Content uploaded by Gavin Price
Author content
All content in this area was uploaded by Gavin Price
Content may be subject to copyright.
Behavioral/Cognitive
Why Mental Arithmetic Counts: Brain Activation during
Single Digit Arithmetic Predicts High School Math Scores
Gavin R. Price,
1
Michèle M. M. Mazzocco,
2,3
and Daniel Ansari
1
1
Numerical Cognition Laboratory, Department of Psychology and Brain and Mind Institute, University of Western Ontario, London, Ontario, Canada N6G
2K3,
2
Department of Psychiatry and Behavioral Sciences, and School of Education, Johns Hopkins University, Baltimore, Maryland 21287, and
3
Institute of
Child Development, University of Minnesota, Minneapolis, Minnesota 55455
Do individual differences in the brain mechanisms for arithmetic underlie variability in high school mathematical competence? Using
functional magnetic resonance imaging, we correlated brain responses to single digit calculation with standard scores on the Preliminary
Scholastic Aptitude Test (PSAT) math subtest in high school seniors. PSAT math scores, while controlling for PSAT Critical Reading
scores, correlated positively with calculation activation in the left supramarginal gyrus and bilateral anterior cingulate cortex, brain
regions known to be engaged during arithmetic fact retrieval. At the same time, greater activation in the right intraparietal sulcus during
calculation, a region established to be involved in numerical quantity processing, was related to lower PSAT math scores. These data
reveal that the relative engagement of brain mechanisms associated with procedural versus memory-based calculation of single-digit
arithmetic problems is related to high school level mathematical competence, highlighting the fundamental role that mental arithmetic
fluency plays in the acquisition of higher-level mathematical competence.
Introduction
School-entry math skills are a stronger predictor of later aca-
demic achievement than early reading or socio-emotional skills
(Duncan et al., 2007), and low mathematical competence is asso-
ciated with lower indices of life success (Parsons and Bynner,
2005). Improvements in mathematical competence are related to
growth of gross domestic product (Organisation for Economic
Co-operation and Development, 2010, p 17) and are identified as
essential for boosting U.S. global competitiveness (National
Academies, 2007, p 5). These factors demonstrate the fundamen-
tal significance of mathematical competence and highlight the
importance of identifying sources of its variability.
A potential source of individual differences in math competence
is the neural architecture supporting the performance of simple
arithmetic problem solving. Arithmetic fluency, the speed and effi-
ciency with which correct solutions to numerical computations are
generated, is thought to represent a scaffold upon which higher-level
mathematical skills are built. Initially, students rely on procedural
strategies, such as counting aloud, finger counting, or decomposi-
tion to solve calculations. These explicit procedures are gradually
replaced by more efficient strategies, such as the retrieval of solutions
from memory (Ashcraft, 1982). This shift toward memory-based
calculation is a hallmark of successful arithmetic development. In-
deed, children with mathematical learning difficulties exhibit imma-
ture procedural strategies and poor math fact performance
(Mazzocco et al., 2008) long after their typically developing peers
begin using fact retrieval (Geary, 1993). Thus, it appears that early
arithmetic abilities support the acquisition of higher mathematical
competence, yet little is known about whether individual differences
in arithmetic fluency continue to scaffold broader mathematical
competence throughout secondary school and, if so, what neural
mechanisms underlie this relationship.
By investigating whether the function of brain circuits under-
lying simple arithmetic problem solving are predictive of vari-
ability in mathematical achievement, it is possible to better
understand the mechanisms by which the proposed scaffolding
between arithmetic fluency and higher-level skills might occur. A
deeper understanding of such mechanisms will support the de-
velopment of educational interventions that optimally exploit the
neurocognitive architectures supporting math achievement and,
at the very least, provide some explanation for individual differ-
ences in mathematics achievement outcomes.
In the present study, we adopted an educational neuroscience
approach (Carew and Magsamen, 2010) using functional mag-
netic resonance imaging (fMRI) to investigate the relationship
between brain activation during single-digit arithmetic and
mathematical competence as measured by the Preliminary Scho-
lastic Aptitude Test (PSAT) Math subtest, a nationally adminis-
tered exam designed to predict college readiness.
Received June 20, 2012; revised Oct. 15, 2012; accepted Nov. 2, 2012.
Author contributions: G.R.P., M.M.M.M., and D.A. designed research; G.R.P., M.M.M.M., and D.A. performed
research; G.R.P. and D.A. analyzed data; G.R.P., M.M.M.M., and D.A. wrote the paper.
This research was supported by funding from the Canadian Institutes of Health Research (CIHR), The Natural
Sciences and Engineering Research Council of Canada (NSERC), the Canada Research Chairs Program (CRC) to D.A.,
and Ontario Ministry of Research and Innovation (OMRI) Postdoctoral Fellowship to G.R.P. We thank Carolyn Ahart,
who oversaw efforts to obtain the PSAT data, Kate Semeniak, for assistance with collection of behavioral data, and
Bea Goffin, for proofreading the manuscript.
The authors declare no competing financial interests.
Correspondence should be addressed to Daniel Ansari, Department of Psychology and Brain and Mind Institute,
University of Western Ontario, London, ON, Canada N6G 2K3. E-mail: daniel.ansari@uwo.ca.
G. R. Price’s present address: Department of Psychology and Human Development, Peabody College, Vanderbilt
University, Nashville, TN 37203.
M. M. M. Mazzocco’s present address: Institute of Child Development, University of Minnesota, Minneapolis, MN
55455.
DOI:10.1523/JNEUROSCI.2936-12.2013
Copyright © 2013 the authors 0270-6474/13/330156-08$15.00/0
156 •The Journal of Neuroscience, January 2, 2013 •33(1):156 –163
If arithmetic fluency serves as a scaffold for mathematical
competence, individual differences in performance on the PSAT
Math test should be associated with variation in the brain mech-
anisms associated with retrieval versus procedural calculation:
the left inferior parietal lobe (LIP) and bilateral intraparietal sul-
cus (IPS), respectively (Grabner et al., 2007;Grabner et al., 2009).
We predict that individuals with higher PSAT Math scores will
demonstrate increased activation of LIP regions during single
digit calculations compared to individuals with lower PSAT Math
scores, who are expected to exhibit greater activation of the IPS.
We predict that such individual differences in brain activation
patterns will be specific to PSAT Math and thus unrelated to
PSAT Critical Reading scores.
Materials and Methods
Participants
Participants were drawn from a larger, longitudinal study described else-
where in more detail (Mazzocco and Myers, 2003). When the cohort
reached 12th grade, we recruited a sample of these students representa-
tive of those with consistently deficient, low average, average, or above
average levels of mathematics achievement from kindergarten to grade 9.
A total of 43 participants took part in the fMRI experiment. From these
participants we requested and received authorization to obtain their of-
ficial PSAT score report from their high school registrar. Of those partic-
ipants whose data were not excluded due to excess motion (3 mm total
motion in a given run), 33 students sat for the PSAT in grade 10. Thus, a
total of 33 participants was included in the final analysis (14 females;
mean age: 17 years, 11.5 months).
Tasks
Tasks were presented in separate runs during the scanning session.
Arithmetic verification. Single digit arithmetic is an elementary math-
ematical ability, with improvements in efficiency already evident be-
tween first and second grade (Geary et al., 1991). Accordingly, it
represents an ideal task with which to investigate the neural mechanisms
of elementary arithmetic fluency and how these mechanisms relate to
individual differences in comprehensive mathematical achievement at
the end of secondary education.
Participants were presented with a series of single digit addition and
subtraction equations in the standard a⫹/⫺b⫽cformat (Fig. 1A), the
solution to which was either correct or incorrect (e.g., 5 ⫹3⫽8or5⫹
3⫽7). In the construction of arithmetic trials, all single digit numbers,
with the exception of 1, were used as either the left or right operand across
the paradigm, and solutions were always a single digit operand. Incorrect
solutions deviated from the correct solution by either ⫹1or⫺1, and
participants were required to indicate via button press whether the pre-
sented solution was correct or incorrect. This task, modeled on the task
reported by Rivera et al. (2005), had a total of 40 trials presented in a
single run comprising 20 subtraction and 20 addition trials. Subtraction
and addition trials were intermixed in a pseudorandom order so that the
same trial type never occurred for more than three consecutive trials.
Each trial was presented for 2 s, followed by a fixation screen comprising
a single white dot at the center of the screen (font size, 60). Poststimulus
fixation duration was6sonaverage but was varied between trials to
improve deconvolution of the hemodynamic response function (HRF).
Thus, interstimulus intervals (ISI) could be 4, 5, 6, 7, or 8 s, with a mean
ISI across the run of 6 s. Varying the ISI in this way ensures that stimulus
onset is not locked with the time to repeat (TR), as trial duration is not
consistently an integer multiple of the TR and therefore allows for over-
sampling of the HRF. ISI length and trial type (subtraction correct and
incorrect, addition correct and incorrect) were balanced such that no ISI
length was more frequently associated with a given trial type.
Digit matching. Participants were presented with three single digits
separated by an equal sign (⫽) and rotated 90° into a vertical rather than
horizontal orientation (Fig. 1B). Participants were required to indicate
via dual button press whether or not the third digit was identical to either
of the preceding digits. Each trial was presented for 2 s, followed by a
fixation screen comprising a single white dot at the center of the screen
(font size, 60). Stimulus duration and spacing were identical to that for
the Arithmetic Verification task described above.
Non-symbolic number comparison. The non-symbolic comparison
paradigm used in the present study was based on that reported by Hal-
berda et al. (2008). Participants were presented with a single array of blue
and yellow dots in intermixed locations (Fig. 2) and required to select, via
button press, whether there were more blue or more yellow dots in the
array. Trials varied according to the ratio between the dot sets (ratio
calculated as the larger number divided by the smaller number, so that in
a trial with 17 yellow dots and 13 blue dots, the ratio was 1.308). A total of
160 trials was presented across two runs, with the number of dots per
color ranging from 5 to 21, and ratios ranging from 1.182 to 3.6. Each
trial was presented for 500 ms. In half the trials the yellow dots were more
numerous, and in the other half the blue dots were more numerous. Trial
presentation order was randomized with respect to ratio, but fixed across
participants. Poststimulus fixation duration was6sonaverage but was
varied between trials to improve deconvolution of the HRF. Thus, an ISI
could be 4, 5, 6, 7, or 8 s, with a mean ISI across the run of 6 s. ISI length
and trial type (ratio) were balanced such that no ISI length was more
frequently associated with a given trial type. Following the method
described by Halberda et al. (2008) to limit the influence of non-
numerical continuous physical variables, the following controls were
put in place. For each ratio, half the trials were “dot-size controlled,”
meaning that the size of the average blue dot was equal to the size of
the average yellow dot. On these trials, the set with more dots neces-
sarily also had a larger total area on screen. The other half of trials
were “area controlled,” meaning that the total number of pixels for
blue and yellow dots was equal, resulting in an equivalent total surface
area for both sets of dots. Therefore, in this condition the more nu-
merous set therefore had a smaller average dot size.
PSAT. As our measure of mathematical competence, we used standard
scores from the PSAT Math subtest sat during grade 10. The PSAT math
subtest is part of a nationally administered test taken by over 3.5 million
high school students in the U.S.A. each year as reported by “Colleg-
eBoard.” It is designed to reliably predict college entrance exam scores
and serves as the qualifying test for the U.S. Merit-Based Scholarship
Program, and it is thus also known as the “National Merit Scholarship
Figure 1. Arithmetic verification (A) and digit matching (B) example stimuli and paradigm timing.
Price et al. •Arithmetic Brain Activation Predicts Math Skills J. Neurosci., January 2, 2013 •33(1):156 –163 • 157
Qualifying Test” (PSAT/NMSQT). Therefore,
performance on the PSAT is profoundly rele-
vant to higher education success among stu-
dents in the U.S. Most individuals who take the
PSAT are tenth graders, and in most states (in-
cluding Maryland, where most of the partici-
pants resided) tenth graders are enrolled in a
mathematics course. Beginning in Grade 11,
some students choose not to pursue elective
mathematics coursework (Updegraff et al.,
1996). Thus, Grade 10 PSAT Math was chosen
as a measure of broad achievement outcomes
at the latest school grade during which all par-
ticipants are likely to be receiving ongoing
math instruction.
The PSAT Math contains 38 items, includ-
ing word problems, geometry, algebraic equa-
tions, and complex (no single-digit simple
calculations) arithmetic, and it therefore repre-
sents a broad test of mathematical competence
of significant importance to an individual’s ac-
ademic success. As control measure for broad
academic achievement, we used standard
scores from the Grade 10 PSAT Critical Reading subtest. The PSAT Crit-
ical Reading includes reading comprehension questions about full-
length and paragraph-length passages, such as speculating on the origin
of the passage, as well as questions requiring students to fill in missing
words from a range of sentences.
fMRI acquisition parameters
All MR imaging was acquired with a 3T Phillips MRI scanner using an
8-channel head coil with parallel imaging capability. Using multislice 2D
SENSE T2* gradient-echo, echo planar imaging (EPI) pulse sequence,
functional images were obtained in the axial plane. Higher order shim-
ming was applied to the static magnetic field (B0). The EPI parameters
were as follows: echo time, 30 ms; TR, 2000 ms; flip angle, 75°; acquisition
matrix, 80 ⫻80 voxels; field of view (FOV), 240 mm; SENSE factor of 2.
This protocol acquired 34 axial brain slices per TR (3 mm thick slices with
1 mm slice gap) and a time course of 176 temporal whole brain image vol-
umes after discarding the first five volumes to ensure steady state. Anatom-
ical scan parameters were performed using an 8-channel head coil, 240 cm
FOV, anda1mmisotropic MP-RAGE (magnetization-prepared rapid ac-
quisition with gradient echo), which takes 6 min with SENSE factor 2. Axial
T1-weighted FSPGR (TR/TE, 215/12), axial diffusion-weighted (10,000/13),
and axial T2-weighted FRFSE with fat saturation (3440/68) fast spin echo
scans were obtained through the brain.
fMRI analyses
Structural and functional images were analyzed using Brain Voyager QX
2.4.1 (Brain Innovation). Any runs in which head motion exceeded a
total of 3 mm were excluded, which resulted in the exclusion of a total of
four runs across all participants. All participants were presented with and
completed a single run of arithmetic verification, a single run of digit
matching, and at least one run of nonsymbolic number comparison. The
remaining functional images were corrected for differences in slice time
acquisition, head motion, and low-frequency drifts in signal intensity
(high-pass filtering). In addition, functional images were spatially
smoothed witha6mmfull width at half maximum Gaussian smoothing
kernel. Following initial automatic alignment, the alignment of func-
tional images to the high resolution T1 structural images was manually
fine-tuned. The realigned functional dataset was then transformed
into Talairach space (Talairach and Tournoux, 1988). A two gamma
hemodynamic response function was used to model the expected BOLD
signal (Friston et al., 1998).
Statistical maps were corrected for multiple comparisons using false
discovery rate (FDR) in the case of whole brain, first level contrasts (e.g.,
arithmetic verification versus digit matching; see below, this section and
Results). In the case of second level covariate analyses (e.g., calculation
versus digit matching correlated with PSAT scores), where no voxels
survived correction using FDR, cluster-level correction (Forman et
al., 1995;Goebel et al., 2006) was used to correct for multiple com-
parisons. In this method, an initial voxel level (uncorrected) thresh-
old is set. Then, thresholded maps are submitted to a whole-slab
correction criterion based on the estimate of the map’s spatial
smoothness and on an iterative procedure (Monte Carlo simulation)
for estimating cluster-level, false-positive rates. After 1000 iterations,
the minimum cluster size that yielded a cluster level false-positive rate
(
␣
) of 0.05(0.5%) was used to threshold the statistical maps. Put
another way, this method calculates the size that a cluster would need
to be (the cluster threshold) to survive a correction for multiple com-
parisons at a given statistical level. Only activations whose size meets
or exceeds the cluster threshold are allowed to remain in the statistical
maps. Statistical maps resulting from first level contrasts (i.e., calcu-
lation ⬎digit matching) were thresholded at FDR corrected p⬍0.05.
Maps resulting from covariate analyses due to the more complex
nature of the analysis were thresholded at p⬍0.005 uncorrected and
then submitted to cluster-level correction as described above, yield-
ing clusters whose corrected significance level is p⬍0.05.
The analysis of nonsymbolic numerical comparison data were carried out
at the region of interest (ROI) level. ROIs were defined from the correlation
of brain activation during “arithmetic minus digit matching” with PSAT
Math scores residualized for PSAT Critical Reading. We performed an ROI
GLM in each region, testing for a parametric effect of ratio against baseline.
The parametric regressor was defined by assigning as a weight to each cor-
rectly solved trial the ratio between the two dot sets (larger set divided by
smaller set). In other words, this analysis tested whether a significant rela-
tionship between numerical ratio and brain activation (i.e., a positive corre-
lation between ratio and activation) existed in the regions identified by the
correlation between arithmetic verification-related activation and PSAT
Math scores.
Results
Behavioral data
The two primary behavioral variables of interest from the fMRI
tasks were reaction time (milliseconds) for correct responses and
percent accuracy across all trials. Paired ttests were used to com-
pare arithmetic verification and digit-matching performance.
Reaction time for correctly answered arithmetic verification
items (mean ⫽1540.57; SD ⫽398.61; range ⫽887.86 –2922.15)
was significantly longer than reaction time for digit-matching
items (mean ⫽1093.14; SD ⫽314.96; range ⫽683.04 –2542.44),
t
(32)
⫽10.11; p⬍0.001.
Arithmetic verification performance was also significantly
more accurate (mean ⫽96.21; SD ⫽2.43; range ⫽87.50 –100)
Figure 2. Nonsymbolic numerical magnitude comparison example stimuli and paradigm timing.
158 •J. Neurosci., January 2, 2013 •33(1):156 –163 Price et al. •Arithmetic Brain Activation Predicts Math Skills
than digit matching (mean ⫽92.12; SD ⫽2.35; range ⫽87.50 –
95), t
(32)
⫽6.89; p⬍0.001. Despite this difference, the mean
accuracy was high for both tasks.
The sample mean PSAT standard score (possible range ⫽20 –
80) was 49.15 (SD ⫽10.14; range ⫽35–72) for Math and 45.7
(SD ⫽9.412; range ⫽29 – 68) for Critical Reading. The national
norms (average scores) based on over 1.1 million tenth graders
who completed the PSAT in 2008, the same year as our sample,
were, 44.3 (SD ⫽11.1) for PSAT Math and 41.9 (SD ⫽11.4) for
PSAT Critical Reading (CollegeBoard, 2008). One-sample ttests,
comparing standard scores in the current sample to the national
average for the respective test, revealed that the mean PSAT Math
score in the current sample was significantly higher than the national
average for that year (t
(32)
⫽2.75; p⬍0.05), as was the PSAT Critical
Reading mean score (t
(32)
⫽2.32; p⬍0.05). The current sample was
representative of the normative variation, however, because the range of
scores observed in the current sample spanned ⬎3 SD, and their means
fell within 1 standard deviation of the national average score. Thus, while
on average the scores from our sample were higher than the national
norm, the scores were within the range of the national norms. Further-
more, the standard scores for PSAT math were normally distributed in
our sample (Shapiro–Wilk, p⫽0.074), as were the standard scores for
PSAT Critical Reading (Shapiro–Wilk; p⫽0.622).
To estimate behavioral performance specifically related to
arithmetic processing, we calculated difference scores by sub-
tracting accuracy or reaction time for digit matching from
accuracy or reaction time for arithmetic verification, respec-
tively (mean reaction time difference ⫽411.42; SD ⫽233.68;
range ⫽⫺2.39 –953.51; Shapiro–Wilk, p⫽0.772; mean accu-
racy difference ⫽411.42; SD ⫽3.41; range ⫽⫺7.5–10.00;
Shapiro–Wilk, p⫽0.001)). The difference scores served to
isolate variance in performance specific to calculation and
were thus closely aligned with the brain-imaging data de-
scribed subsequently.
Relationship to PSAT scores
Bivariate correlation analyses revealed no significant relationship be-
tween accuracy difference score and PSAT Math (r
(31)
⫽⫺0.11, p⬎
0.05) or PSAT Critical Reading (r
(31)
⫽⫺0.03) p⬎0.05. By
contrast, reaction time difference was negatively correlated
with PSAT Math (r
(31)
⫽⫺0.35; p⬍0.05), but for PSAT
Critical Reading the correlation was not significant (r
(31)
⫽
⫺0.29; p⬎0.05). However, the relationship between PSAT
Math and RT difference was no longer significant when con-
trolling for PSAT Critical Reading (r
(30)
⫽⫺0.25; p⬎0.05).
These results suggest that the reaction time difference between
arithmetic and digit matching captures variance associated
with cognitive processes common to PSAT Math and PSAT
Critical Reading, rather than processes specific to arithmetic.
Therefore, the relationship between calculation reaction time
and PSAT Math does not provide insight into any cognitive
mechanisms specific to math competence, but instead it re-
flects cognitive mechanisms general to academic achievement.
Indeed, from these behavioral data alone it could be con-
cluded that there is no domain-specific relationship between
performance on single-digit arithmetic verification and indi-
vidual differences on the PSAT Math test.
fMRI data
Calculation versus digit matching
To confirm that the current arithmetic verification task activated
typical calculation brain networks, we conducted whole brain
random effects General Linear Model testing for regions showing
greater activation for calculation relative to digit matching
(incorrect trials were modeled as separate predictors for both
conditions and excluded from further analysis). This analysis re-
vealed a number of regions, including the left intraparietal sulcus/
superior parietal lobe, bilateral insula, and bilateral superior
frontal gyri (p⬍0.05, FDR corrected; Table 1), many of which
are commonly found to be active during arithmetic verification
relative to control tasks (Rueckert et al., 1996;Menon et al.,
2000).
PSAT correlations
To create a measure of math competence controlling for the variance
related to reading ability (a nonmathematical academic domain), we
computed a linear regression with PSAT Math as the dependent
variable and PSAT Critical Reading as the independent variable to
derive residualized PSAT Math scores. We entered these residualized
PSAT Math scores (mean ⫽⫺3.03E⫺07; SD ⫽8.7; range ⫽
⫺15.42–27.65; Shapiro–Wilk, p⫽0.193) into a whole brain corre-
lation analysis, testing for an association between the residualized
PSAT Math scores and calculation-specific brain activation
(i.e., residualized PSAT Math scores were correlated with the
difference in brain activation between arithmetic verification
and digit matching).
This analysis revealed positive correlations between PSAT
Math and individual differences in the brain activation associated
with arithmetic verification (arithmetic verification ⬎digit
matching) in the left supramarginal gyrus (Talairach coordinates
(Tal): ⫺55, ⫺30, 30; k⫽959; Fig. 3) and the anterior cingulate
gyrus (Tal: 1, 23, 21, k⫽1090). In other words, greater activation
of these brain regions during calculation relative to digit match-
ing was associated with higher PSAT Math scores.
In addition, a negative correlation was revealed between
PSAT Math scores and arithmetic activation in the right intra-
parietal sulcus (Tal: 29, ⫺71, 41; k⫽583) (Fig. 4). Specifically,
those individuals with lower PSAT Math scores exhibited
Table 1. Significant regions of difference resulting from contrast of calculation and digit matching
Contrast Region
Peak Tal coordinate
(x,y,z)
Cluster size
(voxels)
Average
t-statistic
Calculation ⬎Control Right precentral Gyrus/inferior frontal gyrus 50, 4, 34 1923 4.19
Right insula 34, 19, 5 1917 4.79
Left insula ⫺31, 22, 6 929 4.09
Bilateral medial occipital gyrus 3, ⫺67, 2 2780 3.93
Bilateral thalamus/medial dorsal nucleus ⫺2, ⫺14, 11 2290 4.17
Bilateral superior frontal Gyrus/dorsal anterior cingulate cortex ⫺2, 4, 52 8878 4.50
Left superior frontal gyrus ⫺30, ⫺8, 64 506 3.88
Left intraparietal sulcus/ superior parietal ⫺32, ⫺62, 40 2569 4.01
Left precentral gyrus/Inferior frontal gyrus ⫺49, 3, 30 6594 4.31
Price et al. •Arithmetic Brain Activation Predicts Math Skills J. Neurosci., January 2, 2013 •33(1):156 –163 • 159
greater activation of the right IPS during
single digit arithmetic relative to digit
matching.
Several studies have shown that re-
gions of the left inferior parietal lobe,
including and proximal to the left su-
pramarginal gyrus (SMG) as well as the
anterior cingulate cortex (ACC), are as-
sociated with arithmetic fact retrieval
(Delazer et al., 2005;Grabner et al.,
2007;Grabner et al., 2009), while the
right IPS has been widely implicated in
the representation and processing of
numerical magnitude information
(Dehaene et al., 2003;Cohen Kadosh et
al., 2008) and is associated with proce-
dural problem solving strategies (De-
lazer, 2003;Delazer et al., 2005). Thus,
the current results suggest that individ-
uals with higher PSAT Math standard
scores are engaging neural mechanisms
associated with memory retrieval to
solve single-digit equations, while those
with lower scores are engaging systems as-
sociated with processing numerical quan-
tity, and likely relying on procedural
computations.
To further empirically constrain our
interpretation of this finding, we tested
the activation of the above regions in a
nonsymbolic numerical comparison
task completed by the same participants
during the same scanning session. Par-
ticipants were presented with an array of
blue and yellow dots and asked to decide
whether there were more blue or yellow
dots. The numerical ratio between the
blue and yellow dot sets was parametri-
cally varied, allowing us to test for the
“numerical ratio effect” reliably ob-
served at both the behavioral (Moyer
and Landauer, 1967) and brain levels
(Pinel et al., 2001;Holloway et al., 2010)
and used as a marker of basic numerical
magnitude processing. This analysis re-
vealed that activation strength in the
right IPS region, whose activity during
mental arithmetic negatively correlated
with PSAT Math scores, was parametrically
modulated by numerical ratio (t
(32)
⫽2.27;
p⫽0.03; see Materials and Methods for de-
tails). Specifically, this region showed
greater activation for trials in which the number of blue versus yellow
dots was harder to discriminate due to smaller ratio. By contrast, no
significant parametric ratio effect was observed in either the anterior
cingulate (t
(32)
⫽0.67; p⫽0.051) or the left SMG (t
(32)
⫽0.13; p⫽
0.89), suggesting that these regions were not involved in the process-
ing of numerical magnitude information (although it should be
noted that the parametric effect of ratio approached significance in
the ACC). These data suggest that the brain circuitry engaged by
individuals with lower PSAT scores during single-digit arithmetic is
also engaged during basic quantity processing, while brain mecha-
nisms engaged by individuals with higher PSAT Math scores are not.
These findings bolster the interpretation that the continued reliance
on quantity/procedural mechanisms to solve arithmetic problems is
associated with lower math competence even into high school.
Discussion
Summary and interpretation
The present findings reveal that during single-digit arithmetic
computations, individuals with higher standardized scores on
the PSAT Math test engage calculation brain mechanisms as-
sociated with arithmetic fact retrieval in the left SMG and
bilateral ACC to a greater extent than individuals with rela-
Figure 3. Positive correlation between calculation activation and PSAT Math standard scores in the left SMG. A, Correlation overlaid on an
inflated cortical surface constructed from the average of all participants. B–D, Correlation shown in volume space (radiological convention, left
hemisphereshownonrightandviceversa)insagittal(SAG)(B),coronal (COR) (C), and axial (TRA) (D)orientation,respectively.
Figure 4. Negative correlation between calculation activation and PSAT Math standard scores in the right IPS. A, Correlation overlaid on an
inflated cortical surface constructed from the average of all participants. B–D, Correlation shown in volume space (radiological convention, left
hemisphereshownonrightandviceversa)insagittal(SAG)(B),coronal (COR) (C), and axial (TRA) (D)orientation,respectively.
160 •J. Neurosci., January 2, 2013 •33(1):156 –163 Price et al. •Arithmetic Brain Activation Predicts Math Skills
tively lower PSAT Math scores, who activate quantity-
processing mechanisms in the right IPS.
Each of these regions has been previously associated with
numerical and mathematical processing. Specifically, the left
SMG has been associated with age-related increases in activa-
tion during single-digit arithmetic verification (Rivera et al.,
2005), and several studies have shown that regions of the left-
inferior parietal lobe, including and proximal to the left SMG,
are associated with arithmetic fact retrieval relative to proce-
dural calculations (Delazer et al., 2005;Grabner et al., 2007;
Grabner et al., 2009).
In addition to its activation during arithmetic retrieval,
previous studies have reported involvement of the SMG in the
subjective perception of timing (Wiener et al., 2010a), implicit
timing mechanisms (Wiener et al., 2010b), and phonological
processing during reading (Church et al., 2011). Such findings
could suggest a role for the left SMG in the processing of
rhythmic, phonologically encoded arithmetic facts in memory
(i.e., processing deeply encoded arithmetic facts as a type of
rhyme). However, other studies point to a role for the SMG in
the processing of semantic associations, both in the context of
arithmetic (Grabner et al., 2012) and linguistic processing
(Kim et al., 2011), suggesting a more complex and abstract
function underlying SMG activity. Thus, SMG’s involvement
in arithmetic fact retrieval may represent a more “mature”
calculation mechanism comprising semantic memory search
processes that rely on phonological, timing, and semantic pro-
cessing mechanisms. However, a great deal of future research
is required to fully explicate its precise function.
Likewise, the ACC has previously shown greater activation
during arithmetic retrieval relative to number matching and
during trained versus novel arithmetic problems (Delazer et
al., 2003). This region has a well documented role in conflict-
monitoring, and specifically in the top-down regulation of
cognitive control (Botvinick et al., 2004), suggesting that the
region may play a role in modulating the response to incorrect
equations.
By contrast, activation in the right IPS, found here to correlate
negatively with PSAT Math scores, is frequently observed during
the mental manipulation of numerical quantities in tasks such as
numerical comparison (Dehaene et al., 2003;Cohen Kadosh et
al., 2008). Indeed, in this study, unlike the ACC and left SMG,
activation of the right IPS showed a parametric numerical ratio
effect during nonsymbolic number comparison, suggesting that
high school students with relatively lower mathematical compe-
tence appear to be engaging numerical quantity processing
mechanisms to solve single digit calculations to a greater extent
than their peers with relatively higher PSAT Math scores. It is
possible that these individuals were not relying exclusively on
magnitude processing mechanisms to solve the task, but may
have developed additional alternative strategies not fully eluci-
dated by the current data.
Consistent with the present findings are data from a recent
study using multivoxel pattern analysis in which Cho et al.
(2011) showed that activation patterns in brain regions, in-
cluding the left SMG and right IPS, reliably distinguished be-
tween retrieval versus counting calculation strategies in 7–9
year olds. While those findings reveal a brain network related
to arithmetic strategy use, the present data are the first to
demonstrate that individual differences in the relative engage-
ment of the nodes of that network are associated with perfor-
mance on a high school mathematical competence test. Thus,
we suggest that successful encoding of arithmetic facts con-
tributes, in combination with other factors not investigated in
the present study, to the successful acquisition of higher level
mathematical competence affecting the ontogenetic construc-
tion of brain networks facilitating the learning of higher level
mathematical skills.
Developmental scaffolding
The interpretation of the present results is supported by a large
body of behavioral research showing that children typically
undergo a process of development in arithmetic skill whereby
simple calculations are initially computed through procedural
strategies, but then gradually come to be solved by memory
retrieval (Ashcraft, 1982;Geary et al., 1991). Children with
mathematical learning difficulties fail to show this develop-
mental shift (Geary, 1993), suggesting that arithmetic fluency
plays a key role in the acquisition of higher math skills. The
present data support such a link, providing the first neurosci-
entific evidence that the functional brain networks associated
with arithmetic fluency are related to higher-level math skills.
In contrast, behavioral performance measures did not reveal
specific associations, thereby highlighting the value added by
neuroimaging to our understanding of the cognitive founda-
tions of mathematical competence.
The association between activation of the IPS and poorer
math competence may seem counterintuitive, as previous be-
havioral evidence suggests that numerical magnitude process-
ing serves as a foundation for the acquisition of early
arithmetic skills (Halberda et al., 2008). Furthermore, neuro-
imaging data have revealed that in children with mathematical
learning difficulties, the right IPS region thought to support
numerical magnitude processing shows atypical responses
during numerical magnitude processing (Price et al., 2007;
Mussolin et al., 2010). Thus, the functional maturity of the
neural substrates for numerical magnitude processing appears
to serve as a foundation for early arithmetic learning. The
present results, however, in concert with previous findings
(De Smedt et al., 2011), demonstrate that while such quantity
processing mechanisms may have an important role in the
development of elementary arithmetic skills, individuals who
continue to rely upon them into adolescence and beyond
achieve poorer mathematical competence than their peers
who do not. Migration away from quantity-based calculation
strategies appears essential for the development of mathemat-
ical competence beyond simple arithmetic.
Alternative interpretations
It should be noted that the IPS is also known to be involved in
visuospatial working memory, which in turn plays a role in arithme-
tic performance (Dumontheil and Klingberg, 2012), so the present
results could reflect working memory rather than numerical magni-
tude processing mechanisms. However, arithmetic problem solving
involves the mental manipulation of quantities for which both work-
ing memory and the engagement of quantity representations are
required. Furthermore, the same IPS region found to correlate neg-
atively with PSAT Math scores showed a parametric ratio effect dur-
ing a nonsymbolic number comparison task that carried no working
memory demands. Thus, it is unlikely that working memory can be
the sole factor in explaining the present results.
A further constraint on the interpretation of the current data
is that they are correlational, and thus it is impossible to draw
resolute causal inferences. Since single-digit arithmetic is learned
during the very first years of schooling and the PSAT is taken in
the last years of high school, it seems logical that single-digit
Price et al. •Arithmetic Brain Activation Predicts Math Skills J. Neurosci., January 2, 2013 •33(1):156 –163 • 161
arithmetic skills and their associated neural mechanisms would
exert an influence on the acquisition of high school level math
skills as opposed to the reverse. However, the present data
cannot rule out the possibility that those individuals who
scored more highly on the PSAT Math spent more time en-
gaged in practice activities involving mental calculation and
thus developed more fluent mental arithmetic processing,
which was reflected in the brain activation patterns reported
above.
Conclusions and applications
A better understanding of the sources of variability in mathe-
matical skills may inform educational approaches to improv-
ing math achievement. Although the present data do not allow
us to speculate as to what pedagogical methods are best suited
to facilitate the successful encoding of arithmetic facts into
memory, they do have important educational implications. In
2005 the Fordham Foundation issued a critique on state math-
ematics standards (Klein et al., 2005) and reported that even
the highest ranked American state curricula spend signifi-
cantly less time on arithmetic than the “A⫹countries” (Sin-
gapore, Japan, Korea, Hong Kong, Flemish Belgium, and the
Czech Republic). From an educational perspective, our results
provide the first neuroscientific evidence demonstrating the
fundamental importance of fluency in basic mental arithmetic
in the acquisition of college-level mathematical skills. Fur-
thermore, they significantly extend our understanding of the
relationship between simple arithmetic and higher level math
competence beyond that revealed by behavioral data alone.
Specifically, the relationship between PSAT Math and func-
tional brain activation during single-digit arithmetic was sig-
nificant even when controlling for PSAT Critical Reading,
revealing neurocognitive mechanisms specific to PSAT Math
not evident from reaction time analysis alone.
In conclusion, the present data are the first to demonstrate
that brain mechanisms associated with elementary arithmetic
skills are related to performance on a broad ranging, education-
ally relevant measure of math competence at the end of high
school. Thus, the importance of early arithmetic skills for math
competence is not only evident at a behavioral level. Their
acquisition appears to impact the construction of neurobio-
logical architectures across development, which may in turn
support the acquisition of high school-level math skills that
have significant consequences for progression into higher ed-
ucation. Finally, the present findings demonstrate how neuro-
imaging data can inform our understanding of educationally
relevant issues and thus demonstrate the power of an educa-
tional neuroscience framework.
References
Ashcraft MH (1982) The development of mental arithmetic: a chronomet-
ric approach. Dev Rev 2:213–236. CrossRef
Botvinick MM, Cohen JD, Carter CS (2004) Conflict monitoring and ante-
rior cingulate cortex: an update. Trends Cogn Sci 8:539–546. CrossRef
Medline
Carew TJ, Magsamen SH (2010) Neuroscience and education: an ideal part-
nership for producing evidence-based solutions to guide 21st century
learning. Neuron 67:685– 688. CrossRef Medline
Cho S, Ryali S, Geary DC, Menon V (2011) How does a child solve 7⫹8?
Decoding brain activity patterns associated with counting and retrieval
strategies. Dev Sci 14:989–1001. CrossRef Medline
Church JA, Balota DA, Petersen SE, Schlaggar BL (2011) Manipulation of
length and lexicality localizes the functional neuroanatomy of phonolog-
ical processing in adult readers. J Cogn Neurosci 23:1475–1493. CrossRef
Medline
Cohen Kadosh R, Lammertyn J, Izard V (2008) Are numbers special? An
overview of chronometric, neuroimaging, developmental and compara-
tive studies of magnitude representation. Prog Neurobiol 84:132–147.
CrossRef Medline
CollegeBoard (2008) PSAT/NMSQT percentiles and mean scores. In: Un-
derstanding 2008 PSAT/NMSQT scores. New York: CollegeBoard.
Dehaene S, Piazza M, Pinel P, Cohen L (2003) Three Parietal Circuits for
Number Processing. Cogn Neuropsychol 20:487–506. CrossRef Medline
Delazer M, Domahs F, Bartha L, Brenneis C, Lochy A, Trieb T, Benke T
(2003) Learning complex arithmetic–an fMRI study. Brain Res Cogn
Brain Res 18:76 –88. CrossRef Medline
Delazer M, Ischebeck A, Domahs F, Zamarian L, Koppelstaetter F, Siedentopf
CM, Kaufmann L, Benke T, Felber S (2005) Learning by strategies and
learning by drill– evidence from an fMRI study. Neuroimage 25:838– 849.
CrossRef Medline
De Smedt B, Holloway ID, Ansari D (2011) Effects of problem size and
arithmetic operation on brain activation during calculation in children
with varying levels of arithmetical fluency. Neuroimage 57:771–781.
CrossRef Medline
Dumontheil I, Klingberg T (2012) Brain activity during a visuospatial work-
ing memory task predicts arithmetical performance 2 years later. Cereb
Cortex 22:1078–1085. CrossRef Medline
Duncan GJ, Dowsett CJ, Claessens A, Magnuson K, Huston AC, Klebanov P,
Pagani LS, Feinstein L, Engel M, Brooks-Gunn J, Sexton H, Duckworth K,
Japel C (2007) School readiness and later achievement. Dev Psychol 43:
1428–1446. CrossRef Medline
Forman SD, Cohen JD, Fitzgerald M, Eddy WF, Mintun MA, Noll DC (1995)
Improved assessment of significant activation in functional magnetic res-
onance imaging (fMRI): use of a cluster-size threshold. Magn Reson Med
33:636–647. CrossRef Medline
Friston KJ, Fletcher P, Josephs O, Holmes A, Rugg MD, Turner R (1998)
Event-related fMRI: characterizing differential responses. Neuroimage
7:30–40. CrossRef Medline
Geary DC (1993) Mathematical disabilities: cognitive, neuropsychological,
and genetic components. Psychol Bull 114:345–362. CrossRef Medline
Geary DC, Brown SC, Samaranayake VA (1991) Cognitive addition: a
short longitudinal study of strategy choice and speed-of-processing
differences in normal and mathematically disabled children. Dev Psy-
chol 27:787–797. CrossRef
Goebel R, Esposito F, Formisano E (2006) Analysis of functional image
analysis contest (FIAC) data with brainvoyager QX: From single-subject
to cortically aligned group general linear model analysis and self-
organizing group independent component analysis. Hum Brain Mapp
27:392–401. CrossRef Medline
Grabner RH, Ansari D, Reishofer G, Stern E, Ebner F, Neuper C (2007)
Individual differences in mathematical competence predict parietal brain
activation during mental calculation. Neuroimage 38:346–356. CrossRef
Medline
Grabner RH, Ansari D, Koschutnig K, Reishofer G, Ebner F, Neuper C
(2009) To retrieve or to calculate? Left angular gyrus mediates the re-
trieval of arithmetic facts during problem solving. Neuropsychologia 47:
604–608. CrossRef Medline
Grabner RH, Ansari D, Koschutnig K, Reishofer G, Ebner F (2012) The
function of the left angular gyrus in mental arithmetic: evidence from the
associative confusion effect. Hum Brain Mapp. Advanced online publica-
tion. Retrieved July 17, 2012. CrossRef Medline
Halberda J, Mazzocco MM, Feigenson L (2008) Individual differences in
non-verbal number acuity correlate with maths achievement. Nature 455:
665–668. CrossRef Medline
Holloway ID, Price GR, Ansari D (2010) Common and segregated neural
pathways for the processing of symbolic and nonsymbolic numerical
magnitude: An fMRI study. Neuroimage 49:1006–1017. CrossRef
Medline
Kim KK, Karunanayaka P, Privitera MD, Holland SK, Szaflarski JP (2011)
Semantic association investigated with functional MRI and independent
component analysis. Epilepsy Behav 20:613–622. CrossRef Medline
Klein D, Braams BJ, Parker T, Quirk W, Wilson WS (2005) The state of
math standards. Washington, DC: Fordham Foundation.
Mazzocco MM, Myers GF (2003) Complexities in identifying and defining
mathematics learning disability in the primary school-age years. Ann
Dyslexia 53:218 –253. CrossRef Medline
Mazzocco MM, Devlin KT, McKenney SJ (2008) Is it a fact? Timed arith-
162 •J. Neurosci., January 2, 2013 •33(1):156 –163 Price et al. •Arithmetic Brain Activation Predicts Math Skills
metic performance of children with mathematical learning disabilities
(MLD) varies as a function of how MLD is defined. Dev Neuropsychol
33:318–344. CrossRef Medline
Mazzocco MM, Feigenson L, Halberda J (2011) Preschoolers’ precision of
the approximate number system predicts later school mathematics per-
formance. PLoS One 6:e23749. CrossRef Medline
Menon V, Rivera SM, White CD, Glover GH, Reiss AL (2000) Dissociating
prefrontal and parietal cortex activation during arithmetic processing.
Neuroimage 12:357–365. CrossRef Medline
Moyer RS, Landauer TK (1967) Time required for judgements of numerical
inequality. Nature 215:1519–1520. CrossRef Medline
Mussolin C, De Volder A, Grandin C, Schlo¨gel X, Nassogne MC, Noe¨l MP
(2010) Neural correlates of symbolic number comparison in develop-
mental dyscalculia. J Cogn Neurosci 22:860–874. CrossRef Medline
National Academies (2007) Rising above the gathering storm: energizing
and employing America for a brighter economic future (Committee on
Prospering in the Global Economy of the 21st Century and Committee on
Science, Engineering and Public Policy, ed). Washington, DC: National
Academy Press.
Organisation for Economic Co-operation and Development (2010) The
high cost of low education performance: the long-run economic impact
of improving PISA outcomes. Paris: OECD.
Parsons S, Bynner J (2005) Does numeracy matter more? London: National
Research and Development Centre for Adult Literacy and Numeracy.
Pinel P, Dehaene S, Rivie`re D, LeBihan D (2001) Modulation of parietal
activation by semantic distance in a number comparison task. Neuroim-
age 14:1013–1026. CrossRef Medline
Price GR, Holloway I, Ra¨sa¨nen P, Vesterinen M, Ansari D (2007) Impaired
parietal magnitude processing in developmental dyscalculia. Curr Biol
17:R1042–R1043. CrossRef
Rivera SM, Reiss AL, Eckert MA, Menon V (2005) Developmental changes
in mental arithmetic: evidence for increased functional specialization in
the left inferior parietal cortex. Cereb Cortex 15:1779–1790. CrossRef
Medline
Rueckert L, Lange N, Partiot A, Appollonio I, Litvan I, Le Bihan D, Grafman J
(1996) Visualizing cortical activation during mental calculation with
functional MRI. Neuroimage 3:97–103. CrossRef Medline
Talairach J, Tournoux P (1988) Co-planar atlas of the human brain. New
York: Thieme.
Updegraff KA, Eccles JS, Barber BA, O’Brien KM (1996) Course enrollment
As self-regulatory behaviour: who takes optional high school math
courses? Learn Indiv Diff 8:239–259. CrossRef
Wiener M, Hamilton R, Turkeltaub P, Matell MS, Coslett HB (2010a)
Fast forward: supramarginal gyrus stimulation alters time measure-
ment. J Cogn Neurosci 22:23–31. CrossRef Medline
Wiener M, Turkeltaub PE, Coslett HB (2010b) Implicit timing activates the
left inferior parietal cortex. Neuropsychologia 48:3967–3971. CrossRef
Medline
Price et al. •Arithmetic Brain Activation Predicts Math Skills J. Neurosci., January 2, 2013 •33(1):156 –163 • 163
