filler performance
almost perfectly predicts Hebb performance (r ⫽ .998 in children and r ⫽ .919 in adolescents). When combining the results
of both experiments, we obtained a Pearson correlation of .946
with a bias-corrected and accelerated (BCa) 95% confidence
interval (CI) of [.855, .982] (with 10,000 bootstrap samples).
Even if we included the Szmalec et al. (2011) results in this
analysis (also comprising the problematic visuospatial condition), the combined correlation was .885 with a BCa 95% CI of
[.710, .961] (with 10,000 bootstrap samples). Note that these
numbers should be considered as lower limits as differing
circumstances over studies and experiments introduce error
variance. This implies that either filler performance and Hebb
performance are based on the same memory processes or individual differences in the consolidation process in LTM are
strongly determined by individual differences in STM processes. This finding imposes constraints on models designed to
explain the relationship between immediate serial recall and the
Hebb repetition effect.
Figure 8. State-trace plots for children and adolescents aggregated over participants. Each point in the plot
represents the average performance of typical readers (circles) or dyslexic readers (squares) for each of the
presented conditions.
NO EVIDENCE FOR THE SOLID HYPOTHESIS OF DYSLEXIA
17
Table 7
Summary of Multilevel Models Tested in Different Experiments
Experiment
BICnull
BICfull
N data points
LR
df
p
Bayesian factor
Experiment 1 (adolescents)
Experiment 2 (children)
Szmalec et al. (2011)
Szmalec et al. (2011) w/o visuospatial condition
351.10
389.14
361.41
243.39
358.93
399.31
357.95
245.39
147
165
96
64
2.152
0.04
12.58
6.318
2
2
2
2
.341
.980
.002
.042
7.83 ⫹ Null
10.17 ⫹ Null
3.45 ⫹ Full
2.00 ⫹ Null
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
Note. BIC ⫽ Bayesian information criterion; N ⫽ sample size; LR ⫽ Likelihood Ratio Test (testing the difference between the null model and the full
model). Bayesian factor indicates which model is supported.
Conclusions and Discussion
We begin by summarizing the main conclusion that follows
from the two reported empirical studies examining a differential
Hebb learning effect in dyslexic versus normal readers. Neither in
the study on adolescents nor in the study on children were we able
to detect a specific Hebb-learning deficit in individuals with dyslexia. Given the fact that we used a more sensitive measure to
detect Hebb learning (weighted sums instead of slopes), that the
groups were matched more precisely on filler performance (using
state trace analysis), and given the larger power of our studies
compared to Szmalec et al. (2011) (see Table 7), we can be
confident that the underlying problem of dyslexics, after statistically controlling for attentional functioning, is not to be found in
a specific Hebb learning problem.
As a result, we consider it more likely that it is the finding of a
dyslexic Hebb learning deficit in the Szmalec et al. (2011) study
that is atypical rather than our null finding. There is yet another
reason for this inference. When we performed—as we did for our
own data—an STA averaged out over the subjects on the Szmalec
et al. data, we observed a moderate correlation (r ⫽ .44) between
filler performance and Hebb performance. The lack of a strong
linear relationship, however, was almost entirely due to the outlying data point of the visuospatial condition. When this data point
was left out, the linear relationship increased considerably (r ⫽
.85). In the same vein, when we performed a multilevel analysis,
integrating all data from the three tasks of the Szmalec study, the
Bayesian factor favoring the full model (implying differential
Hebb learning) was 3.45; however, when the visuospatial condition was deleted from this multilevel analysis, we observed a
Bayesian factor of 2.00 favoring the null model (see Table 7).
Thus, the Szmalec data of the verbal–visual and the verbal–
auditory conditions together also sustained the absence of a Hebblearning deficit in dyslexic adolescents. In conclusion, the divergent result of the Szmalec study can be pinned down to the
visuospatial condition.
Second, we discuss the extent to which our results on the
relationship between filler performance and Hebb performance
contribute to the current discussion on the mechanisms for encoding sequence information in STM and in LTM. After effectively
eliminating error variance due to individual differences by averaging out over individuals, this almost perfect linear relationship
turned out to be robust over conditions (modalities) and experiments (see Figure 8). It is clear that such an intricate relationship
between immediate serial recall (ISR) and long-term Hebb learning is difficult to reconcile with models conjecturing distinct and
independent coding systems for both kinds of order learning. Even
though (raw) Hebb performance is not the same as the Hebb
effect—the latter being defined as the difference between filler and
Hebb performance—the observed strength of the linear relationship leaves little room for the existence of separate coding mechanisms. From a common sense viewpoint, it can be argued that this
strong relationship was largely to be expected, because the task for
the participants is exactly the same when asked to recall the order
of the presented filler sequence or (repeated) Hebb sequence.
However, this merely begs the question: Is the improved performance on Hebb sequences the result of the processes involved in
the filler performance or is it the consequence of an independent
but closely related process?
Several theoretical memory models have hypothesized a distinction between a learning mechanism of ISR and an additional
mechanism for long-term serial learning. In the influential model
of Burgess and Hitch (2006), it is assumed that long-term serial
recall is exclusively dependent on the context-timing signal, which
is responsible for the encoding of the serial order of the items,
whereas ISR is only mildly dependent on this signal. The main
mechanism thought to be responsible for ISR is competitive queuing (Grossberg, 1987; Houghton, 1990). Hence, recovery of order
information in ISR (filler sequences) is believed to be largely
epiphenomenal to item coding and based on a different mechanism
than recovery of order information in long-term sequence learning
(Hebb sequences). Importantly, our results are not in contradiction
with the idea that item information is coded differently than order
information. What our results point to is that processing order
information in ISR (filler sequences) has a direct influence on the
longer term retention of order information (in Hebb sequences).
The connectionist model put forward by Page and Norris (2009),
on which Szmalec et al. (2011) based their predictions, is probably
in a better position to integrate our findings of a strong relationship
between filler en Hebb performance. Although this model also
implements different learning mechanisms for ISR and for longterm sequence learning, Hebb learning depends heavily on the
strength of the primacy gradient encoded in the order layer (ISR),
because the primacy gradient is copied into connection strengths
that are controlled by a learning rate. For that reason, this model
could probably be adapted to take account of individual differences in learning, in such a way that a strong relationship between
filler and Hebb performance would emerge.
References
Araujo, A. P. (2012). Attention deficit hyperactivity disorder and dyslexia:
A history of overlap. Arquivos de Neuro-Psiquiatria, 70, 83– 84. doi:
10.1590/S0004-282X2012000200001
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
18
STAELS AND VAN DEN BROECK
Baeyens, D., Van Dyck, L., Broothaerts, C., Danckaerts, M., & Kooij, S.
(2012). Zelfrapportage Vragenlijst voor Aandachtsproblemen en Hyperactiviteit (16 –25 jaar) [Self-Report Questionnaire for Attention Problems and Hyperactivity (ages 16 –25)]. Leuven, Belgium: Acco.
Baker, L. (1982). An evaluation of the role of metacognitive deficits in
learning disabilities. Topics in Learning & Learning Disabilities, 2,
27–35.
Bamber, D. (1979). State-trace analysis: A method of testing simple
theories of causation. Journal of Mathematical Psychology, 19, 137–
181. doi:10.1016/0022-2496(79)90016-6
Bentin, S., Deutsch, A., & Liberman, I. (1990). Syntactic competence and
reading ability in children. Journal of Experimental Child Psychology,
49, 147–172. doi:10.1016/0022-0965(90)90053-B
Boada, R., Willcutt, E. G., & Pennington, B. F. (2012). Understanding the
comorbidity between dyslexia and attention-deficit/hyperactivity disorder. Topics in Language Disorders, 32, 264 –284. doi:10.1097/TLD
.0b013e31826203ac
Boets, B., De Smedt, B., Cleuren, L., Vandewalle, J., & Ghesquiere, P.
(2010). Towards a further characterization of phonological and literacy
problems in Dutch speaking children with dyslexia. British Journal of
Developmental Psychology, 28, 5–31. doi:10.1348/026151010X485223
Boets, B., Op de Beeck, H., Vandermosten, M., Scott, S. K., Gillebert,
C. R., Mantini, D., . . . Ghesquière, P. (2013). Intact but less accessible
phonetic representations in adults with dyslexia. Science, 342, 1251–
1254. doi:10.1126/science.1244333
Bogaerts, L., Szmalec, A., Hachmann, W. M., Page, M. P. A., & Duyck,
W. (2013, May). Dyslexia as a dis-order: The SOLID hypothesis. Paper
presented at the International Workshop on Reading and Developmental
Dyslexia, San Sebastian, Spain.
Bowers, P. G. (1995). Tracing symbol naming speed’s unique contributions to reading disabilities over time. Reading and Writing, 7, 189 –216.
doi:10.1007/BF01027185
Bowers, P. G., & Ishaik, G. (2003). RAN’s contribution to understanding
reading disabilities. In H. L. Swanson, K. R. Harris, & S. Graham (Eds.),
Handbook of learning disabilities (pp. 1401–1457). New York, NY:
Guilford Press.
Bradley, L., & Bryant, P. (1983). Categorizing sounds and learning to read:
A causal connection. Nature, 301, 419 – 421. doi:10.1038/301419a0
Brady, S. A. (1991). The role of working memory in reading disability. In
S. A. Brady & D. P. Shankweiler (Eds.), Phonological processes in
literacy: A tribute to Isabelle Y. Liberman (pp. 129 –151). Hillsdale, NJ:
Erlbaum.
Brus, B. T., & Voeten, M. J. M. (1973). Eén-minuut-test [One-minute test].
Nijmegen, the Netherlands: Berkhout.
Burgess, N., & Hitch, G. J. (2006). A revised model of short-term memory
and long-term learning of verbal sequences. Journal of Memory and
Language, 55, 627– 652. doi:10.1016/j.jml.2006.08.005
Chung, K. K. H., Ho, C. S. H., Chan, D. W., Tsang, S. M., & Lee, S. H.
(2013). Contributions of syntactic awareness to reading in Chinesespeaking adolescent readers with and without dyslexia. Dyslexia: An
International Journal of Research and Practice, 19, 11–36. doi:10.1002/
dys.1448
Conway, C. M., & Christiansen, M. H. (2001). Sequential learning in
non-human primates. Trends in Cognitive Sciences, 5, 539 –546. doi:
10.1016/S1364-6613(00)01800-3
Conway, C. M., & Pisoni, D. B. (2008). Neurocognitive basis of implicit
learning of sequential structure and its relation to language processing.
Annals of the New York Academy of Sciences, 1145, 113–131. doi:
10.1196/annals.1416.009
Corsi, P. M. (1972). Human memory and the medial temporal region of the
brain. Dissertation Abstracts International: Section B. Sciences and
Engineering, 34(10), 891.
Deroost, N., Zeischka, P., Coomans, D., Bouzza, S., Depessemier, P., &
Soetens, E. (2010). Intact first- and second-order implicit sequence
learning in secondary-school-aged-children with developmental dyslexia. Journal of Clinical and Experimental Neuropsychology, 32, 561–
572. doi:10.1080/13803390903313556
Destrebecqz, A., & Cleeremans, A. (2001). Can sequence learning be
implicit? New evidence with the process dissociation procedure. Psychonomic Bulletin & Review, 8, 343–350. doi:10.3758/BF03196171
Dunn, J. C., & James, R. N. (2003). Signed difference analysis: Theory and
application. Journal of Mathematical Psychology, 47, 389 – 416. doi:
10.1016/S0022-2496(03)00049-X
Fletcher, C. M., & Prior, M. R. (1990). The rule learning behavior of
reading disabled and normal children as a function of task characteristics
and instruction. Journal of Experimental Child Psychology, 50, 39 –58.
doi:10.1016/0022-0965(90)90031-3
Folia, V., Udden, J., Forkstam, C., Ingvar, M., Hagoort, P., & Petersson,
K. M. (2008). Implicit learning and dyslexia. Annals of New York
Academy of Sciences, 1145, 132–150. doi:10.1196/annals.1416.012
Gombert, J. E. (2003). Implicit and explicit learning to read: Implication as
for subtypes of dyslexia. Current Psychology Letters, 10(1). Retrieved
from http://cpl.revues.org/
Gough, P., & Tunmer, W. (1986). Decoding, reading, and reading disability. RASE: Remedial and Special Education, 7, 6 –10. doi:10.1177/
074193258600700104
Gould, J. H., & Glencross, D. J. (1990). Do children with a specific reading
disability have a general serial-ordering problem? Neuropsychologia,
28, 271–278. doi:10.1016/0028-3932(90)90020-O
Grossberg, S. (1987). Competitive learning: From interactive activation to
adaptive resonance. Cognitive Science, 11, 23– 63. doi:10.1111/j.15516708.1987.tb00862.x
Hachmann, W. M., Bogaerts, L., Szmalec, A., Woumans, E., Duyck, W.,
& Job, R. (2014). Short-term memory for order but not for item information is impaired in developmental dyslexia. Annals of Dyslexia, 64,
121–136. doi:10.1007/s11881-013-0089-5
Harrison, A. G., Edwards, M. J., & Parker, K. C. H. (2008). Identifying
students feigning dyslexia: Preliminary findings and strategies for detection. Dyslexia: An International Journal of Research and Practice,
14, 228 –246. doi:10.1002/dys.366
Hebb, D. (1961). Distinctive features of learning in the higher animal. In
J. F. Delafresnaye (Ed.), Brain mechanisms and learning (pp. 37– 46).
Oxford, England: Blackwell
Houghton, G. (1990). The problem of serial order: A neural network model
of sequence learning and recall. In R. Dale, C. Mellish, & M. Zock
(Eds.), Current research in natural language generation (pp. 287–319).
London, England: Academic Press.
Johnson, D. J. (1993). Relationship between oral and written language.
School Psychology Journal, 22, 595– 609.
Kass, R. E., & Raftery, A. E. (1995). Bayes factors. Journal of the
American Statistical Association, 90, 773–795. doi:10.1080/01621459
.1995.10476572
Kelly, S. W., Griffiths, S., & Frith, U. (2002). Evidence for implicit
sequence learning in dyslexia. Dyslexia, 8, 43–52. doi:10.1002/dys.208
Lindstrom, W., Coleman, C., Thomassin, K., Southall, C. M., & Lindstrom, J. H. (2011). Simulated dyslexia in postsecondary students:
Description and detection using embedded validity indicators. The Clinical Neuropsychologist, 25, 302–322. doi:10.1080/13854046.2010
.537280
Loftus, G. R., Oberg, M. A., & Dillon, A. M. (2004). Linear theory,
dimensional theory, and the face-inversion effect. Psychological Review,
111, 835– 863. doi:10.1037/0033-295X.111.4.835
Long, J. S. (1997). Regression models for categorical and limited dependent variables. Thousand Oaks, CA: Sage.
Manis, F. R., Savage, P. L., Morrison, F. J., Horn, C. C., Howell, M. J.,
Szeszulski, P. A., & Holt, L. J. (1987). Paired associate learning in
reading-disabled children: Evidence for a rule-learning deficiency. Jour-
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
NO EVIDENCE FOR THE SOLID HYPOTHESIS OF DYSLEXIA
nal of Experimental Child Psychology, 43, 25– 43. doi:10.1016/00220965(87)90049-X
Manly, T., Robertson, I. H., Anderson, V., & Nimmo-Smith, I. (1999). The
Test of Everyday Attention for Children (TEA–CH). Bury St. Edmunds,
England: Thames Valley Test Co.
McKelvie, S. J. (1987). Learning and awareness in the Hebb digits task.
Journal of General Psychology, 114, 75– 88. doi:10.1080/00221309
.1987.9711057
Menghini, D., Hagberg, G. E., Caltagirone, C., Petrosini, L., & Vicari, S.
(2006). Implicit learning deficits in dyslexic adults: An fMRI study.
NeuroImage, 33, 1218 –1226. doi:10.1016/j.neuroimage.2006.08.024
Morrison, F. J. (1984). Word decoding and rule-learning in normal and
disabled readers. RASE: Remedial and Special Education, 5, 20 –27.
doi:10.1177/074193258400500307
Mosse, E. K., & Jarrold, C. (2008). Hebb learning, verbal short-term
memory, and the acquisition of phonological forms in children. The
Quarterly Journal of Experimental Psychology, 61, 505–514. doi:
10.1080/17470210701680779
Muthén, L. K., & Muthén, B. O. (2010). MPlus user’s guide (6th ed.). Los
Angeles, CA: Muthén & Muthén
Newell, B. R., & Dunn, J. C. (2008). Dimensions in data: Testing psychological models using state-trace analysis. Trends in Cognitive Sciences,
12, 285–290. doi:10.1016/j.tics.2008.04.009
Nicolson, R. I., & Fawcett, A. J. (1990). Automaticity: A new framework
for dyslexia research? Cognition, 35, 159 –182. doi:10.1016/00100277(90)90013-A
Nissen, M. J., & Bullemer, P. (1987). Attentional requirements of learning:
Evidence from performance measures. Cognitive Psychology, 19, 1–32.
doi:10.1016/0010-0285(87)90002-8
Orban, P., Lungu, O., & Doyon, J. (2008). Motor sequence learning and
developmental dyslexia. Annals of New York Academy of Sciences,
1145, 151–172. doi:10.1196/annals.1416.016
Page, M. P., Cumming, N., Norris, D., Hitch, G., & McNeil, A. (2006).
Repetition learning in the immediate serial recall of visual and auditory
materials. Journal of Experimental Psychology: Learning, Memory, and
Cognition, 32, 716 –733. doi:10.1037/0278-7393.32.4.716
Page, M. P. A., & Norris, D. (2008). Is there a common mechanism
underlying word-form learning and the Hebb repetition effect? Experimental data and a modelling framework. In A. Thorn & M. Page (Eds.),
Interactions between short-term and long-term memory in the verbal
domain (pp. 136 –156). Hove, England: Psychology Press.
Page, M. P. A., & Norris, D. (2009). A model linking immediate serial
recall, the Hebb repetition effect and the learning of phonological
word-forms. Philosophical Transactions of the Royal Society: B. Biological Sciences, 364, 3737–3753. doi:10.1098/rstb.2009.0173
Perez, T. M., Majerus, S., Mahot, A., & Poncelet, M. (2012). Evidence for
a specific impairment of serial order short-term memory in dyslexic
children. Dyslexia: An International Journal of Research and Practice,
18, 94 –109. doi:10.1002/dys.1438
Petersson, K. M., & Reis, A. (2006). Characteristics of illiterate and literate
cognitive processing: Implications of brain-behavior co-constructivism.
In P. B. Baltes, F. Rösler, & P. A. Reuter-Lorenz (Eds.), Lifespan
development and the brain: The perspective of biocultural constructivism (pp. 279 –305). New York, NY: Cambridge University Press. doi:
10.1017/CBO9780511499722.015
Prince, M., Brown, S., & Heathcote, A. (2012). The design and analysis of
state-trace experiments. Psychological Methods, 17, 78 –99. doi:
10.1037/a0025809
Raftery, A. E. (1996). Bayesian model selection in social research. In P. V.
Marsden (Ed.), Sociological methodology (Vol. 25, pp. 111–163). Oxford, England: Blackwell.
Ramus, F., Rosen, S., Dakin, S. C., Day, B. L., Castellote, J. M., White, S.,
& Frith, U. (2003). Theories of developmental dyslexia: Insights from a
19
multiple case study of dyslexic adults. Brain: A Journal of Neurology,
126, 841– 865. doi:10.1093/brain/awg076
Ramus, F., & Szenkovits, G. (2008). What phonological deficit? The
Quarterly Journal of Experimental Psychology, 61, 129 –141. doi:
10.1080/17470210701508822
Raudenbush, S. W., & Bryk, A. S. (2002). Hierarchical linear models.
Applications and data analysis methods. Thousand Oaks, CA: Sage.
Roodenrys, S., & Dunn, N. (2008). Unimpaired implicit learning in children with developmental dyslexia. Dyslexia: An International Journal of
Research and Practice, 14, 1–15. doi:10.1002/dys.340
Rousseeuw, P. J., & van Zomeren, B. C. (1990). Unmasking multivariate
outliers and leverage points. Journal of the American Statistical Association, 85, 633– 639. doi:10.1080/01621459.1990.10474920
Schacter, D. L. (1987). Implicit memory: History and current status.
Journal of Experimental Psychology: Learning, Memory, and Cognition, 13, 501–518. doi:10.1037/0278-7393.13.3.501
Schneider, W., Eschman, A., & Zuccolotto, A. (2002). E-Prime (Version
1.2.) [Computer software]. Pittsburgh, PA: Psychology Software Tools.
Scholte, E. M., & Van der Ploeg, J. D. (2005). AVL-vragenlijst [ADHD
Questionnaire]. Houten, the Netherlands: Bohn Stafleu Van Loghum.
Share, D. L. (1994). Deficient phonological processing in disabled readers
implicates processing deficits beyond the phonological module. In K. P.
Van den Bos, L. S. Siegel, D. J. Bakker, & D. L. Share (Eds.), Current
directions in dyslexia research (pp. 149 –167). Lisse, the Netherlands:
Swets & Zeitlinger.
Siegel, L. S. (1994). Working memory and reading: A life-span perspective. International Journal of Behavioral Development, 17, 109 –124.
doi:10.1177/016502549401700107
Siegel, L. S., & Ryan, E. B. (1984). Reading disability as a language
disorder. RASE: Remedial and Special Education, 5(3), 28 –33. doi:
10.1177/074193258400500308
Snowling, M. J. (2001). From language to reading and dyslexia. Dyslexia,
7, 37– 46. doi:10.1002/dys.185
Snowling, M. J. (2012). Changing concepts of dyslexia: Nature, treatment
and comorbidity. Journal of Child Psychology and Psychiatry, 53,
e1– e3. doi:10.1111/j.1469-7610.2009.02197.x
Stadler, M. A. (1993). Implicit serial learning: Questions inspired by Hebb
(1961). Memory & Cognition, 21, 819 – 827. doi:10.3758/BF03202749
Staels, E., & Van den Broeck, W. (2013). Order short-term memory is not
impaired in dyslexia and does not affect orthographic learning. Manuscript submitted for publication.
Stanovich, K. E. (1986). Cognitive processes and the reading problems of
learning disabled children: Evaluating the assumption of specificity. In
J. Torgesen & B. Wong (Eds.), Psychological and educational perspectives on learning disabilities (pp. 87–131). New York, NY: Academic
Press.
Stanovich, K. E. (1988). Explaining the differences between the dyslexic
and the garden-variety poor reader. The phonological-core variabledifference model. Journal of Learning Disabilities, 21, 590 – 604. doi:
10.1177/002221948802101003
Stanovich, K. E., & Siegel, L. S. (1994). Phenotypic performance profile of
children with reading disabilities: A regression-based test of the
phonological-core variable-difference model. Journal of Educational
Psychology, 86, 24 –53. doi:10.1037/0022-0663.86.1.24
Stichting Dyslexie Nederland. (2008). Diagnose dyslexie [Dyslexia diagnosis]. Bilthoven, the Netherlands: Author.
Stoodley, C. J., Harrison, E. P. D., & Stein, J. F. (2006). Implicit motor
learning deficits in dyslexic adults. Neuropsychologia, 44, 795–798.
doi:10.1016/j.neuropsychologia.2005.07.009
Stoodley, C. J., Ray, N. J., Jack, A., & Stein, J. F. (2008). Implicit learning
in control, dyslexic, and garden-variety poor readers. Annals of New
York Academy of Sciences, 1145, 173–183. doi:10.1196/annals.1416.003
Szmalec, A., Duyck, W., Vandierendonck, A., Mata, A. B., & Page,
M. P. A. (2009). The Hebb repetition effect as a laboratory analogue of
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
20
STAELS AND VAN DEN BROECK
novel word learning. The Quarterly Journal of Experimental Psychology, 62, 435– 443. doi:10.1080/17470210802386375
Szmalec, A., Loncke, M., Page, M. P. A., & Duyck, W. (2011). Order or
disorder? Impaired Hebb learning in dyslexia. Journal of Experimental
Psychology: Learning, Memory, and Cognition, 37, 1270 –1279. doi:
10.1037/a0023820
Torgesen, J. (1978). Performance of reading disabled children on serial
memory tasks. Reading Research Quarterly, 14, 57– 87. doi:10.2307/
747294
Turner, M. (1997). Psychological assessment of dyslexia. London, England: Whurr. doi:10.1002/9780470777947
Van den Bos, K. P., Lutje Spelberg, H. C., Scheepsma, A. J. M., & de
Vries, J. R. (1994). De Klepel. Vorm A en B. Een test voor de leesvaardigheid van pseudowoorden. Verantwoording Handleiding Diagnostiek
en behandeling [The Klepel, Forms A and B: A test for readability of
pseudowords. Justification manual diagnostics and treatment]. Nijmegen, the Netherlands: Berkhout.
Van den Broeck, W., & Geudens, A. (2012). Old and new ways to study
characteristics of reading disability: The case of the nonword-reading
deficit. Cognitive Psychology, 65, 414 – 456. doi:10.1016/j.cogpsych
.2012.06.003
Van den Broeck, W., Geudens, A., & van den Bos, K. P. (2010). The
nonword-reading deficit of disabled readers: A developmental interpretation. Developmental Psychology, 46, 717–734. doi:10.1037/a0019038
Van der Ven, A. H., & Smit, J. C. (1989). Het geheim van de BourdonWiersma test ontraadseld? [The secret of the Bourdon-Wiersma test
unraveled?]. Nederlands Tijdschrift voor de Psychologie en Haar Grensgebieden, 44, 260 –270.
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, 45, 2– 40.
doi:10.1046/j.0021-9630.2003.00305.x
Vicari, S., Finzi, A., Menghini, D., Marotta, L., Baldi, S., & Petrosini, L.
(2005). Do children with developmental dyslexia have an implicit learning deficit? Journal of Neurology, Neurosurgery & Psychiatry, 76(10),
1392–1397. doi:10.1136/jnnp.2004.061093
Vicari, S., Marotta, L., Menghini, D., Molinari, M., & Petrosini, L. (2003).
Implicit learning deficit in children with developmental dyslexia. Neuropsychologia, 41, 108 –114. doi:10.1016/S0028-3932(02)00082-9
Waber, D. P., Marcus, D. J., Forbes, P. W., Bellinger, D. C., Weiler, M. D.,
Sorenson, L. G., & Curran, T. (2003). Motor sequence learning and
reading ability: Is poor reading associated with sequencing deficits?
Journal of Experimental Child Psychology, 84, 338 –354. doi:10.1016/
S0022-0965(03)00030-4
Wechsler, D. (1998). Wechsler Intelligentietest voor Kinderen [Wechsler
Intelligence Scale for Children (3rd ed.)]. London, England: Psychological Corporation.
Wechsler, D. (2005). Wechsler intelligence scale for children (3rd ed.)
(WISC-III-NL). Amsterdam, the Netherlands: Pearson.
Weitz, D., O’Shea, G., Zook, N., & Needham, W. (2011). Working
memory and sequence learning in the Hebb digits task: Awareness is
predicted by individual differences in operation span. The American
Journal of Psychology, 124, 49 – 62. doi:10.5406/amerjpsyc.124.1.0049
Wimmer, H., Mayringer, H., & Raberger, T. (1999). Reading and balancing: Evidence against the automatization deficit explanation of developmental dyslexia. Journal of Learning Disabilities, 32, 473– 478. doi:
10.1177/002221949903200513
Wolf, M., & Bowers, P. G. (1999). The double-deficit hypothesis for
developmental dyslexia. Journal of Educational Psychology, 91, 415–
438. doi:10.1037/0022-0663.91.3.415
Wong, B. Y. L. (1991). The relevance of metacognition to learning
disabilities. In B. Y. L. Wong (Ed.), Learning about learning disabilities
(pp. 231–258). San Diego, CA: Academic Press. doi:10.1016/B978-012-762530-0.50014-X
Ziegler, J. C., & Goswami, U. C. (2005). Reading acquisition, developmental dyslexia and skilled reading across languages: A psycholinguistic
grain size theory. Psychological Bulletin, 131(1), 3–29. doi:10.1037/
0033-2909.131.1.3
Received September 17, 2013
Revision received June 24, 2014
Accepted June 26, 2014 䡲