Content uploaded by Brian Butterworth
Author content
All content in this area was uploaded by Brian Butterworth on Oct 16, 2017
Content may be subject to copyright.
Developmental dyscalculia and basic numerical
capacities: a study of 8–9-year-old students
Karin Landerl
a,b
, Anna Bevan
a
, Brian Butterworth
a,
*
a
Institute of Cognitive Neuroscience, University College London, 17 Queen Square, London WC1N 3AR, UK
b
Department of Psychology, University of Salzburg, Salzburg, Austria
Received 21 March 2003; revised 17 September 2003; accepted 13 November 2003
Abstract
Thirty-one 8- and 9-year-old children selected for dyscalculia, reading difficulties or both, were
compared to controls on a range of basic number processing tasks. Children with dyscalculia only
had impaired performance on the tasks despite high-average performance on tests of IQ, vocabulary
and working memory tasks. Children with reading disability were mildly impaired only on tasks that
involved articulation, while children with both disorders showed a pattern of numerical disability
similar to that of the dyscalculic group, with no special features consequent on their reading or
language deficits. We conclude that dyscalculia is the result of specific disabilities in basic numerical
processing, rather than the consequence of deficits in other cognitive abilities.
q2004 Published by Elsevier B.V.
Keywords: Dyscalculia; Numerical processing; Intelligence; Working memory; Arithmetic development
1. Introduction
Mathematics is a complex subject, involving language, space and quantity. Much
research into the development of mathematical skills has focused upon relatively
basic numerical abilities, such as arithmetic or counting (Bisanz, 1999a), but even at
such early levels many complex abilities are involved. These include transcoding
between spoken number words and Arabic numerals, relating these to semantic
representations of set size (“numerosity”), reasoning about relative set sizes (if 1 is
added to 2, the result should be 3); and understanding the relations between set size
and counting order.
0022-2860/$ - see front matter q2004 Published by Elsevier B.V.
doi:10.1016/j.cognition.2003.11.004
Cognition 93 (2004) 99–125
www.elsevier.com/locate/COGNIT
*Corresponding author. Fax: þ44-71-580-1100.
E-mail address: b.butterworth@ucl.ac.uk (B. Butterworth).
The complexity of numerical processing has made defining what it means to have a
specific mathematical learning disability (dyscalculia) difficult. Traditional definitions
(e.g. DSM-IV, American Psychiatric Association, 1994) state that the child must
substantially underachieve on a standardized test relative to the level expected given age,
education and intelligence, and must experience disruption to academic achievement or
daily living. Standardized achievement tests, however, generally test a range of skills,
which may include spatial and verbal abilities, before collapsing the total into one global
score of ‘maths achievement’. There is thus a substantial risk of Type I errors. In addition
standardized tests are diverse, so what is meant by ‘maths achievement’ may vary
substantially between tests. For this reason it has been hard for researchers to pinpoint the
key deficits in dyscalculia, or to be sure how to define dyscalculics for study.
A range of terms for referring to developmental maths disability has emerged, along
with different criteria. Geary and colleagues use the term “mathematical disabilities” and
include all children who fall below the 30th percentile (Geary, Hoard, & Hamson, 1999)or
35th percentile (Geary, Hoard, & Hamson, 2000)ontheWoodcock –Johnson
Mathematics reasoning test (Woodcock & Johnson, 2001). Jordan and colleagues
(Hanich, Jordan, Kaplan, & Dick, 2001; Jordan, Hanich, & Kaplan, 2003a; Jordan,
Kaplan, & Hanich, 2002) refer to “mathematics difficulties”, and include all children
below the 35th percentile of the Woodcock– Johnson Broad Mathematics Composite
Score. The 35th percentile criterion means that the best children will be about 0.39 SD
units below the mean, and that 90% of the sample will be better than 2 SDs below the
mean. Koontz and Berch (1996) use the term “arithmetic learning disabilities” and include
children below the 25th percentile on the Iowa Test of Basic Skills. Most children so
classified would fall between 0.67 and 1.18 SDs below the expected mean, and could thus
be regarded as in the low average or even the average range. These authors’ terminology,
as well as criteria, make it clear that they are considering a range of causes for low
mathematics achievement, not just the clinical condition of dyscalculia.
Several authors refer to “dyscalculia” or “developmental dyscalculia”. They tend to use
their own tests and a much more stringent criterion. Shalev, Manor, and Gross-Tsur (1997),
who have carried out the most extensive study of this condition, use the criterion of two
grades below chronological age. Butterworth’s (2003) Dyscalculia Screener requires scores
on two tests to be in the lowest two stanines (11th percentile). The study reported here used
teachers’ report and a criterion of 3 SDs below the mean of the control group to avoid false
positives. For this group, we will use the term ‘dyscalculia’ (abbreviated to DC).
1.1. Features of dyscalculia
The most generally agreed upon feature of children with dyscalculia is difficulty in
learning and remembering arithmetic facts (Geary, 1993; Geary & Hoard, 2001; Ginsburg,
1997; Jordan, Hanich, & Kaplan, 2003b; Jordan & Montani, 1997; Kirby & Becker, 1988;
Russell & Ginsburg, 1984; Shalev & Gross-Tsur, 2001). A second feature of children with
dyscalculia is difficulty in executing calculation procedures, with immature problem-
solving strategies, long solution times and high error rates (Geary, 1993).
Temple (1991) has demonstrated using case studies that fact retrieval and procedural
difficulties are dissociable in dyscalculia. However, case studies, while providing
K. Landerl et al. / Cognition 93 (2004) 99–125100
important theoretical information on cognitive structures, are not necessarily representa-
tive of the majority of dyscalculic children: such dissociations may be rare. Ashcraft,
Yamashita, and Aram (1992) found no dissociation between arithmetic fact ability and
procedural ability in children with numerical processing difficulties. Russell and Ginsburg
(1984) found that a group of children with “mathematics difficulties” struggled with both
written calculation and arithmetic fact retrieval. Geary (1993) suggests that procedural
problems are likely to improve with experience, whereas retrieval difficulties are less
likely to do so. Geary proposes that this dissociation emerges because procedural problems
are due to lack of conceptual understanding, while retrieval difficulties are the result of
general semantic memory dysfunction. However, it is possible that both difficulties result
from a lack of conceptual understanding. It may be easier for a child to memorise one or
two meaningless procedures than the multitude of arithmetic facts (from simple number
bonds to multiplication tables) which, without understanding of cardinality, are simply
unrelated word strings.
Researchers, then, agree that dyscalculia manifests itself as a problem in learning
arithmetic facts and calculation procedures. The question which remains unanswered
relates to the underlying deficits which cause these problems. Various candidates have
been put forward, including dyslexic difficulties, memory difficulties, spatial difficulties
and attentional difficulties. However, many of the studies which have been designed to
relate these “underlying” abilities to dyscalculia have confounded them with numerical
processing abilities.
1.2. Underlying processing deficits
One approach to the study of developmental dyscalculia involves identifying the
cognitive and neuropsychological correlates of dyscalculia in an attempt to extrapolate
causal features or to identify dyscalculia subtypes with differential causes. Suggested
cognitive causes include abnormal representations in semantic memory and difficulty with
working memory.
Geary and colleagues (Geary et al., 2000; Geary & Hoard, 2001) have suggested that
semantic memory difficulties may underlie the problems experienced by developmental
dyscalculics in learning number facts, and may also underlie the comorbid reading
difficulties frequently found with dyscalculia. The argument is based on evidence that
dyscalculic children have difficulty learning and remembering arithmetic facts. However,
if this theory is correct, we should expect all dyslexic children to have number fact
problems and vice versa. In addition, the argument confounds semantic memory with
numerical processing. There is little empirical evidence for a non-numerical semantic
deficit in dyscalculic children. To our knowledge only one study has found such evidence:
Temple and Sherwood (2002) found that a group of children with arithmetic difficulties
were slower at colour and object naming than controls, evidence for a generalized speed of
access difficulty in this sample. However, the authors argued against a causal relationship
between speed of access and arithmetic ability, one reason for caution being the small size
of the group (four participants). Another problem with the hypothesis arises in the light of
evidence that semantic memory for numbers is mediated by a different system than general
semantic memory. Neuropsychological studies indicate that number knowledge is
K. Landerl et al. / Cognition 93 (2004) 99–125 101
dissociable from verbal semantic memory (Cappelletti, Butterworth, & Kopelman, 2001),
and that the semantic memory systems for numerical and non-numerical information are
localized in different areas of the brain (Thioux, Seron, & Pesenti, 1999). This functional
and anatomical dissociation between the two memory systems makes it unlikely that the
same semantic deficit can account for both maths and reading disability.
Working memory difficulties have also been associated with developmental
dyscalculia. Geary (1993) suggests that poor working memory resources not only
lead to difficulty in executing calculation procedures, but may also affect learning of
arithmetic facts. In general the aspect of working memory that has been focused on is
the phonological loop (Baddeley, Lewis, & Vallar, 1984), normally assessed by the
number of spoken items (generally digits) which can be remembered in the correct
sequence. However, empirical evidence for a correlation between reduced span in
phonological working memory and developmental dyscalculia is conflicting, and the
issue is confused by potential confounds. Reduced phonological span is associated with
general academic difficulties and with dyslexia, both of which are also associated with
difficulties in mathematics, and, presumably, dyscalculia. Therefore, any study
examining Working memory in dyscalculic children should control for both reading
ability and general IQ.
Siegel and Ryan (1989) found that children with dyscalculia did less well than controls
on a working memory task involving counting and remembering digits, but not on a non-
numerical Working memory task. This led them to speculate that there is a working
memory system specialised for numerical information, and that children with dyscalculia
have specific problems with this system. Similarly, McLean and Hitch (1999) found a
trend towards poorer digit span in dyscalculic children, while there was no difference on a
non-numerical task testing phonological working memory (non-word repetition). No
evidence was found for a faster decay rate of phonological representations in dyscalculic
children. These authors concluded that dyscalculic children do not have reduced
phonological working memory capacity in general, although they may have a specific
difficulty with working memory for numerical information. On the other hand, they found
that spatial working memory and some aspects of central executive function were poorer
in dyscalculic children.
Koontz and Berch (1996) tested children with and without dyscalculia using both digit
and letter span (the latter being a measure of phonological working memory capacity
which is not confounded with numerical processing). This study found that dyscalculic
children performed below average on both span tasks, indicating general working memory
difficulties. However, although children in this study were above the 30th percentile on a
standardized reading achievement test, general IQ was not controlled. In contrast, Geary
et al. (1999) did not find a difference between dyscalculic children and controls on the
forward digit span measure, but did find a difference on backward digit span, thought to tap
central executive processes.
Temple and Sherwood (2002) attempted to resolve this confusion, testing children with
dyscalculia and controls on forward and backward digit span, word span and Corsi blocks.
This study found no differences between groups on any of the working memory measures,
and no correlation between the working memory measures and measures of arithmetic
ability.
K. Landerl et al. / Cognition 93 (2004) 99–125102
On balance, although various forms of working memory difficulty may well co-occur
with maths difficulties, there is little agreement between these studies, and no convincing
evidence implicating any form of working memory as a causal feature in dyscalculia.
1.3. Subtyping
Another approach to the study of developmental dyscalculia has involved subtyping
dyscalculics according to the presence or absence of other disorders, in an attempt to
highlight underlying processes which may contribute to the comorbidity of the disorders.
An important correlate of maths disability is reading disability. It is estimated that 40%
of dyslexics also have maths disability (Lewis, Hitch, & Walker, 1994). One of the most
common ways of subtyping dyscalculic children is according to whether or not they have
a comorbid reading disability. Hanich et al. (2001) found that “children with MD-only
seem to be superior to children with MD/RD in areas that may be mediated by language
but not in ones that rely on numerical magnitudes, visuospatial processing, and
automaticity”.
Rourke and his colleagues (see Rourke, 1993 for a review) have compared children
with arithmetic difficulties only and children with better arithmetic scores than reading
scores. Children with arithmetic difficulties only were more likely to have difficulties with
spatial and psychomotor abilities, whilst children with reading difficulties were more
likely to struggle with verbal tasks. The authors suggest that these findings indicate that
comorbid maths and reading difficulties result from left-hemisphere dysfunction, while
specific difficulty with maths stems from right-hemisphere dysfunction. However,
Rourke’s constellation of ‘right-hemisphere’ symptoms is similar to the ‘left-parietal’
constellation found in Gerstmann’s syndrome (Gerstmann, 1940). In addition, a recent
attempt to replicate Rourke’s findings (Shalev et al., 1997) failed; Shalev et al. found no
qualitative difference between children with both reading and maths disability and
children with maths disability only. Children with both disorders scored more poorly on
several measures, but the authors concluded that this was unsurprising, given that the
presence of more than one disorder indicates relatively widespread brain dysfunction.
Fayol, Barrouillet, and Marinthe (1998) attempted to test Rourke’s hypothesis regarding
the causal relationship between neuropsychological deficits and arithmetic difficulties.
They conducted a longitudinal study in which nursery school children were given tests of
finger agnosia, graphisthesia and simultagnosia. These neuropsychological measures
correlated with simple arithmetic tests given at the same time. However, except (oddly) for
word problem solving, general intelligence in nursery school was a better predictor of
arithmetic in the 1st year of school than were the neuropsychological tests. This finding
suggests that correlation, in this case, is not causation.
Another set of deficits which are associated with developmental dyscalculia are finger
agnosia, dysgraphia and difficulties with left– right discrimination. Taken together this
symptom complex constitutes developmental Gerstmann’s syndrome. However, since it
appears that the four symptoms can appear individually and in any combination, and are
frequently associated with other conditions such as reading disability (Kinsbourne &
Warrington, 1963; Spellacy & Peter, 1978) it is unlikely that the symptoms are related in
terms of a single underlying deficit.
K. Landerl et al. / Cognition 93 (2004) 99–125 103
Other conditions which have been associated with dyscalculia are ADHD (Badian,
1983; Rosenberg, 1989; Shalev et al., 1997), poor hand – eye co-ordination (Siegel &
Feldman, 1983), poor memory for non-verbal material (Fletcher, 1985), and poor social
skills (Rourke, 1989). Shalev and Gross-Tsur (1993) examined a group of seven children
with developmental dyscalculia who were not responding to intervention. All seven were
suffering numbers, additional neurological conditions, ranging from petit mal seizures
through dyslexia for numbers, attention deficit disorder and developmental Gerstmann’s
syndrome. In summary, while it is clearly the case that dyscalculia is frequently comorbid
with other disabilities, causal relationships between the disorders have not been proven. In
addition, the utility of subtying dyscalculics according to neuropsychological or cognitive
correlates will not be clear until it has been shown that the different subtypes display
qualitatively different patterns of numerical deficit.
Relatively few studies have examined differences between subtypes on tasks involving
numerical processing. Shalev et al. (1997) found that children with comorbid maths and
reading difficulties were more profoundly impaired than children with specific maths
problems on subtraction and division and had lower verbal IQ scores. They also scored
consistently lower on most of the WISC subtests, although this difference did not reach
statistical significance. However, the pattern of numerical impairment was the same for
both groups. This study found no evidence for a dissociation between the two groups in
numerical processing, although children with comorbid maths and reading difficulties
tended to be more impaired than children with specific maths problems.
Jordan and Montani (1997) compared a group of children with specific maths disability
with a group of children who had maths disability in the context of more general academic
difficulties. Children with maths disability only were better able to execute backup
strategies in arithmetic, and were able to perform at a normal level under untimed
conditions, although their performance dropped under timed conditions. Children with
more general difficulties struggled under both conditions. The authors suggested that
children with specific maths difficulties are able to compensate under untimed conditions
because of relatively good verbal or conceptual skills. However, although this study also
indicates that children with general difficulties have quantitatively more difficulty than
children with specific maths disability, again there is no evidence that the pattern of
numerical impairment is qualitatively different between the two groups. More detailed
examination of the numerical abilities of groups of children are in order before it is certain
that subtyping developmental dyscalculics according to this framework is a useful
approach.
It is clear that dyscalculia frequently co-occurs with a range of other disabilities.
However, it is still far from clear that these disabilities play any causal role in
developmental maths disability. Not only has no single underlying process been
identified which predicts dyscalculia, but also there is no evidence for qualitatively
different patterns of impairment across dyscalculia subtypes, as would be expected if
different subtypes corresponded to different underlying causes. There is no robust
empirical evidence causally relating any of these correlates to numerical ability. In
addition there is very little coherent theory which could explain such causal
relationships. Currently the most likely explanations for overlap between different
disabilities are anatomical or genetic: damage to a brain area or failure of that area to
K. Landerl et al. / Cognition 93 (2004) 99–125104
develop normally may affect one or more cognitive functions depending on the extent
and severity of the damage (Shalev & Gross-Tsur, 1993).
1.4. Dyscalculia as a deficit in a specialized brain system
The studies described above have attempted to get at the root of dyscalculia by
examining various abilities, not obviously related to number processing, which are
hypothesized to underlie dyscalculia. This approach involves an implicit assumption that
the representation and manipulation of numerical information is a higher-order Rinction
which is dependent upon the abilities described. However, evidence from neuropsychol-
ogy and research with animals and very young children suggests that number processing is
not only independent of other abilities, but is also manifested at a very basic level.
Numerical abilities, including arithmetic, are mediated by areas in the parietal lobe
(Dehaene, Dehaene-Lambertz, & Cohen, 1998). Neuropsychological evidence has shown
that the ability to understand numbers and to calculate is dissociable from language
(Cohen, Dehaene, Cochon, Lehericy, & Naccache, 2000); from semantic memory for non-
numerical information (Cappelletti et al., 2001); and from working memory (Butterworth,
Cipolotti, & Warrington, 1996). There is also evidence for a genetic basis for this
specialization in studies of Turner’s Syndrome (Butterworth et al., 1999).
Not only are numerical abilities independent of other abilities, but also appear to be
‘hardwired’ and are manifested even in infants (Starkey, Spelke, & Gelman, 1990; Wynn,
1995). A number of studies have found numerical processing abilities in animals (see
Gallistel, 1990 for a review). Thus number processing appears to be a function which
emerges in infants at a very early age, and is independent of other abilities. This argues
against a role for language-related abilities such as semantic or working memory in
developmental dyscalculia. It seems likely that basic numerical functions, such as
comprehension of numerical symbols, counting, and simple calculation, are built primarily
upon early mechanisms for processing small numerosities. These mechanisms also seem
to be a good candidate for a basic deficit underlying dyscalculia.
If dyscalculia is the result of a fundamental difficulty with numerical processing, as is
proposed here, dyscalculic children should have problems with even the most basic
functions involving numbers such as subitizing, counting small numbers of objects, using
number names and numerals, and comparing numerical magnitudes, as well as more
advanced arithmetical skills.
Some evidence for this comes from a study by Koontz and Berch (1996), who found
that the dyscalculic children appeared to be counting to 3 rather than subitizing in a dot-
matching task. Kirby and Becker (1988) found a trend for dyscalculic children to be slower
at number naming relative to controls, while Geary et al. (1999) found small but
systematic group differences between 1st grade dyscalculic children and controls in
number naming, number writing and magnitude comparison. Geary, Bow-Thomas, and
Yao (1992) found that dyscalculic children are less likely to detect counting errors than
control children, and are delayed in understanding the principles of counting (Gelman &
Gallistel, 1978).
However, Russell and Ginsburg (1984) tested children with “mathematics difficulties”
on tasks involving ‘informal’ numerical concepts, including indicating which of two
K. Landerl et al. / Cognition 93 (2004) 99–125 105
numbers is more, indicating which of two numbers is closer to a stated number x, and
estimation of quantities. Their criterion for mathematics difficulty is three stanines, or one
grade below CA on the Iowa Mathematics Achievement Test, with exclusions for low IQ
or sensory or motor disabilities. This yields a sample of about 7% of the test population
before attrition, which would be in line with the dyscalculia prevalence estimates, so these
subjects seem to be similar to the studies where children are described as “dyscalculic”.
They found that, despite arithmetical difficulties, these children have normal ‘informal’ or
conceptual understanding. This study is of theoretical importance, since the findings imply
that the problems faced by dyscalculic children are specifically to do with memory rather
than with numerical understanding. However, very few (between four and six) items were
presented for each task; there was a 50% probability of answering correctly by chance; and
error rate, a relatively insensitive measure, was used as the only indicator of performance.
As a matter of fact, on the ‘which is more’ task, all groups performed near ceiling, and on
the ‘which is closer to x’ task, all groups performed near chance. In an estimation task six
items were presented, and while the difference between groups was not statistically
significant, it should be noted that there was a difference of nearly an item between the
mean scores of control and dyscalculic children. These findings are in need of replication
using more items and more sensitive measures such as reaction time latencies.
In conclusion, information regarding dyscalculics’ numerical abilities is rather
scattered and conflicting, but does indicate that general difficulties with basic number
processing may be a feature of dyscalculia. Attempts to discover the root cause of
dyscalculia by correlating it with other disabilities have not so far been successful. This
approach assumes that number skills are relatively of higher-order and predicated on other
abilities such as verbal or spatial skills. However, neuropsychological and developmental
research points to the existence of a ‘number module’ based in the parietal lobe
(Butterworth, 1999) for dealing with numerical representations. We propose that the
underlying cause of dyscalculia is likely to be related to dysfunction of this system.
1.5. Dyscalculia as a deficit in the concept of numerosity and its processing
Although to our knowledge there has been no systematic attempt to investigate
specifically the numerical skills of dyscalculic children, studies have consistently found
evidence of difficulties even on simple number tasks.
Further examination of the basic number processing abilities of dyscalculic children is
necessary for two reasons. First a more fine-grained picture of their difficulties is needed in
order to make theoretical headway. If, as proposed here, the deficit is numerical in origin,
all aspects of numerical processing are likely to be affected. If it is not, then more
information regarding the specific numerical deficits displayed by dyscalculic children
will be helpful in relating arithmetic deficits to other underlying abilities. In addition, the
utility of subtyping dyscalculics according to neuropsychological or cognitive correlates
will not be clear until it has been shown that the different subtypes display qualitatively
different patterns of numerical deficit. The current study examines basic numerical
processing in detail, and employs a subtyping procedure based on comorbid reading
difficulties. It is predicted that dyscalculic children will demonstrate a broad range of
K. Landerl et al. / Cognition 93 (2004) 99–125106
number processing deficits, but that this pattern will not be affected by the presence of
comorbid reading disabilities.
The basic numeric processing skills that are examined in the present study include the
socially necessary skills of transcoding between a verbal and an Arabic number code
(naming and writing of 1- to 3-digit numbers), but the primary focus is on how well
children understand numerosity.
Numerosity is a property of a set and is not, therefore, tied to properties of the
individual objects that make up that set. The basic principles derived from the concept of
numerosity have been elaborated by Gelman and Gallistel (1978) and Piaget (1952).
Piaget stressed an understanding of the kinds of transformation that would and would not
affect the numerosity of a set in his “conservation” tasks. Thus, moving objects around
would not affect numerosity, while adding or subtracting objects would. He also laid
emphasis on understanding that two sets would have the same number if and only if their
members could be put in one-to-one correspondence. Gelman and Gallistel focused on the
procedures for enumerating sets using counting words, but drew attention also to the kinds
of procedural variations that would affect the outcome of the count (one-to-one
correspondence with the counting words) and which would not (e.g. which object begins
the count, and what kind of object is being counted). One should note also that
understanding numerosity implies understanding that each numerosity, say 4, is distinct,
and has a one successor, in this case 5, and that numerosities are ordered by magnitude—
that is 5 has a larger magnitude than 4.
There is evidence that children are born with a capacity for recognizing and even
mentally manipulating small numerosities (Starkey et al., 1990; Wynn, 1992, 1995), and
adults seem able to carry out judgements on small numerosities with greater speed and
accuracy. In this study, we focused on two tasks that required an understanding of
numerosity and the ability to recognize and judge small numerosities, and which did not
require high levels of formal mathematical achievement. Indeed, all that was needed, was
an ability to understand numerosities up to 9. We therefore used a simple number
comparison, which would test whether children understood number magnitudes, and
simple counting skill. Response speed was measured for each of the tasks as deficits in
dyscalculic children are not always detected in untimed tasks (Jordan & Montani, 1997).
One adult dyscalculic, “Charles”, reported by Butterworth (1999) was abnormally slow on
both of these tasks.
In addition to tasks assessing basic numeric processing we also carried out a number of
non-numeric tasks for which we do not expect specific deficits in dyscalculia. Number
naming speed is contrasted with the non-numerical control task of colour naming
presented in exactly the same format. The number comparison task includes a non-
numeric control condition where participants had to compare the physical size of the
presented numbers. Phonological short-term and working memory (WISC-III digit span
forward and backward) and vocabulary (BPVS) were also assessed. Psychomotor
functions were assessed by the WISC-III Mazes subtest which requires adequate visuo-
motor co-ordination as well as executive functions (planning the way through the maze,
monitoring if you are still on the right track and inhibition of going into blind alleys).
Although all of these skills were suspected to be deficient in dyscalculia in the literature,
K. Landerl et al. / Cognition 93 (2004) 99–125 107
we predict that the deficits of the dyscalculic children will be confined to numerical
processing.
A timed arithmetic test was used rather than scores on an achievement test to classify
dyscalculic children. Numerical achievement tests yield a composite score based on a
range of different abilities (for example number reading and writing, arithmetic, ability to
carry and borrow, spatial and organizational skills) so that it is not clear which abilities are
problematic. Achievement test scores may fluctuate over time and may not be reliable. In
addition, achievement tests are generally untimed, although evidence indicates that
dyscalculic children may perform as well as controls in untimed conditions (Jordan &
Montani, 1997). Ashcraft et al. (1992) found no correlation between scores on a test of
mathematical achievement and numerical abilities for their left-lesioned maths disabled
subjects. It was felt that a timed arithmetic test would be a more sensitive index of
dyscalculia, given that difficulty with arithmetic facts is a defining feature of the disorder.
2. Method
2.1. Participants
Fifty-four participants were selected from a larger group of 89 4-year-old children from
11 middle schools in the London area. Nine of the schools were state schools, one was a
private school specialized for dyslexic children and one was a normal private school.
Initial selection was based upon teacher assessment: teachers were asked to nominate
children who they felt were of normal intelligence but had serious difficulties with reading,
numeracy or both. For each of these children a control child of average ability was tested
who was of the same gender and from the same class, in order to minimize as much as
possible effects of instruction. Clearly controls could not be matched in this way for the
children from the specialized private school, so controls were taken from a nearby private
school.
IQ was assessed using Ravens CPM (Raven, Court, & Raven, 1986). For seven children
recent scores from a full WISC-III (Wechsler, 1992) assessment were available, so their
WTSC performance IQ scores were used as measures of non-verbal IQ rather than the
CPM for these children. (The CPM is highly correlated with WISC-R scores, with
correlations up to þ0.86 in some studies; Raven et al., 1986.) All children were
administered the Number Skills and the Word Reading subscale from the British Ability
Scales (BAS II, Elliot, Smith, & McCulloch, 1997). A customized timed test involving
simple addition, subtraction and multiplication was also administered (see Tasks below).
Data were discarded if the child was bilingual ðN¼4Þor had been diagnosed with ADHD
ðN¼2Þ:
2.1.1. Classification scheme
To qualify for the control group ðN¼28Þ;children had to have been nominated as
‘average’ by their teacher, and to have scored between the 25th and 90th percentiles
for their age on the BAS subtests and non-verbal IQ. To qualify as ‘dyslexic’ children
had to have been nominated as learning disabled by their teacher, to have scored below
K. Landerl et al. / Cognition 93 (2004) 99–125108
the 25th percentile on the BAS word-reading subscale, and above the 25th percentile on
non-verbal IQ.
Arithmetic skills were assessed by a timed test described below (Tasks). Median
reaction times for each of three tasks (addition, subtraction and multiplication) were
calculated for each child. The mean of these medians for each child was then used as a
measure of speed on the mental arithmetic tasks overall. The number of errors made on
each task was summed, giving a total number of errors per child on the mental arithmetic
tasks overall. The group means and standard deviations of the reaction times and errors
were calculated for the control group only. Any child scoring more than 3 standard
deviations above the control mean on reaction time or error was designated as dyscalculic.
Two measures (error and reaction time) were used because it had been noted that children
who struggled with mental arithmetic tended to adopt one of the two strategies: they would
painstakingly work out each sum on their fingers or by counting, leading to generally
accurate answers but extremely long RT latencies; or they would simply guess quickly,
leading to inaccurate answers but short RT latencies. A composite score for the facts was
used because it was a more conservative criterion than using scores on the individual
operations. A composite criterion ensured that children designated as having dyscalculia
were either struggling with all three arithmetic operations, or were having severe
difficulties with at least one operation.
A cutoff of 3 standard deviations is conservative, but was used to ensure a low
incidence of false positives. In addition, all children categorized as dyscalculic had been
nominated as having learning difficulties by their teacher. Children who scored between 2
and 3 standard deviations above the mean on either of the facts measures (RT and
accuracy) were considered ‘borderline’ and removed from the analysis.
Three groups were formed on the basis of these criteria: a group of 10 dyslexic children
(6 boys, 4 girls), who scored below the 25th percentile on the BAS word-reading task but
were within the 2nd standard deviation above the mean for both arithmetic facts measures
(RT and accuracy); a group of 10 dyscalculic children (8 boys, 2 girls) who scored above
the 25th percentile on the BAS word-reading and whose performance was more than 3
standard deviations below the mean on at least one of the facts measures; and a group of 11
double deficit (dyslexic and dyscalculic) children (8 boys, 3 girls) who scored below the
25th percentile on the reading test and whose performance was also more than 3 standard
deviations below the mean for at least one of the facts measures. Although these groups are
small, it was felt that identification of children with serious difficulties was a higher
priority than large sample sizes, since patterns of ability may differ between children with
genuine learning disabilities and those with relatively minor problems.
Once the three LD groups had been identified, an attempt was made to match the
control group as closely as possible to the dyscalculic and dyslexic groups for reading and
arithmetic facts, respectively. Ten controls with reading ages a year or more ahead of their
chronological ages were discarded, since dyscalculic children tended to be average or
somewhat below on the reading test. Thus the matched control group consisted of 18
children (8 boys, 10 girls).
The means and standard deviations for the new matched control group for arithmetic
facts were checked against the criteria based on the old control group. There was very
little difference between the two sets of criteria, except that those based on the new
K. Landerl et al. / Cognition 93 (2004) 99–125 109
matched group were marginally more lenient, due to reduced variability. However, this
difference was not large enough to affect the classifications. The subject details are
presented in Table 1.
Age differences between groups did not reach statistical significance; however,
dyscalculic and double deficit children were on average 5 months younger than control
children and 7 months younger than the dyslexic group. Given that younger children might
be expected to be poorer at arithmetic and at number processing generally, age was entered
as a covariate into all subsequent analyses.
An independent samples ANCOVA (controlling for age) indicated that there were no
differences between groups on the Raven CPM ðF,1Þ:There was a statistically significant
difference between groups on Word Reading: Fð3;48Þ¼29:7;P,0:001:t-Tests indicated
that the double deficit group had lower scores than controls: tð27Þ¼7:2;P,0:001;and
dyscalculic children: tð19Þ¼4:7;P,0:001:The dyslexic group also scored lower than
controls: tð26Þ¼7:6;P,0:001;and dyscalculic children: tð18Þ¼5:4;P,0:001:There
were no differences between the dyscalculic group and controls, or between the double
deficit and the dyslexic group. Separate ANCOVAs with age as a covariate were carried out
on speed and accuracy of arithmetic facts. There were statistically significant differences
between groups for both speed: Fð3;40Þ¼12:2;P,0:001;and accuracy: Fð3;43Þ¼10:7;
P,0:001:Group comparisons controlling for age indicated that the double deficit and the
dyscalculic group were both slower and less accurate than the control group and the dyslexic
group (all P,0:05). There were no differences in either speed or accuracy between the
double deficit group and the dyscalculic group, or between the dyslexic group and the
control group. The BAS Number Skills test was included as a comparison measure, although
it was not used for selection. There was a significant difference between groups:
Fð3;48Þ¼9:1;P,0:001:t-Tests indicated that the double deficit group had lower scores
than controls: tð27Þ¼4:1;P,0:001;and lower scores than the dyslexic group: tð19Þ¼
2:7;P¼0:01:The dyscalculic group also had lower scores than controls: tð26Þ¼3:9;
P¼0:001;and lower scores than the dyslexic group: tð18Þ¼2:5;P¼0:02:There was no
difference between the double deficit group and the dyscalculic group; or between the
dyslexic and the control groups.
Table 1
Subject details
Control
ðN¼18Þ
Dyslexic
ðN¼10Þ
Dyscalculic
ðN¼10Þ
Double deficit
ðN¼11Þ
Age (months) 108.7 (8.6) 110.1 (5.9) 103.7 (6.0) 103.9 (5.7)
Raven CPM
(raw scores)
28.8 (3.4)
a
29.7 (3.7) 28.5 (3.8)
b
27.0 (3.4)
c
BAS reading
(RA-CA in months)
20.94 (6.9) 219.90 (4.8) 26.30 (6.4) 219.73 (6.6)
BAS numeracy
(NA-CA in months)
5.72 (8.1) 0.90 (5.5) 28.20 (10.4) 27.18 (8.3)
a
N¼17;one child had a WISC performance IQ score.
b
N¼8;two children had WISC performance IQ scores.
c
N¼7;four children had WISC performance IQ scores.
K. Landerl et al. / Cognition 93 (2004) 99–125110
The above analysis involving the two criteria for categorization (word reading and
arithmetic facts) indicates that the groups have been classified correctly and are reasonably
well matched. The analysis of the BAS Number Skills test is also encouraging, since this
measure was not used to categorize the children. This suggests that arithmetic fact ability,
at least in our sample, is associated with the global numerical ability tapped by
standardized tests.
2.2. Tasks
2.2.1. Standardized tests
Ravens Coloured Progressive Matrices was administered as a measure of non-verbal
reasoning ability, and the British Picture Vocabulary Scale (Dunn, Dunn, & Whetton,
1997) as a measure of verbal ability. The Word Reading scale from the British Ability
Scale (BAS II) was used as measure of reading ability. The Number Skills scale from the
BAS was also administered. Phonological short-term and working memories were
measured using the Digit Span scale from the WISC-III, which includes forwards and
backwards digit span. Finally, the Mazes subscale from the WISC-III was used.
2.2.2. Computer tasks
The computer tasks were presented on a Macintosh PowerBook running SuperLab
software. Unless otherwise specified, numbers were 0.6 £1 cm in size, presented in an
8£7 cm field in the centre of the screen. Items were presented after a 400 ms fixation
point accompanied by an acoustic signal. Responses were verbal and triggered a voice-
activated key which measured reaction latencies from the onset of presentation. All items
remained visible until the voice key was activated. The experimenter pressed a key in
order to trigger the next item.
Mental arithmetic. 12 simple additions, 12 simple subtractions and 12 simple
multiplications were presented in 3 separate blocks. All involved single-digit numbers
from 2 to 9, excluding 0 and 1, since number facts involving 0 and 1 can be solved by
application of a rule rather than calculation or retrieval. No ties (e.g. 3 þ3, 5 £5) were
presented, and items were not repeated. Items were presented on the computer screen in
the form “3 þ4”. Six practice addition sums were given before the first block (additions);
four practice subtractions before the start of the subtraction block, and four practice
multiplications before the start of the multiplication block. Children were asked to say the
answer as quickly as they could without making any mistakes. They were also told that if
they didn’t already know the answer, they could use whatever way they found easiest to
work it out. Reaction latencies were recorded using a voice-activated key, and errors were
recorded by the experimenter.
2.3. Basic number processing skills
2.3.1. Number reading and naming
1- and 2-digit numbers were presented individually on the computer. Children were
asked to name the numbers as quickly as they could without making any mistakes, and
response latencies were recorded using a voice-activated key. Errors were recorded by
K. Landerl et al. / Cognition 93 (2004) 99–125 111
the experimenter. Six practice items were given, followed by a block of 18 single-digit
items (each of the numbers from 1 to 9 appeared twice in a pseudo-random order with the
proviso that no item appeared twice in succession). A block of 16 2-digit items was then
presented. Numbers were between 20 and 90, avoiding teens.
Colour naming was used as a non-numerical control for the number naming tasks.
Rectangles of five different colours (red, green, brown, blue and black) were presented on
the computer. Each colour was presented four times in pseudo-random order with the
proviso that no item appeared twice in succession. Five practice items were given.
Response latencies were measured using a voice key, and errors were noted by the
experimenter.
A list of eight 3-digit numbers was also presented. These numbers were presented
simultaneously on a sheet of paper, and children were asked to read them as quickly as
they could without making any mistakes. A stopwatch was used to measure the time taken
by the child to read the whole list. Simultaneous presentation was used for this task in
order to record the amount of time taken to articulate 3-digit numbers: a voice key
measures only speech onset.
2.3.2. Number comparison
Children were presented with two digits (1 – 9, not including 5) on the computer, one to
the left and one to the right of the screen. The digits could vary along two dimensions:
numerical size and physical size. In one variant of the task children were asked to select
the numerically largest number, in the other variant they were asked to select the
physically largest number. Congruence between physical and numerical size was
counterbalanced; so that there were 12 congruent trials, 12 neutral trials (in which either
the numerical or the physical size of the two numbers was the same) and 12 incongruent
trials for each variant of the task. There were two numerical size differences: 1 (e.g. 2, 3)
and 5 (e.g. 7, 2) and two physical sizes: 0.3 £0.5 cm, for small numbers, and 0.6 £1cm
for large numbers. Items were presented in a pseudo-random order. The numbers were
presented in an 11 £8.5 cm field in the centre of the screen. Each item was preceded by a
fixation cross that lasted for 300 ms. Response keys were ‘f’ and ‘J’ on the computer
keyboard. Children were asked to hit the key to the same side (left or right) to the side
where the larger number had appeared. Reaction times and errors were recorded by the
computer.
Order of presentation of the two variants of the task was counterbalanced as far as
possible, each LD child being presented with the opposite order to the last, and each
control child being presented with the same order as the matched LD child.
2.3.3. Number writing
The experimenter dictated a number to the child who immediately wrote it on a piece of
paper. 10 single-digit numbers (in random order); eight 2-digit numbers and eight 3-digit
numbers were presented. Errors were scored as substitutions, reversals, or place-value
errors. An error was scored as a substitution if an alternative digit was substituted for the
correct one (e.g. 26 written as 22). Reversals were when a digit was written the wrong way
around (e.g. 3 written as
1
). Place value errors involved syntax confusion (e.g. 724 written
as 7024).
K. Landerl et al. / Cognition 93 (2004) 99–125112
2.3.4. Number sequences/counting
Children were asked to count as quickly as they could from 1 to 20; 45 to 65; 2 to 20 in
twos; and 20 to 1 backwards. Each counting task was timed using a stopwatch from the
first to the last item. Errors were noted.
2.3.5. Dot counting
Groups of randomly arranged dots ranging from 1 to 10 were presented on the
computer. Children were asked to count the dots and to respond as quickly as they could
without making any mistakes. A voice key recorded reaction times. Errors were recorded
by the experimenter. 20 trials were presented altogether, with each number from 1 to 9
being represented twice in a pseudo-random order with the proviso that no item occurred
twice in succession.
2.3.6. Procedure
Head teachers and special needs co-ordinators were contacted either directly or via
their Local Education Authority. Those who agreed to take part in the study asked teachers
in their school to nominate children according to the criteria, and sent consent forms to
parents. Children were tested individually in a quiet room in their school. Testing was
done as far as possible in a single session, otherwise it was completed the next day.
Sessions generally lasted from 1 to 2 h, depending on the ability of the child.
3. Results
The data were analysed using separate ANCOVAs (with age as a covariate) for each
task. When indicated by statistically significant effects in the main analysis, planned
comparisons (also controlling for age) were carried out between the control group and the
two groups with number fact deficits (dyscalculic and double deficit), between the control
group and the dyslexic group and between the dyscalculic group and the double deficit
group.
3.1. Standardized tests
Mean scores and standard deviations for the standardized tests are presented in Table 2.
Digit span. In both conditions (forward and backward), Table 2 shows a trend of lower
performance of the two reading disabled groups (dyslexic and double deficit) compared to
the two other groups. However, in a repeated measures ANCOVA (controlling for age) the
group effect did not reach statistical significance (Fð3;38Þ¼2:5;P¼0:08), and neither
did the interaction between group and condition (Fð3;38Þ¼0:1;P.0:1).
Mazes. There was also no statistically significant difference between groups on the
Mazes.
Vocabulary. With most children, the British Picture Vocabulary Test (Dunn et al.,
1997) could be carried out. In an ANCOVA controlling for age no group differences could
be found. For 7 children (1 control, 2 dyscalculic and 4 double deficit children) recent
scores of the WISC-vocabulary scales were available so that carrying out the BPVS was
K. Landerl et al. / Cognition 93 (2004) 99–125 113
superfluous. All of these children had scaled scores above 10, once again confirming that
there were no serious verbal deficits.
3.1.1. Mental arithmetic
Errors. A repeated measures ANCOVA, controlling for age, was performed on
errors with each operation (addition, subtraction and multiplication) as a separate
level. There was a significant effect of level, Fð2;86Þ¼6:2;P¼0:003:Examination
of the mean scores presented in Fig. 1 indicates that all groups were less accurate
with multiplication than with addition or subtraction. There was also an interaction
Table 2
Mean scores (standard deviations) for standardized tests
Control
ðN¼18Þ
Dyslexic
ðN¼10Þ
Dyscalculic
ðN¼10Þ
Double deficit
ðN¼11Þ
Digit span forward
(raw score)
8.8 (l.5) 7.7 (0.7) 8.7 (2.6) 7.6 (l.1)
Digit span backward
(raw score)
4.5 (l.4) 3.7 (l.1) 4.3 (l.3) 3.6 (l.5)
Mazes (raw score) 18.1 (3.6) 19.0 (2.3) 18.7 (3.8) 17.0 (3.8)
Vocabulary
(BPVS raw scores)
88.1 (14.3)
a
89.3 (12.8) 91.3 (3.9)
b
78.6 (13.5)
c
a
N¼17;one child had a WISC vocabulary score.
b
N¼8;two children had WISC vocabulary scores.
c
N¼7;four children had WISC vocabulary scores.
Fig. 1. Mental arithmetic: mean number (SE) of correct responses for each operation.
K. Landerl et al. / Cognition 93 (2004) 99–125114
between level and group, Fð6;86Þ¼3:2;P¼0:007:Examination of the mean scores
for the groups suggests that the double deficit and dyscalculic groups had relatively
greater difficulty with multiplications than the other two groups. As expected there
was a difference between groups, Fð3;43Þ¼10:7;P,0:001:Planned contrasts
indicated that the double deficit group were less accurate than controls on subtraction
and multiplication ðP,0:001Þbut not addition. The dyscalculic group was less
accurate than controls on addition and subtraction ðP¼0:01Þand multiplication ðP,
0:001Þ:No reliable differences were found between dyslexic and control groups and
between dyscalculic and double deficit groups.
RTs. The median reaction time (for correct answers only) was calculated for each child.
As response accuracy was high for additions and subtractions, the median RTs for these
conditions are on average based on a reasonable number of 10 responses per child. The
average number of correct responses—and therefore the number of RTs was clearly lower.
Three children (all from the double deficit group) failed to answer any multiplication
questions correctly and had no valid reaction time for multiplication. In order not to lose
them from the repeated measures ANCOVA (controlling for age and with addition,
subtraction and multiplication as separate levels) their multiplication RT was replaced by
the mean score for the double deficit group.
First inspection of the descriptive statistics (see Fig. 2) showed that the variability
was high, especially for the dyscalculic and double deficit groups. Therefore, the
statistical analysis was carried out on log transformed scores. There was no statistically
significant main effect of level and no interaction between level and group. As
expected, there was a significant difference between groups, Fð3;43Þ¼12:2;
Fig. 2. Mental arithmetic: mean RTs (SE) for correct responses for each operation.
K. Landerl et al. / Cognition 93 (2004) 99–125 115
P,0:001:The double deficit group was slower than controls on addition and
subtraction ðP,0:001Þand multiplication ðP¼0:03Þ:The dyscalulic group was also
slower than controls on all operations ðP,0:005Þ:The dyslexic group was not
statistically different from controls and the dyscalculic group was not statistically
different from the double deficit group.
3.2. Basic number processing skills
3.2.1. Number naming and reading
There was no attempt at statistical analysis of errors for these tasks, due to the very low
proportion of errors made.
Median reaction times for correct answers were calculated for each child for 1-digit
number naming, 2-digit number naming, and colour naming (see Fig. 3). As response
accuracy was high, these median RTs are based on a reasonable number of responses in
each condition (18 1-digit numbers, 16 2-digit numbers and 20 colour patches). A repeated
measures ANCOVA controlling for age was performed on the data, with 1-digit number
naming, 2-digit number naming and colour naming as separate levels. There was no main
effect of level, and no effect of group. However, there was a statistically significant
interaction between level and group: Fð6;0:88Þ¼3:1;P¼0:009:Examination of the
mean scores suggested that while the dyslexic group was generally slower at naming than
controls, the dyscalculic and double deficit groups appeared to have a particular problem
with 2-digit number naming.
Fig. 3. Naming: mean RTs (SE) for 1- and 2-digit numbers and colours.
K. Landerl et al. / Cognition 93 (2004) 99–125116
This interaction was investigated further by reanalysing the number naming data with
colour naming entered as a covariate. This analysis compares number naming speed across
groups whilst controlling for factors such as speed of processing, articulation and access to
semantic memory (processes which are also involved in colour naming). A repeated
measures ANCOVA with 1- and 2-digit number naming as separate levels and colour
naming and age as covariates was performed. Now the only significant effect was group:
Fð3;43Þ¼3:3;P¼0:03:Contrasts between each group and the control group indicated
that the double deficit and the dyscalculic group were slower at number naming than the
control group ðP,0:05Þ;even with colour naming controlled. The dyslexic group,
however, was no different from controls on number naming once general naming ability
was controlled. In a final planned contrast it was confirmed that the dyscalculic group did
not differ from the double deficit group.
From the analysis of reading a list of eight 3-digit numbers as quickly as possible one
child from the dyscalculic group was excluded due to missing data. A one-way ANCOVA
controlling for age was performed on the reading times taken. There was a statistically
significant difference between groups: Fð3;47Þ¼4:2;P¼0:01:Contrasts ðP,0:05Þ
indicated that the double deficit group (M¼20:3s;SD ¼4.6) and the dyscalculic group
(M¼20:7s;SD ¼6.7) did not differ from each other but both showed higher number
reading times than controls (M¼15:0s;SD ¼4.1). The dyslexic group (M¼17:8s;
SD ¼4.5) performed not reliably different from the control group.
3.2.2. Number comparison
Median reaction times for correct answers were calculated for each child for number
comparison and size comparison (Fig. 4). A repeated-measures ANCOVA controlling for
age was performed on the data. There was a main effect of task (numerical or physical):
Fð1;43Þ¼4:1;P¼0:05:Comparison of the means indicates that physical comparison
was faster than numerical comparison. There was also an interaction between condition
and group: Fð3;43Þ¼6:0;P¼0:002:Contrasts indicated that the double deficit and the
dyscalculic group did not differ from each other in any condition but were both slower than
controls on the numerical comparison task ðP,0:05Þthough not the physical comparison
task. The dyslexic group did not differ from controls on either task. An analysis of the
errors found no effect of task, and no differences between groups.
In the numerical condition an effect of congruency of physical and numerical size could
be observed (Fð1;43Þ¼6:0;P¼0:02), however, this effect did not interact with group.
Planned comparisons showed that RTs for congruent items (1186 ms) were lower than for
neutral (1340 ms, tð47Þ¼5:1;P,0:001) and incongruent items (1362 ms, tð47Þ¼8:0;
P,0:001). For the physical condition no effects of congruency were found. In the
numerical condition, RTs for items with a larger numerical difference (1199 ms) were
lower than for items with a smaller numerical difference (1415), however, this difference
was not reliable and did not interact with group.
3.2.3. Number writing
Nine children committed one or more substitution errors. Four were from the double
deficit group (36%); 2 from each of the dyscalculic and dyslexic groups (20%) and 1 from
K. Landerl et al. / Cognition 93 (2004) 99–125 117
the control group (6%). An independent samples ANCOVA was performed on the
substitutions but did not reach statistical significance.
No statistical analysis was performed on reversals and place value errors, due to the fact
that no errors of these types were made by any of the children in the control or dyslexic
groups. Variance was thus far from equal between groups. 11 children in total reversed
numbers; of these 5 were in the double deficit group (45%), 6 in the dyscalculic group
(60%). Five children made place value errors: 1 was in the dyscalculic group (10%) and 4
were in the double deficit group (36%).
3.2.4. Number sequences/counting
One dyscalculic child was excluded from this analysis because of missing data. A
repeated-measures ANCOVA controlling for age was performed on the times taken by
each child to count in the four conditions: from 1 to 20, from 45 to 65, from 2 to 20 in twos,
and from 20 to 1 backwards (see Fig. 5). There was no main effect of counting condition,
but there was an effect of group: Fð3;43Þ¼7:0;P¼0:001;and an interaction between
group and condition: Fð9;129Þ¼2:3;P¼0:021:Contrasts indicated that the double
deficit and the dyscalculic group did not differ from each other and were both slower than
controls in all conditions (P,0:007 except for dyscalculics in the 1 –20 condition when
P¼0:01). There was a trend for dyslexic children to be slower at counting although this
Fig. 4. Size and number comparison: mean RTs (SE) for correct responses.
K. Landerl et al. / Cognition 93 (2004) 99–125118
only reached statistical significance in two conditions: counting 2–20 in 2 s ðP¼0:02Þ
and counting from 45 to 65 ðP¼0:04Þ:
3.2.5. Dot counting
Only 7% of all responses were miscounts, almost without exception for six or more
dots, deviating from the correct number of dots by one or two. In a non-parametric
Kruskal– Wallis test, no group differences were found.
Altogether, the analysis of RTs for correct responses is based on 847 RTs, 17 responses
per child on average. Reliability proved to be high with 0.82. A first inspection of the mean
scores showed that for the low range of one to three dots comparably small differences
between RTs could be observed while from four dots on a larger and systematic increase of
RTs was evident for increasing dot numbers. This was taken as indication for a subitizing
range (1– 3 dots) and a counting range (4 – 10 dots). For theoretical reasons and because
the distribution of RTs for these two ranges were very different, subitizing range and
counting range were analysed separately. For both ranges, the best fitting regression lines
were calculated for each child. Fig. 6 presents the mean regression lines for each group for
the subitizing and the counting range.
ANCOVAs with age and colour naming as covariates were carried out on the slopes
and intercepts of both ranges. Colour naming was introduced as a covariate to control
for general differences in naming speed. Fig. 6 shows that within the subitizing range,
the slopes for the dyscalculic, and the double deficit group are steeper than those of
Fig. 5. Number sequences: mean response times (SE) for counting from 1 to 20, 45– 65, 2– 20 in twos and 20 – 1
backwards.
K. Landerl et al. / Cognition 93 (2004) 99–125 119
the dyslexic and the control groups; however, this was not statistically reliable. For the
counting range, the group difference was marginally significant (Fð3;47Þ¼2:7;
P¼0:06). Fig. 6 indicates that this is due to a steeper slope of dyscalculic compared to
control children. There was no difference between the intercepts of the four groups, neither
for subitizing nor for counting range.
Fig. 6. Dot counting: regression lines for subitizing and counting range.
K. Landerl et al. / Cognition 93 (2004) 99–125120
4. General discussion
In summary, the dyscalculic children identified in this study demonstrated general
deficits in number processing, including accessing verbal and semantic numerical
information, counting dots, reciting number sequences and writing numbers. Despite this,
dyscalculic children without reading disability were normal or above average on tasks
involving phonological working memory, accessing non-numerical verbal information,
non-verbal intelligence, language abilities, and psychomotor abilities. As predicted, this
pattern of deficits is at once too broad and too specific to be readily explainable in terms of
general spatial, verbal or memory abilities. In terms of these results, dyscalculia can best
be defined as a deficit in the representation or processing of specifically numerical
information.
Neuropsychological evidence indicates that numerical processing is localized to the
parietal lobes bilaterally, in particular the intra-parietal sulcus (Dehaene, Piazza, Pinel, &
Cohen, 2003), and is independent of other abilities. Developmental dyscalculia is likely to
be the result of the failure of these brain areas to develop normally, whether because of
injury or because of genetic factors.
Children with reading disability only performed similarly to controls on the numerical
tasks. They were slower than controls in reciting number sequences (although less so than
dyscalculic children) and there were non-significant trends towards slowness in number
reading and number naming. However, unlike with the dyscalculic groups, the number
naming trend disappeared once general naming ability was controlled for. Dyslexic
children were also identical to controls on non-verbal (or non-phonological) tasks such as
number writing and number comparison. This pattern of results suggests that children with
reading difficulties only do not have number processing deficits, although difficulties with
verbal or phonological aspects of some of these tasks may have affected their performance.
The patterns of performance of the two dyscalculic groups on the numerical tasks were
very similar. This study has found no evidence for a qualitative difference in the numerical
abilities of dyscalculic children with and without reading disabilities. In many tasks the
double deficit group’s performance was slower or more error-prone than that of the
dyscalculic group, suggesting that their difficulties may be more severe; this is in keeping
the findings of Jordan and Montani (1997) and Shalev et al. (1997). However, the pattern
of impairment was the same for both groups: they each appeared to struggle with every
aspect of numerical processing tested in this study. This finding is further evidence against
theories which posit differential causes for different subgroups of dyscalculic children.
There are interesting parallels with and divergences from recent studies by Jordan and
colleagues (Jordan et al., 2003a,b) of second and third grade children using “achievement
groups” with similar names to our own: “math difficulties”, “math and reading
difficulties”, “reading difficulties”, and “normal achievement”. However, their groups
were defined by the 35th percentile, a far more inclusive criterion than the highest
estimates for dyscalculia prevalence (see above for a discussion). Their affected groups
would therefore include children whose mathematics difficulties are not severe enough to
fall outside the usual definition of the average range, as well as some proportion of
dyscalculics as determined by more stringent criteria, and a large proportion of children
whose maths achievement is likely to be poor for a wide variety of reasons.
K. Landerl et al. / Cognition 93 (2004) 99–125 121
One important finding of Jordan et al.’s studies is that the growth curves of performance
on mathematics achievement tasks are similar for all groups. This is in contrast with the
findings that dyscalculia is a persistent deficit involving the maintenance of immature
strategies, which causes dyscalculic children to fall farther and farther behind their peers
(e.g. Ostad, 1998, p. 359; Ostad, 1999, p. 360).
A second important finding was that there were no differences between groups with and
without reading difficulties on fact retrieval—which supports our general claim about the
functional independence of learning to read and learning arithmetic, and that fact retrieval
is not, in essence, a verbally mediated process. Indeed, they write that “our data do not
support the suggestion that difficulties in reading and fact retrieval share a core-underlying
deficit related to phonological processing”. (Jordan et al., 2003a, p. 24)
In five of their seven tasks, IQ “added little information beyond our initial group
classification” (p. 26) reinforcing our own conclusion that low general cognitive ability is
neither necessary nor sufficient to cause dyscalculia.
We suggest that the key deficit in developmental dyscalculia is a failure to represent
and process numerosity in a normal way. Numerical expressions do not seem to have the
same meaning for these children, as is evidenced by the relative difficulty they have with
the number comparison and dot counting. Failure to develop normal representations may
account for the difficulty experienced by dyscalculic children in memorizing arithmetical
facts: these facts lack meaning for them, or, at least, they do not carry the usual systematic
meanings that make for well-ordered and accessible memories. This account would be
consistent with the finding that there are no general short-term or long-term memory
deficits observed in the DC children. However, this proposal requires further investigation.
In conclusion, the most likely candidate for an underlying cause of dyscalculia is a
congenital failure to understand basic numerical concepts, especially the idea of
numerosity, a capacity which is independent of other abilities. This is revealed by deficits
in very basic numerical capacities, dot counting, and number comparison; there was also a
trend towards a difference in subitizing. This study has demonstrated that dyscalculic
children without verbal or psychomotor difficulties have a range of numerical difficulties
relative to controls. (The small sample-size defined by tight criteria means that the
possibility cannot be excluded that there are mathematical difficulties with a different
cognitive basis.)
We suggest that lack of understanding of numerosity, and a poor capacity to recognize
and discriminate small numerosities—as revealed in performance on dot tasks—may
prevent dyscalculics developing the normal meanings for numerical expressions and lead
to their difficulties in learning and retaining information regarding numbers. Geary (1993)
proposes that dyscalculic children may well have trouble with representation of numbers
in semantic memory, although we would argue that this does not imply anything about
their semantic memory for anything else.
We suggest that future research on dyscalculia should focus upon the numerical basis of
dyscalculia rather than upon its correlates in other cognitive domains. The study described
here was based upon small groups of children, and the definition of “dyscalculia” that we
used was based upon difficulties specifically with number facts. Further research may
identify other subgroups of dyscalculic, who may display other patterns of disability.
K. Landerl et al. / Cognition 93 (2004) 99–125122
However, on the basis of the evidence so far, numerical abilities should take centre stage in
further research into the nature of dyscalculia.
Acknowledgements
This research was supported by the Department for Education and Skills SEN Small
Programmes grant to UCL and the British Dyslexia Association and an APART grant of
the Austrian Academy of Science to Karin Landerl. We thank Dr Lindsay Peer of the
British Dyslexia Association and Eva-Maria Ebner for their contributions to this project.
We are grateful to the students and teachers of the participating schools in the London
Boroughs of Harrow and Camden.
References
American Psychiatric Association (1994). Diagnostic and statistical manual of mental disorders (4th ed).
Washington, DC: American Psychiatric Association.
Ashcraft, M. H., Yamashita, T. S., & Aram, D. M. (1992). Mathematics performance in left and right brain-
lesioned children and adolescents. Brain and Cognition,19, 208–252.
Baddeley, A. D., Lewis, V., & Vallar, G. (1984). Exploring the articulatory loop. Journal of Experimental
Psychology,36A, 233– 252.
Badian, N. A. (1983). Arithmetic and nonverbal learning. In H. R. Myklebust (Ed.), (Vol. 5) (pp. 235– 264).
Progress in learning disabilities, New York: Grune and Stratton.
Bisanz, J. (1999a). The development of mathematical cognition: Arithmetic. Journal of Experimental Child
Psychology,74, 153– 156.
Butterworth, B. (1999). The mathematical brain. London: Macmillan.
Butterworth, B., Cipolotti, L., & Warrington, E. K. (1996). Short-term memory impairments and arithmetical
ability. Quarterly Journal of Experimental Psychology,49A, 251– 262.
Butterworth, B., Grana, A., Piazza, M., Girelli, L., Price, C., & Skuse, D. (1999). Language and the origins of
number skills: karyotypic differences in Turner’s syndrome. Brain and Language,69(3), 486– 488.
Cappelletti, M., Butterworth, B., & Kopelman, M. D. (2001). Spared numerical abilities in a case of semantic
dementia. Neuropsychologia,39/11, 1224– 1239.
Cohen, L., Dehaene, S., Cochon, F., Lehericy, S., & Naccache, L. (2000). Language and calculation within the
parietal lobe: A combined cognitive, anatomical and fMRI study. Neuropsychologia,38, 1426– 1440.
Dehaene, S., Dehaene-Lambertz, G., & Cohen, L. (1998). Abstract representations of numbers in the animal and
human brain. Trends in Neurosciences,21(8), 355– 361.
Dehaene, S., Piazza, M., Pinel, P., & Cohen, L. (2003). Three parietal circuits for number processing. Cognitive
Neuropsychology,20, 487– 506.
Dunn, L. M., Dunn, L. M., & Whetton, C. (1997). The British picture vocabulary scales (2nd ed.). Windsor:
NFER-Nelson.
Elliot, C., Smith, P., & McCulloch, K. (1997). British ability scales II. Windsor: NFER-Nelson.
Fayol, M., Barrouillet, P., & Marinthe, C. (1998). Predicting arithmetical achievement from neuro-psychological
performance: A longitudinal study. Cognition,68, 63– 70.
Fletcher, J. F. (1985). Memory for verbal and nonverbal stimuli in learning disabled subgroups: Analysis by
selective reminding. Journal of Experimental Child Psychology,40, 244–259.
Gallistel, C. C. (1990). The organization of learning. Cambridge, MA: MIT Press.
Geary, D. C. (1993). Mathematical disabilities: Cognition, neuropsychological and genetic components.
Psychological Bulletin,114, 345– 362.
K. Landerl et al. / Cognition 93 (2004) 99–125 123
Geary, D. C., Bow-Thomas, C. C., & Yao, Y. (1992). Counting knowledge and skill in cognitive addition: A
comparison of normal and mathematically disabled children. Journal of Experimental Child Psychology,54,
372– 391.
Geary, D. C., Hamson, C. O., & Hoard, M. K. (2000). Numerical and arithmetical cognition: A longitudinal study
of process and concept deficits in children with learning disability. Journal of Experimental Child
Psychology,77, 236– 263.
Geary, D. C., & Hoard, M. K. (2001). Numerical and arithmetical deficits in learning-disabled children: Relation
to dyscalculia and dyslexia. Aphasiology,15(7), 635– 647.
Geary, D. C., Hoard, M. K., & Hamson, C. O. (1999). Numerical and arithmetical cognition: Patterns of functions
and deficits in children at risk for a mathematical disability. Journal of Experimental Child Psychology,74,
213– 239.
Gelman, R., & Gallistel, C. R. (1978). The child’s understanding of number (1986 ed.). Cambridge, MA: Harvard
University Press.
Gerstmann, J. (1940). Syndrome of Finger Agnosia: disorientation for right and left, agraphia and acalculia.
Archives of Neurology and Psychiatry,44, 398–408.
Ginsburg, H. P. (1997). Mathematics learning disabilities: A view from developmental psychology. Journal of
Learning Disabilities,30, 20– 33.
Hanich, L. B., Jordan, N. C., Kaplan, D., & Dick, J. (2001). Performance across different areas of mathematical
cognition in children with learning difficulties. Journal of Educational Psychology,93, 615– 626.
Jordan, N., Hanich, L. B., & Kaplan, D. (2003a). A longitudinal study of mathematical competencies in children
with specific mathematics difficulties versus children with co-morbid mathematics and reading difficulties.
Child Development,74, 834– 850.
Jordan, N., Hanich, L. B., & Kaplan, D. (2003b). Arithmetic fact mastery in young children: A longitudinal
investigation. Journal of Experimental Child Psychology,85, 103– 119.
Jordan, N., Kaplan, D., & Hanich, L. B. (2002). Achievement growth in children with learning difficulties in
mathematics: Findings of a two-year longitudinal study. Journal of Educational Psychology,94, 586 – 597.
Jordan, N., & Montani, T. (1997). Cognitive arithmetic and problem solving: A comparison of children with
specific and general mathematics difficulties. Journal of Learning Disabilities,30(6), 624– 634.
Kinsbourne, M., & Warrington, E. K. (1963). The developmental Gerstmann Syndrome. Annals of Neurology,8,
490– 501.
Kirby, J. R., & Becker, L. D. (1988). Cognitive components of learning problems in arithmetic. Remedial and
Special Education,9, 7– 16.
Koontz, K. L., & Berch, D. B. (1996). Identifying simple numerical stimuli: Processing inefficiencies exhibited by
arithmetic learning disabled children. Mathematical Cognition,2(1), 1– 23.
Lewis, C., Hitch, G., & Walker, P. (1994). The prevalence of specific arithmetic difficulties and specific reading
difficulties in 9- and 10-year old boys and girls. Journal of Child Psychology and Psychiatry,35, 283– 292.
McLean, J. F., & Hitch, G. J. (1999). Working memory impairments in children with specific arithmetical
difficulties. Journal of Experimental Child Psychology,74, 240–260.
Ostad, S. E. (1998). Comorbidity between mathematics and spelling difficulties. Logopedics Phoniatrics
Vocology,23, 145– 154.
Ostad, S. E. (1999). Developmental Progression of subtraction studies: a comparison of mathematically normal
and mathematically disabled children. European Journal of Special Needs Education,14(1), 21–36.
Piaget, J. (1952). The child’s conception of number. London: Routledge & Kegan.
Raven, J. C., Court, J. H., & Raven, J. (1986). Manual for Raven progressive matrices and vocabulary scales.
London: Lewis.
Rosenberg, P. B. (1989). Perceptual-motor and attentional correlates of developmental dyscalculia. Annals of
Neurology,26, 216– 220.
Rourke, B. (1989). Nonverbal learning disabilities. The syndrome and the model. New York: Guilford Press.
Rourke, B. P. (1993). Arithmetic disabilities, specific and otherwise: A neuropsychological perspective. Journal
of Learning Disabilities,26, 214– 226.
Russell, R. L., & Ginsburg, H. P. (1984). Cognitive analysis of children’s mathematical difficulties. Cognition
and Instruction,1, 217– 244.
K. Landerl et al. / Cognition 93 (2004) 99–125124
Shalev, R. S., & Gross-Tsur, V. (1993). Developmental dyscalculia and medical assessment. Journal of Learning
Disabilities,26, 134– 137.
Shalev, R. S., & Gross-Tsur, V. (2001). Developmental dyscalculia. Review article. Pediatric Neurology,24,
337– 342.
Shalev, R. S., Manor, O., & Gross-Tsur, V. (1997). Neuropsychological aspects of developmental dyscalculia.
Mathematical Cognition,3(2), 105– 120.
Siegel, L. S., & Feldman, W. (1983). Non-dyslexic children with combined writing and arithmetic difficulties.
Clinical Pediatrics,22, 241– 244.
Siegel, L. S., & Ryan, E. B. (1989). The development of working memory in normally achieving and subtypes of
learning disabled children. Child Development,60, 973– 980.
Spellacy, F., & Peter, B. (1978). Dyscalculia and elements of the developmental Gerstmann syndrome in school
children. Cortex,14(2), 197–206.
Starkey, P., Spelke, E. S., & Gelman, R. (1990). Numerical abstraction by human infants. Cognition,36, 97– 128.
Temple, C. (1991). Procedural dyscalculia and number fact dyscalculia: Double dissociation in developmental
dyscalculia. Cognitive Neuropsychology,8, 155– 176.
Temple, C. M., & Sherwood, S. (2002). Representation and retrieval of arithmetical facts: Developmental
difficulties. Quarterly Journal of Experimental Psychology,55A(3), 733– 752.
Thioux, M., Seron, X., & Pesenti, M. (1999). Functional neuroanatomy of the semantic system: The case for
numerals. Brain and Language,69(3), 488– 490.
Wechsler, D. (1992). WISC III UK. Sidcup, Kent: The Psychological Corporation.
Woodcock, R. W., McGrew, K. S., & Mather, N. (2001). Woodcock-Johnson III. Itasca, IL: Riverside Publishing.
Wynn, K. (1992). Addition and subtraction by human infants. Nature,358, 749– 751.
Wynn, K. (1995). The origins of numerical knowledge. Mathematical Cognition,1, 35 – 60.
K. Landerl et al. / Cognition 93 (2004) 99–125 125
