Journal of Experimental Child Psychology 152 (2016) 173–191

Contents lists available at ScienceDirect

Journal of Experimental Child

Psychology
journal homepage: www.elsevier.com/locate/jecp

Cognitive factors affecting children’s nonsymbolic

and symbolic magnitude judgment abilities: A
latent profile analysis
Cindy S. Chew ⇑, Jason D. Forte, Robert A. Reeve ⇑
Melbourne School of Psychological Sciences, University of Melbourne, Parkville, Victoria 3010, Australia

a r t i c l e i n f o a b s t r a c t

Article history: Early math abilities are claimed to be linked to magnitude repre-
Received 4 May 2016 sentation ability. Some claim that nonsymbolic magnitude abilities
Revised 1 July 2016 scaffold the acquisition of symbolic (Arabic number) magnitude
abilities and influence math ability. Others claim that symbolic
magnitude abilities, and ipso facto math abilities, are independent
Keywords:
of nonsymbolic abilities and instead depend on the ability to pro-
Nonsymbolic and symbolic ability patterns
Visuospatial working memory
cess number symbols (e.g., 2, 7). Currently, the issue of whether
Number naming ability symbolic abilities are or are not related to nonsymbolic abilities,
Math ability differences and the cognitive factors associated with nonsymbolic–symbolic
Latent profile analysis relationships, remains unresolved. We suggest that different non-
Magnitude representation symbolic–symbolic relationships reside within the general magni-
tude ability distribution and that different cognitive abilities are
likely associated with these different relationships. We further sug-
gest that the different nonsymbolic–symbolic relationships and
cognitive abilities in combination differentially predict math abili-
ties. To test these claims, we used latent profile analysis to identify
nonsymbolic–symbolic judgment patterns of 124, 5- to 7-year-
olds. We also assessed four cognitive factors (visuospatial working
memory [VSWM], naming numbers, nonverbal IQ, and basic reac-
tion time [RT]) and two math abilities (number transcoding and
single-digit addition abilities). Four nonsymbolic–symbolic ability
profiles were identified. Naming numbers, VSWM, and basic RT
abilities were differentially associated with the different ability
profiles and in combination differentially predicted math abilities.
Findings show that different patterns of nonsymbolic–symbolic
magnitude abilities can be identified and suggest that an adequate
account of math development should specify the inter-relationship

⇑ Corresponding authors.
E-mail addresses: cindy.chew@unimelb.edu.au (C.S. Chew), r.reeve@unimelb.edu.au (R.A. Reeve).

http://dx.doi.org/10.1016/j.jecp.2016.07.001
0022-0965/Ó 2016 Elsevier Inc. All rights reserved.
174 C.S. Chew et al. / Journal of Experimental Child Psychology 152 (2016) 173–191

between cognitive factors and nonsymbolic–symbolic ability

patterns.
Ó 2016 Elsevier Inc. All rights reserved.

Introduction

It is claimed that magnitude representation abilities support emerging math abilities. This claim is
predicated in part on the proposition that early developing nonsymbolic magnitude abilities support
the acquisition of later acquired symbolic magnitude abilities (Dehaene, 2007; Piazza, 2010). Nonsym-
bolic and symbolic magnitude representations are typically assessed by the ability to compare quan-
tities (e.g.,  vs. ) and Arabic digits (e.g., 3 vs. 5) that differ in magnitude, respectively. Similar
error and reaction time (RT) response signatures for nonsymbolic and symbolic magnitude judgments
support the claim that they share a common representation system (Gebuis, Cohen Kadosh, de Haan, &
Henik, 2009; Halberda, Mazzocco, & Feigenson, 2008; Moyer & Landauer, 1967). However, nonsym-
bolic and symbolic magnitude abilities are not always related to each other, nor is nonsymbolic ability
always related to math abilities (Holloway & Ansari, 2009; Sasanguie, De Smedt, Defever, & Reynvoet,
2012; Vanbinst, Ghesquière, & De Smedt, 2012). Indeed, some claim that symbolic magnitude ability is
independent of nonsymbolic ability and that symbolic judgments reflect an ability to connect number
symbols to the magnitude information (De Smedt, Verschaffel, & Ghesquière, 2009; Rousselle & Noël,
2007). It is evident that the circumstances under which symbolic magnitude abilities are or are not
related to nonsymbolic abilities are currently underspecified. We suggest that one way to resolve this
impasse would be to determine whether different patterns of nonsymbolic–symbolic magnitude abil-
ities reside within a general nonsymbolic–symbolic magnitude ability distribution and to identify the
cognitive markers linked to these different patterns to better understand individual differences in chil-
dren’s math abilities.

Nonsymbolic–symbolic magnitude abilities and math abilities

Some researchers have found an association between nonsymbolic and symbolic magnitude abil-
ities and math abilities in typically developing children (Bonny & Lourenco, 2013; Gilmore, Attridge,
De Smedt, & Inglis, 2014; Mussolin, Nys, Leybaert, & Content, 2012) as well as in children with math
learning difficulties (Landerl & Kölle, 2009; Mazzocco, Feigenson, & Halberda, 2011a; Mussolin, Mejias,
& Noël, 2010; Piazza et al., 2010). This pattern of findings is interpreted as showing that nonsymbolic
abilities affect symbolic abilities, which in turn affect math abilities.
In contrast, others have found only an association between symbolic magnitude abilities and math
abilities (De Smedt & Gilmore, 2011; De Smedt et al., 2009; Holloway & Ansari, 2009; Iuculano, Tang,
Hall, & Butterworth, 2008; Landerl & Kölle, 2009). The lack of a relationship between nonsymbolic and
symbolic magnitude abilities has led some to suggest that they are independent abilities (Le Corre &
Carey, 2007; Lyons, Ansari, & Beilock, 2012; Noël & Rousselle, 2011) and exert independent effects on
math abilities (Fazio, Bailey, Thompson, & Siegler, 2014). Still others have suggested that nonsymbolic
magnitude abilities play a noncrucial role in math development (De Smedt, Noël, Gilmore, & Ansari,
2013). Children may possess adequate nonsymbolic magnitude abilities but poor symbolic abilities.
This position is often characterized as a symbolic access deficit, which argues that children’s symbolic
magnitude judgment and math difficulties stem from difficulties connecting number symbols with
their corresponding nonsymbolic quantities (Rousselle & Noël, 2007).

Can different nonsymbolic–symbolic ability relationship findings be reconciled?

Although findings suggest that nonsymbolic–symbolic magnitude relationships may be affected by

factors such as age (Inglis, Attridge, Batchelor, & Gilmore, 2011; Rousselle & Noël, 2008), magnitude
 C.S. Chew et al. / Journal of Experimental Child Psychology 152 (2016) 173–191 175

judgment measures (Gilmore et al., 2014; Price, Palmer, Battista, & Ansari, 2012), and sample size,
these factors fail to account for all differences in findings (see Chen & Li, 2014; Fazio et al., 2014,
and Schneider et al., 2016, for discussions). We suggest that differences in the pattern of
nonsymbolic–symbolic magnitude abilities may reflect mixtures of different nonsymbolic–symbolic
relationships residing within a general nonsymbolic–symbolic magnitude ability distribution and
may be differentially affected by these different factors.

Individual differences in nonsymbolic–symbolic magnitude abilities

Several researchers have emphasized the importance of focusing on individual differences in math
development (Ansari, 2010; Dowker, 2008; Gray & Reeve, 2016; Jordan, Mulhern & Wylie, 2009). Most
researchers use variable-centered methods in their analysis of nonsymbolic–symbolic magnitude rela-
tionships and employ correlation and regression techniques. These approaches focus on identifying
similar processes on the assumption that ‘‘populations” are homogeneous with respect to how vari-
ables of interest influence each other (Magnusson, 2003). However, correlations between nonsym-
bolic–symbolic magnitude abilities and math abilities are often modest (Chen & Li, 2014; Fazio
et al., 2014), suggesting significant between- and within-age variability. For instance, a recent
meta-analysis found that the correlation between math competence and nonsymbolic magnitude abil-
ities was r = .24 and the correlation between math competence and symbolic magnitude abilities was
r = .30 (Schneider et al., 2016). Although variable-centered approaches yield interesting findings about
the associations among variables, they do not account for the unexplained variance (91% in the case of
a .30 correlation).
Aggregated data methods raise three issues. First, if a group of children shows different patterns of
nonsymbolic–symbolic magnitude abilities, the average pattern comprises a summary of a mixture of
patterns (Paul & Reeve, 2016; Siegler, 1987). Second, averaged or median scores ignore the relevance
of different ability patterns. Individual deviations from the mean or median score are often assumed as
measurement error when, in some circumstances they may contain important information about indi-
viduals’ abilities (Weinert & Helmke, 1998). Whether children are similarly relatively high or low on
nonsymbolic–symbolic magnitude abilities is not distinguished in these analytical models because
they occupy the same ‘‘distance to the mean.” An examination of nonsymbolic–symbolic magnitude
relative strengths/weaknesses is attenuated by an absence of an appropriate measurement structure
taking into account whether or not children’s abilities lie on different nonsymbolic–symbolic magni-
tude ability dimensions. Third, significant within-age variability in children’s abilities is typically trea-
ted as error when it may be related to inter-individual development (Dowker, 2008; Gray & Reeve,
2016; Wohlwill, 1973). Several studies have shown that assuming age-based differences masks differ-
ent patterns in data (Bouwmeester & Verkoeijen, 2012; Paul & Reeve, 2016; Reeve, Reynolds,
Humberstone, & Butterworth, 2012).

Cognitive factors affecting nonsymbolic–symbolic magnitude relationships

Magnitude information is claimed to be encoded spatially (see de Hevia, Vallar, & Girelli, 2008;
Dehaene, Piazza, Pinel, & Cohen, 2003). Indeed, visuospatial working memory (VSWM) is thought to
support the representation and manipulation of numerical magnitudes (Dehaene, 1992; Dehaene &
Cohen, 1997; Zorzi, Priftis, & Umiltà, 2002). Moreover, VSWM is implicated in early math ability
(Bull, Espy, & Wiebe, 2008; De Smedt et al., 2009; Holmes & Adams, 2006; Mammarella, Lucangeli,
& Cornoldi, 2010). Paul and Reeve (2016), for example, found a link between VSWM and single-
digit addition problem-solving strategy sophistication.
Differences in processing speed have also been linked to numerical processing and math abilities
(Bull & Johnston, 1997; Geary, 2011; Hecht, Torgesen, Wagner, & Rashotte, 2001), which may affect
nonsymbolic–symbolic magnitude ability patterns. Basic RT ability may affect nonsymbolic–symbolic
magnitude abilities in two ways. First, nonsymbolic and symbolic magnitude tasks are often analyzed
in terms of reaction times. Second, the speed and accuracy in naming Arabic digits is often used to
176 C.S. Chew et al. / Journal of Experimental Child Psychology 152 (2016) 173–191

assess the ability to access number symbols information (i.e., symbolic number knowledge) (Berteletti,
Lucangeli, Piazza, Dehaene, & Zorzi, 2010; De Smedt & Gilmore, 2011; Vanbinst, Ghesquière, & De
Smedt, 2015). Indeed, the ability to access symbolic numerical information is invoked as an explana-
tion for differences in symbolic magnitude abilities and ipso facto math abilities (Rousselle & Noël,
2007).

The current study

The aim of this study was threefold. The first aim was to determine whether different profiles of
nonsymbolic–symbolic magnitude ability patterns can be identified from the overall ability distribu-
tion. The heterogeneity with respect to nonsymbolic–symbolic magnitude abilities may reflect differ-
ent mixtures of nonsymbolic–symbolic relationships. Latent profile analysis (LPA) is a statistical
model-based approach to dealing with heterogeneity by identifying a small group of homogeneous
latent subgroups embedded within a set of measures (Lanza & Cooper, 2016). Individuals are assigned
to the subgroup for which the posterior probability of belonging to it is the highest, calculated as a
function of the observed data and parameter estimates (Lanza & Cooper, 2016; Vermunt &
Magidson, 2013). The conditional probabilities provide the measurement structure that defines the
latent subgroups (e.g., relative good/poor nonsymbolic–symbolic magnitude abilities) and allows an
examination of nonsymbolic–symbolic magnitude abilities. Age can be examined, a posteriori, in rela-
tion to subgroups rather than assuming age-based differences (Gray & Reeve, 2014; Paul & Reeve,
2016).
We employed LPA to identify and describe subgroups from a large group of 5- to 7-year-olds’ non-
symbolic–symbolic magnitude accuracy and RT response signatures. We suggest that LPA is useful in
analyzing performance differences in this age range because there are known large individual differ-
ences in math-related abilities in children of this age (Dowker, 2008; Gray & Reeve, 2016; Rittle-
Johnson, Fyfe, Loehr, & Miller, 2015), cautioning against the use of age-based analysis. We anticipated
that children will exhibit poor and/or good nonsymbolic and symbolic magnitude abilities (i.e., similar
nonsymbolic–symbolic magnitude abilities) and exhibit a pattern of better nonsymbolic magnitude
ability relative to poorer symbolic ability.
A second aim of the study was to identify the cognitive factors associated with different nonsym-
bolic–symbolic magnitude ability profiles. Because the ability to access symbolic numerical informa-
tion is argued to affect symbolic judgment performance (Rousselle & Noël, 2007), we expected a
profile characterized by relatively poorer symbolic magnitude abilities to have poorer naming number
ability. On the other hand, VSWM capacity is thought to support processing of magnitudes, and hence
we would expect a profile characterized by relatively poorer nonsymbolic magnitude abilities to have
poorer VSWM ability and vice versa (i.e., profile with relatively good nonsymbolic abilities to have bet-
ter VSWM ability). A measure of young children’s nonverbal IQ was included to account for potential
effects of intellectual ability for nonsymbolic–symbolic magnitude abilities and math abilities (Luwel,
Foustana, Onghena, & Verschaffel, 2013). We also assessed basic RT as a measure of processing speed
(Kail & Ferrer, 2007) because it might account for differences in nonsymbolic–symbolic magnitude
judgments.
Third, we expected math abilities to be associated with nonsymbolic–symbolic magnitude profile
membership. Magnitude representation abilities are claimed to be important for early emerging math
abilities (e.g., number fact knowledge but not more complex abilities: Fuchs et al., 2010; Geary, Hoard,
& Bailey, 2012). We focused on single-digit addition (SDA) and transcoding (‘‘reading” number strings)
because both are considered to be important early abilities for later math abilities (Deloche & Seron,
1987; Geary, 2000; Organization for Economic Cooperation and Development [OECD], 2012; Vanbinst,
Ceulemans, Ghesquière, & De Smedt, 2015). We anticipated that profiles representing relatively
poorer symbolic and/or poorer nonsymbolic magnitude abilities will perform poorer on SDA and
transcoding.
 C.S. Chew et al. / Journal of Experimental Child Psychology 152 (2016) 173–191 177

Method

Participants

Participants were 124 children comprising 43 in kindergarten (Mage = 5.8 years, SD = 4.6 months),
39 in Year 1 (Mage = 7.1 years, SD = 3.58 months), and 42 in Year 2 (Mage = 7.8 years, SD = 3.6 months).
All children spoke English, had normal or corrected-to-normal vision, and had no identified learning
difficulties. The study was conducted with the approval of, and in accordance with, the university’s
human research ethics committee.

Procedure

All children individually completed nonsymbolic and symbolic judgments, naming numbers,
single-digit addition, reading numbers, Corsi blocks (VSWM), Raven’s Coloured Progressive Matrices
(RCPM, nonverbal IQ), and basic RT tasks. Tasks were completed in three sessions (nonsymbolic and
symbolic tasks were completed on separate days to avoid inter-task priming effects). With the excep-
tion of the IQ and VSWM tasks, stimuli were presented on a 15-inch screen laptop computer running
E-Prime software (Version 2.0). The screen was at eye level approximately 40 cm in front of children. A
fixation point (a cross) appeared in the center of the screen prior to a target stimulus appearing. A
stimulus remained on the screen until a response was made except for nonsymbolic and symbolic
judgment tasks, in which the stimuli disappeared after 5000 ms.

Nonsymbolic and symbolic judgment tasks

In the nonsymbolic judgment task, two sets of blue squares separated by a central vertical line
appeared on the screen. Children selected the set that had the most squares by pressing the corre-
sponding right shift key or left shift key (see Gilmore, Attridge, & Inglis, 2011; Holloway & Ansari,
2009). The task comprised 72 trials with judgment combinations of all quantities between 1 and 9
blue squares except ties (e.g., 9 and 9). The ratios for each trial (i.e., smaller numbers/larger numbers)
were divided into eight ratios: 0.1–0.19, 0.2–0.29, . . ., 0.8–0.89 (Reeve et al., 2012). Stimuli were pre-
sented in a fixed randomized order, with the larger set appearing on the left- and right-hand sides of
the screen equally. To reduce possible reliance on perceptual cues for judgments, individual square
sizes and total area were systematically varied across trials (total area was the same for both sets
within trials) (Dehaene, Izard, & Piazza, 2005). Following recommendations (Bartelet, Ansari,
Vaessen, & Blomert, 2014; Ratcliff, Thompson, & McKoon, 2015; Schneider et al., 2016), we analyzed
two indexes (mean accuracy and median RT), both of which are associated with math ability (De
Smedt et al., 2013). Because there was no significant interaction between ratio and grade on RT, ratios
were represented by a single RT measure.
The symbolic judgment task was procedurally identical to the nonsymbolic task except that black
Arabic digits were presented on a white background (60-point font size).

Digit naming

Children named digits between 1 and 9, with each digit being presented three times in separate
blocks of trials (n = 27 trials overall). The interviewer pressed a response key following each response
and recorded responses verbatim (the interviewer could not see the computer screen). Median RT was
used because children made few errors on this task.

Single-digit addition

Children completed 30 two-term addition problems following 2 practice trials. They were
instructed to answer problems as quickly and accurately as possible. Addends comprised combina-
tions of all digits between 2 and 7 (excluding tied pairs, e.g., 2 + 2) in both orders (e.g., 2 + 7 and 7
178 C.S. Chew et al. / Journal of Experimental Child Psychology 152 (2016) 173–191

+ 2). Single-digit addition problems are widely used as a measure of early computation ability (Bailey,
Littlefield, & Geary, 2012; Paul & Reeve, 2016).

Transcoding: Reading multi-digit numbers

Children read 30 two- to four-digit numbers displayed on the screen (one number per trial; see
online Supplementary material for the stimuli).

Corsi blocks (VSWM)

This task was administered following Kessels, van Zandvoort, Postma, Kappelle, and de Haan
(2000). The interviewer tapped a sequence of blocks in a prespecified order, and children attempted
to repeat the tap sequences. The sequence began with two blocks and increased by one block follow-
ing a correct tap reproduction to a maximum of nine blocks. Testing ceased after 2 failed trials. The
VSWM span comprised the average of the longest correct block tap sequences.

Raven’s Coloured Progressive Matrices (nonverbal IQ)

RCPM was administered following manual instructions and scored using age norms (Cotton et al.,
2005; Raven, Court, & Raven, 1995).

Basic RT

Children pressed a computer key as quickly as possible when a black dot appeared on the screen
approximately 500 ms after a central fixation point. The task comprised 9 trials.

Analytic approach

We used Latent GOLD 5.0’s latent profile analysis (Vermunt & Magidson, 2013) to identify sub-
groups of children who share similar profiles of nonsymbolic–symbolic magnitude accuracy and RT
response patterns. We used 200 random starting sets and 500 replications to ensure that model con-
vergence could be replicated, and we estimated candidate models for one to six profiles. Models are
compared using a combination of statistical criteria, parsimony, and theoretical interpretability
(Collins & Lanza, 2013). Goodness-of-fit statistics, such as Bayesian information criterion (BIC), weigh
the fit of the models relative to the number of parameters, with a lower value indicating a better fitting
model to the data (Vermunt & Magidson, 2013). Models with fewer parameters are considered to be
more parsimonious and have a greater likelihood of generalizability. An R2 entropy value quantified
the degree of separation between subgroups (i.e., less likely to misclassify with greater separation),
with values greater than .80 preferred (range of 0–1) (Clark & Muthen, 2009; Lanza, Collins,
Lemmon, & Schafer, 2007). The proportion of classification errors was used to assess the accuracy
of this allocation procedure (Magidson & Vermunt, 2005).
Latent GOLD’s Step3 module (Vermunt, 2010; Vermunt & Magidson, 2013) was used to examine
the association among age, cognitive factors, and nonsymbolic–symbolic magnitude profile member-
ship and the association between profile membership and math abilities. The underlying statistical
model is analogous to a multinomial regression logistic regression; outcomes are interpreted as the
association between each cognitive factor (treated here as covariates) and profile membership while
accounting for other factors. Similarly, SDA and transcoding abilities (treated here as dependent vari-
ables) were regressed on the latent profile membership. The Step3 modeling approach allows for the
correction of classification errors obtained when assigning profile memberships (maximum likelihood
adjustment method is used to correct for classification errors); a failure to account for classification
errors can lead to an underestimation of the relationship between profile membership and other
variables (Bakk, Tekle, & Vermunt, 2013).
 C.S. Chew et al. / Journal of Experimental Child Psychology 152 (2016) 173–191 179

Results

Consistent with previous research, nonsymbolic and symbolic magnitude RT and error rates
increased with increasing difficulty (i.e., increasing ratios) and decreased with increasing grade
(Lyons, Price, Vaessen, Blomert, & Ansari, 2014; Mussolin et al., 2012). Detailed information is reported
in the Supplementary material.
Correlations between all measures are reported in Table 1, which shows significant correlations
between children’s nonsymbolic–symbolic magnitude measures and the math performance measures
(SDA correct and transcoding ability) and between nonsymbolic and symbolic judgments on accuracy
and RT. Whereas accuracy and RT were correlated within the symbolic task, this was not the case for
the nonsymbolic task. Modest correlations were found among the cognitive measures. Age was asso-
ciated with all measures except nonverbal IQ. The modest and varied correlations suggest significant
variability and support the need to examine the possibility of different patterns of nonsymbolic–
symbolic magnitude abilities.

Identifying nonsymbolic–symbolic magnitude subgroups

LPA models comprising one to six profiles were estimated from nonsymbolic–symbolic magnitude
RT and accuracy. Goodness-of-fit statistics were best for a four-profile solution (see Table 2 for model
comparisons). Model selection was supported by a high entropy value—indicating good classification
of individuals into profiles and low classification errors. All indicators significantly contributed to clas-
sifying RT and accuracy responses into subgroups (R2: nonsymbolic accuracy = .40, symbolic accu-
racy = .59, nonsymbolic RT = .42, symbolic RT = .49; all ps < .001). Fit statistics were identical in
replication analysis, indicating that the solution was robust and did not converge on a local maximum
(i.e., best solution in a localized parameter space but not the global maximum).
Deviations from overall mean proportion accuracy (nonsymbolic = .92, symbolic = .90) and median
RT (nonsymbolic = 1009.88 ms, symbolic = 1400.88 ms) for the four profiles are presented in Fig. 1
(top panel). The utility of LPA is further highlighted by comparing the profiles with grade-based anal-
ysis presented in Fig. 1 (bottom panel). Averaging scores per age group tend to attenuate inter-
individual differences and suppress the power to detect different response patterns in the data. When
examining individual variables, the grade-based graph showed a general linear trend of increasing
nonsymbolic–symbolic accuracy and speed with age, whereas the profile-based graph showed distinct
differences of nonsymbolic–symbolic accuracy and speed.
We examined the ordinal-by-ordinal gamma statistic to determine whether nonsymbolic–
symbolic magnitude profiles and grade showed an ordered association. (Profiles were ordered accord-
ing to sophistication. Profile 2 was judged to be the most sophisticated, followed by Profile 4, Profile 1,

Table 1
Zero-order correlations among NSM–SM measures, cognitive factors, math ability, and age.

Measure 1 2 3 4 5 6 7 8 9 10 11
1. NSM correct 1
2. NSM RT .14 1
3. SM correct .40** .07 1
4. SM RT .18* .54** .18* 1
5. VSWM .01 .26** .15 .30** 1
6. NN .34** .31** .33** .45** .29** 1
7. Basic RT .41** .28** .34** .34** .23* .31** 1
8. RCPM .07 .17 .06 .20* .34** .15 .09 1
9.Transcoding .39** .27** .34** .53** .29** .57** .36** .18* 1
10. SDA correct .47** .24** .37** .37** .22* .49** .37** .14 .74** .1
11. Age (in months) .3** .45** .31** .57** .21* .43** .34** .00 .77** .62** 1

Note. NSM, nonsymbolic magnitude; SM, symbolic magnitude; NN, naming numbers RT.
*
p < .05.
**
p < .01.
180 C.S. Chew et al. / Journal of Experimental Child Psychology 152 (2016) 173–191

Table 2
Fit statistics for models comprising 1 to 6 latent profiles.

Model BIC AIC3 N par Entropy Class.Err

1 3162.012 3147.45 8 1 0
2 2973.014 2942.069 17 0.8 0.04
3 2933.42 2886.092 26 0.83 0.06
4 2935.314 2871.604 35 0.84 0.07
5 2945.249 2865.156 44 0.82 0.11
6 2963.037 2866.562 53 0.83 0.1

Note. Bold indicates best-fitting model. BIC, Bayesian information criterion; AIC3, Akaike’s information criterion with 3 as
penalizing factor; N par, number of parameters in model; Entropy, entropy R2; Class.Err, classification error.

Fig. 1. Deviations from NSM–SM overall mean proportion of accuracy (left y-axis) and median RT (right y-axis) as a function of
profile membership (top panel) and as a function of grade (bottom panel). NSM, nonsymbolic magnitude; SM, symbolic
magnitude.
 C.S. Chew et al. / Journal of Experimental Child Psychology 152 (2016) 173–191 181

and Profile 3 [least sophisticated]. The profiles were associated with math abilities; see below.) Grade
was associated with profiles, v2(6, N = 124) = 48.5, p < .001, c = .70, p < .001; a significant gamma
statistic suggests a greater proportion of older children in the more sophisticated nonsymbolic–
symbolic magnitude profiles and younger children in the less sophisticated profiles. However, as
expected, children from each grade were present in the different profiles, and there were more
children in Profiles 1 and 2 than in Profiles 3 and 4 (see Table 3).
Labels corresponding to the relative nonsymbolic and symbolic magnitude abilities were assigned
to each profile. Profile 1 comprised children who displayed average accuracy on nonsymbolic judg-
ments but higher accuracy on symbolic judgments. They were relatively slow in making both nonsym-
bolic and symbolic judgments; interestingly, they were relatively slower in making symbolic
judgments than in making nonsymbolic judgments (labeled NSM0, SM++, Slow). Profile 2 included chil-
dren who were accurate and fast on both nonsymbolic and symbolic judgments (labeled NSM++, SM++,
Fast). Children in Profile 3 were inaccurate and slow on both nonsymbolic and symbolic judgments;
however, they made far fewer errors and were faster in making nonsymbolic judgments than in mak-
ing symbolic judgments (labeled NSM , SM Slow). Profile 4 children exhibited fast RT on both
nonsymbolic and symbolic judgments but were less accurate on both nonsymbolic and symbolic judg-
ments (labeled NSM , SM , Fast).
One-way analyses of variance (ANOVAs) and Bonferroni-corrected post hoc comparisons showed
that profiles differed from each other in nonsymbolic and symbolic accuracy and RT. For nonsymbolic
accuracy, Welch’s F(3, 45) = 29.09, p < .001, g2 = .42 (Welch’s F is reported because of a violation in the
homogeneity of variance assumption); for symbolic accuracy, Welch’s F(3, 42) = 26.97, p < .001,
g2 = .65. Profile 3 (nonsymbolic: M = .81, SD = .13; symbolic: M = .68, SD = .16) was less accurate than
all other profiles on both nonsymbolic and symbolic judgments (all ps < .001). Profile 2 (nonsymbolic:
M = .97, SD = .02; symbolic: M = .96, SD = .03) was more accurate than Profile 4 (nonsymbolic: M = .90,
SD = .03; symbolic: M = .87, SD = .07) on both nonsymbolic and symbolic judgments (all ps < .01). Of
note, Profile 1 (nonsymbolic: M = .94, SD = .05; symbolic: M = .95, SD = .03) was also more accurate
than Profile 4 on symbolic judgment (p = .001).
For nonsymbolic RT, Welch’s F(3, 51) = 43.41, p < .001, g2 = .44; for symbolic RT, Welch’s F(3, 49)
= 57.09, p < .001, g2 = .51. Profile 2 (nonsymbolic: M = 860.41, SD = 100.04; symbolic: M = 1128.19,
SD = 145.17) and Profile 4 (nonsymbolic: M = 830.86, SD = 98.52; symbolic: M = 1141.19,
SD = 171.43) were faster than Profile 1 (nonsymbolic: M = 1178.68, SD = 196.58; symbolic:
M = 1721.79, SD = 313.77) and Profile 3 (nonsymbolic: M = 1160.64, SD = 272.03; symbolic:
M = 1800.80, SD = 483.39) on both nonsymbolic and symbolic judgments (all ps < .001).

Associations among profiles, cognitive factors, and age

Univariate analyses for cognitive measures/age as a function of profiles are reported in Table 4.
However, these analyses do not account for profile classification errors that can lead to biases. Hence,
we examined the standardized regression coefficients (z-scores) and Wald statistics for each measure
predicting profile memberships in a multivariate model that accounts for classification errors (see
Table 5). The z-scores show the predictive effect of factors for profiles while taking into account other
variables in the model. Findings show that age, VSWM, naming number ability, and basic RT indepen-
dently predicted nonsymbolic–symbolic magnitude profile membership, whereas nonverbal IQ did

Table 3
Numbers assigned to nonsymbolic–symbolic magnitude profiles as a function of grade.

Profile membership Grade n

Kindergarten Year 1 Year 2
Profile 1 (Mage = 78.55 months, SD = 10.37) 27 17 9 53 (43%)
Profile 2 (Mage = 91.66 months, SD = 6.14) 0 11 21 32 (26%)
Profile 3 (Mage = 75 months, SD = 10.23) 15 3 3 21 (17%)
Profile 4 (Mage = 88.33 months, SD = 6.53) 1 8 9 18 (15%)
n 43 39 42 124
182 C.S. Chew et al. / Journal of Experimental Child Psychology 152 (2016) 173–191

Table 4
Means (and standard deviations) of cognitive factors and age across four NSM–SM profiles.

Profile 1 (NSM0, Profile 2 (NSM++, Profile 3 (NSM , Profile 4 (NSM , F-value g2

SM++, Slow) SM++, Fast) SM , Slow) SM , Fast) (and df)
[M (SD)] [M (SD)] [M (SD)] [M (SD)]
VSWM 3.56 (0.95) 4.31 (0.67) 3.69 (0.73) 4.14 (0.82) 6.49*** (3, 120) .14
NN 960.24 (136.19) 839.06 (113.95) 1084.38 (188.69) 887.17 (144.86) 12.19*** (3, 47) .26
Basic RT 535.53 (133.13) 454.29 (82.45) 697.77 (260.03) 539.46 (134.70) 8.72*** (3, 46) .18
RCPM 49.42 (26.98) 62.97 (23.24) 49.52 (24.59) 56.11 (20.55) 2.26 (3, 119) –
Age (in 78.55 (10.37) 91.66 (6.14) 75.00 (10.23) 88.33 (6.53) 26.34*** (3, 51) .36
months)

Note. NSM, nonsymbolic magnitude; SM, symbolic magnitude; NN, naming numbers.
***
Significant at p < .001. Welch’s F reported except for VSWM and RCPM.

Table 5
Three-step latent profile model with covariates predicting NSM–SM profile membership.

NSM–SM profiles
Profile 1 Profile 2 Profile 3 Profile 4
N = 124 Wald p z-score z-score z-score z-score
Age 18.59 <.001 3.82** 2.95** 3.09** 1.73
VSWM 9.19 .027 2.94** 2.06* 0.44 0.85
NN 11.92 .008 0.46 1.88 3.28** 0.92
Basic RT 8.73 .033 1.41 1.59 2.59** 0.91
RCPM 3.83 .28 0.85 1.76 1.58 1.16

Note. NSM, nonsymbolic magnitude; SM, symbolic magnitude; NN, naming numbers.
*
p < .05.
**
p < .01.

not. An increase in age was associated with a reduced likelihood of belonging to Profile 1 (B = 0.09,
SE = 0.02, z = 3.82) and Profile 3 (B = 0.11, SE = 0.04, z = 3.09) and a greater likelihood of belonging
to Profile 2 (B = 0.14, SE = 0.05, z = 2.95). An increase in VSWM was associated with a reduced likeli-
hood of belonging to Profile 1 (B = 0.85, SE = 0.29, z = 2.94) and a greater likelihood of belonging
to Profile 2 (B = 0.72, SE = 0.35, z = 2.06). A poorer naming number ability (B = 0.005, SE = 0.002,
z = 3.28) and longer basic RT (B = 0.003, SE = 0.001, z = 2.59) were associated with a greater likelihood
of belonging to Profile 3.
Children in Profiles 1 and 3 were more likely to be younger, whereas children in Profile 2 were
more likely to be older. Poorer VSWM capacity characterized children in Profile 1, whereas higher
VSWM capacity characterized children in Profile 2. Poorer naming number ability and slower basic
RT characterized children in Profile 3.
The significant model predictors for each profile were compared with the other three profiles, sim-
ilar to a multinomial logistic regression. Children who had poorer VSWM were more likely to belong to
Profile 1 relative to Profile 2 (B = 1.57, SE = 0.55, z = 2.84) and Profile 4 (B = 1.12, SE = 0.52,
z = 2.16). Children who performed poorer on naming number were more likely to belong to Profile
3 compared with Profile 1 (B = 0.005, SE = 0.002, z = 2.70), Profile 2 (B = 0.009, SE = 0.003,
z = 3.00), and Profile 4 (B = 0.008, SE = 0.004, z = 2.10). Children who showed longer basic RT were
also more likely to belong to Profile 3 compared with Profile 1 (B = 0.005, SE = 0.002, z = 2.67) and
Profile 2 (B = 0.006, SE = 0.003, z = 2.27). For an increase in age, children were more likely to belong
to Profile 2 (B = 0.23, SE = 0.06, z = 3.62; B = 0.25, SE = 0.07, z = 3.34) and Profile 4 (B = 0.15, SE = 0.05,
z = 3.19; B = 0.17, SE = 0.06, z = 2.88) relative to Profiles 1 and 3, respectively.

Associations between profiles and math abilities

SDA problems solved correctly differed as a function of profile membership, Welch’s F(3, 47)
= 16.07, p < .001, g2 = .26. Profile 3 (M = 50.32, SD = 37.12) solved fewer SDA problems compared with
 C.S. Chew et al. / Journal of Experimental Child Psychology 152 (2016) 173–191 183

Profile 1 (M = 72.58, SD = 31.6; p = .008), Profile 2 (M = 94.27, SD = 7.39; p < .001), and Profile 4
(M = 90.19, SD = 10.87; p < .001). Children in Profile 1 also had lower SDA accuracy compared with Pro-
file 2 (p = .002).
Accuracy in reading teen, two-digit, three-digit, and four-digit numbers showed reasonably good
internal consistency (Cronbach’s alpha = .80); hence, accuracy was summed across these digit strings.
A one-way ANOVA of transcoding (i.e., reading numbers correctness) as a function of profile member-
ship showed a significant effect, Welch’s F(3, 48) = 27.69, p < .001, g2 = .34. Profile 1 (M = 59.12,
SD = 26.98) and Profile 3 (M = 46.03, SD = 27.34) were less accurate compared with Profile 2
(M = 89.84, SD = 12.02) and Profile 4 (M = 79.63, SD = 18.9; all ps < .001 except p = .008 for Profile 1
vs. Profile 4).
SDA accuracy and transcoding were regressed on profile membership while accounting for classi-
fication errors. The standardized regression coefficients (z-scores) and Wald statistics for each depen-
dent variable predicted by profile membership were examined. SDA correctness (Wald = 52.65,
p < .001) and transcoding (Wald = 86.83, p < .001) were significantly associated with the profiles. An
increase in SDA accuracy was associated with belonging to Profile 2 (B = 18.59, SE = 2.66, z = 6.98)
and Profile 4 (B = 13.97, SE = 3.25, z = 4.3) and, conversely, with a reduced likelihood of belonging to
Profile 3 (B = 27.63, SE = 6.22, z = 4.44). Belonging to Profile 2 (B = 22.52, SE = 2.88, z = 7.82) and
Profile 4 (B = 11.57, SE = 4.15, z = 2.79) was associated with better transcoding ability, whereas poorer
transcoding ability was associated with belonging to Profile 1 (B = 10.34, SE = 3.64, z = 2.84) and
Profile 3 (B = 23.75, SE = 4.74, z = 5.01).
Each profile was compared with the other three profiles, with all possible comparisons being gen-
erated by alternating the reference category (see Table 6). Children with poorer SDA accuracy and
transcoding ability were more likely to belong to Profiles 1 and 3 relative to Profiles 2 and 4. In addi-
tion, children belonging to Profile 3 also had poorer SDA accuracy relative to children belonging to Pro-
file 1. Findings suggest that SDA accuracy and transcoding were differentially associated with profile

Table 6
Significant effects in three-step latent profile model with dependent variables.

Reference Profile 1 Profile 2 Profile 3 Profile 4

group
Profile 1 – SDA: SDA: SDA:
B = 23.53, SE = 5.21, B = 22.69, SE = 9.52, B = 18.90, SE = 5.63,
z = 4.52** z = 2.38* z = 3.36**
Transcoding: Transcoding: Transcoding:
B = 32.86, SE = 4.77, B = 13.41, SE = 7.22, B = 21.9, SE = 6.28,
z = 6.88** z = 1.86 z = 3.49**
Profile 2 SDA: – SDA: SDA:
B = 23.53, SE = 5.21, B = 46.21, SE = 8.05, B = 4.62, SE = 3.38,
z = 4.52** z = 5.74** z = 1.37
Transcoding: Transcoding: Transcoding:
B = 32.86, SE = 4.77, B = 46.27, SE = 6.32, B = 10.96, SE = 5.64
z = 6.88** z = 7.33** z = 1.94
Profile 3 SDA: SDA: – SDA:
B = 22.69, SE = 9.52, B = 46.22, SE = 8.05, B = 41.59, SE = 8.61,
z = 2.38* z = 5.74** z = 4.83**
Transcoding: Transcoding: Transcoding:
B = 13.41, SE = 7.22, B = 46.27, SE = 6.32, B = 35.31, SE = 7.68,
z = 1.86 z = 7.33** z = 4.60**
Profile 4 SDA: SDA: SDA: –
B = 18.90, SE = 5.63, B = 4.62, SE = 3.38, B = 41.59, SE = 8.61,
z = 3.36** z = 1.37 z = 4.83**
Transcoding: Transcoding: Transcoding:
B = 27.90, SE = 6.28, B = 10.96, SE = 5.64, B = 35.31, SE = 7.68,
z = 3.49** z = 1.94 z = 4.60**

Note. Reference groups are based on the four profile subgroups.

*
p < .05.
**
p < .01.
184 C.S. Chew et al. / Journal of Experimental Child Psychology 152 (2016) 173–191

membership, with children in Profiles 1 and 3 being characterized by poorer math abilities relative to
other children.

Discussion

This study examined whether different profiles of nonsymbolic–symbolic magnitude ability pat-
terns could be identified from a general nonsymbolic–symbolic magnitude ability distribution, the
cognitive factors associated with these profiles, and whether these profiles and factors in combination
were associated with math abilities. Three findings are of note. First, four nonsymbolic–symbolic mag-
nitude ability profiles were identified from LPA, three of which provide separate support for claims
that symbolic magnitude ability is dependent on or independent of nonsymbolic magnitude ability.
Second, differences in VSWM capacity, naming number, and basic RT abilities were differentially asso-
ciated with nonsymbolic–symbolic magnitude profiles. Third, the cognitive factors and nonsymbolic–
symbolic magnitude profiles in combination were differentially associated with SDA problem-solving
and transcoding abilities.

Different nonsymbolic–symbolic magnitude ability profiles

The different profiles identified represent mixtures of different nonsymbolic–symbolic magnitude

relationships residing within the overall ability distribution and support differing accounts of the
association, or lack thereof, between nonsymbolic and symbolic magnitude abilities. Differences in
the observed data highlight a concern about the generalizability of findings from previous research
that assume a homogeneous population sample (i.e., equal variances across the population/small stan-
dard errors) with respect to nonsymbolic–symbolic magnitude abilities, particularly because both
measurement error and associations may vary across profiles.
Children in Profile 2 (see Fig. 1 and Tables 3–5 for profile abilities) were highly accurate and fast on
both nonsymbolic and symbolic judgments compared with other children, suggesting that they pos-
sess relatively good nonsymbolic and symbolic magnitude abilities—a profile consistent with claims of
an association between nonsymbolic and symbolic magnitude abilities (i.e., good symbolic abilities
depend on good nonsymbolic abilities; e.g., Dehaene, 2011; Feigenson, Libertus, & Halberda, 2013;
Piazza et al., 2010). Children in Profile 3 were less accurate and slower on nonsymbolic and symbolic
judgments relative to other children; however, they were much more accurate and faster on nonsym-
bolic judgments compared with symbolic judgments. They displayed relatively poorer symbolic mag-
nitude abilities compared with nonsymbolic magnitude abilities, which is consistent with claims that
symbolic magnitude ability is not due to a less adequate nonsymbolic representation (i.e., symbolic
abilities are independent of nonsymbolic abilities; e.g., Holloway & Ansari, 2009; Rousselle & Noël,
2007; Sasanguie, Defever, Maertens, & Reynvoet, 2014).
Although children in Profile 4 were poorer on nonsymbolic and symbolic judgment accuracy rela-
tive to some children (Profile 2; only on symbolic accuracy relative to Profile 1), their accuracy on both
nonsymbolic and symbolic judgments was relatively close to the overall mean and they were similar
to children in Profile 2 in their speed of responding. Their similar nonsymbolic–symbolic magnitude
abilities suggest a likely association between nonsymbolic and symbolic magnitude abilities (akin
to Profile 2); however, children in this profile appear to display a tendency to be fast in their responses,
with some costs incurred in accuracy. Some researchers have made the point that the accuracy–speed
relationship may be modulated by individual predispositions, suggested from studies that modeled
individual differences in decision making and strategy choices (e.g., Ratcliff et al., 2015; Siegler, 1988).
Children in Profile 1 showed superior accuracy on symbolic judgments relative to nonsymbolic
judgments coupled with relatively long response speeds. They were also more accurate on symbolic
magnitude abilities alone compared with children in Profile 4. Interestingly, they exhibited longer
RT on symbolic judgments relative to nonsymbolic judgments, albeit with higher accuracy. Although
we did not hypothesize this particular pattern of nonsymbolic–symbolic magnitude abilities, this pro-
file is not entirely unexpected. Some children may learn by rote practice, suggesting that symbolic
magnitude abilities can, to a degree, be supported by rote recall. Children typically learn count
 C.S. Chew et al. / Journal of Experimental Child Psychology 152 (2016) 173–191 185

words/sequence by rote and acquire principles that govern counting across contexts (one to one,
stable order, and cardinality; see Butterworth & Reeve, 2012, and Gelman & Gallistel, 1978, for reviews
of counting principles). It is plausible that these children complete symbolic judgments with relative
success (i.e., indexed by accuracy), but their relatively longer RT suggests that doing so requires effort
and, hence, is less efficient. Although the pattern of good symbolic magnitude ability relative to poorer
nonsymbolic ability is a novel finding that may have implications for math abilities, its significance
should be treated with caution and awaits replication.

Cognitive factors associated with nonsymbolic–symbolic magnitude profiles

Poorer VSWM capacity was associated with Profile 1 (i.e., better symbolic magnitude abilities rel-
ative to nonsymbolic abilities), and poorer naming number and basic RT abilities were associated with
Profile 3 (i.e., poorer symbolic magnitude abilities relative to nonsymbolic abilities). In contrast, good
VSWM capacity was associated with Profile 2 (i.e., relatively good nonsymbolic–symbolic abilities).
These findings extend previous research that identified a link between VSWM/naming number/basic
RT and math skills (e.g., Berteletti et al., 2010; Bull & Johnston, 1997; Passolunghi & Mammarella,
2012). Findings also suggest that their significance for math abilities may be partially mediated via
their role for nonsymbolic–symbolic magnitude abilities.
The findings are consistent with claims that the ability to access symbolic numerical information is
a specific cognitive factor for children who display poorer symbolic magnitude abilities relative to bet-
ter nonsymbolic abilities (i.e., Profile 3) (Holloway & Ansari, 2009; Rousselle & Noël, 2007). However, it
should be noted that it is possible that naming numbers may access the name of digits (symbolic num-
ber knowledge) without accessing the semantic meaning of digits (magnitudes) (De Smedt & Gilmore,
2011). Symbolic access ability is typically inferred from symbolic judgment performance, with a few
exceptions (e.g., Mundy & Gilmore, 2009), and there is surprisingly little specification of how this abil-
ity connects symbols to magnitude representation. Although the ambiguity in this construct needs to
be clarified, naming numbers is a useful marker for children displaying symbolic magnitude difficul-
ties despite possessing relatively better nonsymbolic abilities. Profile 3 was also associated with
longer basic RT. Although some studies have found a contributing role of processing speed for numer-
ical processing/math abilities (Bull & Johnston, 1997), others have not (Vanbinst, Ghesquière, & De
Smedt, 2015). It is possible that the significance of processing speed is relevant only for some children
and not all.
Findings also suggest that VSWM may play a role in mediating the relationship between nonsym-
bolic–symbolic magnitude representations. Researchers hypothesize that magnitudes are spatially
organized, analogous to a mental number line, and the processing of magnitudes is thought to be sup-
ported by VSWM (Dehaene & Brannon, 2011). However, poor VSWM abilities did not manifest in
weaker nonsymbolic magnitude abilities per se (i.e., children in Profiles 3 and 4 relative to other chil-
dren); rather, it predicted children who displayed poorer nonsymbolic magnitude abilities relative to
symbolic abilities. On the other hand, good VSWM abilities predicted relatively good nonsymbolic–
symbolic magnitude abilities. This set of findings suggests that VSWM processes may be potentially
important for relative success in connecting numerical symbols to their magnitude referents. Better
VSWM capacity may underpin a more optimal symbolic magnitude development (e.g., good nonsym-
bolic/good symbolic magnitude abilities), whereas poorer VSWM capacity may signal difficulties with
symbolic magnitude development.
Although profiles were defined by relative differences in nonsymbolic–symbolic magnitude
strengths/weaknesses, their different associations with cognitive markers highlight the validity of
the profiles. Researchers have often focused on either general cognitive processes (Geary, 2004;
Szucs, Devine, Soltesz, Nobes, & Gabriel, 2013) or number-specific cognitive processes (Butterworth,
2010; Feigenson, Dehaene, & Spelke, 2004) to explain differences in math abilities. Math cognition
likely builds on a number of general and number-specific cognitive abilities (Jordan et al., 2013;
Träff, 2013), although relatively few studies have focused on a combination of these cognitive abilities
in a single framework (but see Gray & Reeve, 2014). Current findings show that VSWM capacity, nam-
ing numbers, and basic RT are independent correlates of nonsymbolic–symbolic magnitude abilities
186 C.S. Chew et al. / Journal of Experimental Child Psychology 152 (2016) 173–191

and highlight a need to consider more complex models examining the independent and mediated
effects of general and number-specific competencies for math development.

Nonsymbolic–symbolic magnitude profiles, cognitive factors, and math abilities

Nonsymbolic–symbolic magnitude ability profiles were differentially associated with SDA problem
solving and transcoding. Findings support claims of a link between nonsymbolic or symbolic magni-
tude abilities and math skills (Bugden & Ansari, 2011; De Smedt et al., 2009; Mazzocco, Feigenson, &
Halberda, 2011b; Mussolin et al., 2012). Relatively good nonsymbolic–symbolic magnitude abilities
and average nonsymbolic–symbolic magnitude abilities (i.e., Profiles 2 and 4) were associated with
better SDA problem-solving and transcoding skills. In contrast, poorer symbolic magnitude abilities
relative to nonsymbolic abilities (Profile 3) displayed poorer SDA problem-solving and transcoding
skills. This finding is consistent with studies showing that poor symbolic number access and symbolic
judgment difficulties are associated with poorer math abilities (De Smedt & Gilmore, 2011; Rousselle
& Noël, 2007).
Profile 1 (better symbolic magnitude abilities relative to nonsymbolic abilities) also displayed
poorer SDA problem-solving and transcoding abilities. Contrary to some studies claiming that only
symbolic magnitude abilities are important for math abilities (Holloway & Ansari, 2009; Vanbinst,
Ceulemans, et al., 2015), the current findings suggest that relatively good symbolic magnitude abilities
alone do not guarantee good SDA and transcoding abilities. Although children in this profile showed
effortful success (i.e., relatively accurate but slow) in making symbolic judgments, they appeared to
experience difficulty with transcoding and SDA problem solving. This pattern of finding is analogous
to that observed in some studies of individuals with learning difficulties, where children are often cap-
able of learning (e.g., the Arabic number system) but are not able to effectively use this information in
relevant situations (e.g., math tasks) (Brown, Campione, Webber, & McGilly, 1992; Gersten, Jordan, &
Flojo, 2005; Namkung & Fuchs, 2012).
Nonsymbolic magnitude abilities are evidently important for good symbolic magnitude abilities
(i.e., good nonsymbolic–symbolic magnitude abilities in Profile 2) and ipso facto math abilities. In con-
trast, relatively poor nonsymbolic magnitude abilities might not support developing symbolic abilities
and in turn math abilities (i.e., Profile 3 with relatively poor nonsymbolic abilities compared with
other children but also possessing even poorer symbolic abilities). This is in line with studies claiming
that nonsymbolic magnitude abilities have a crucial role in early development of symbolic number/
math abilities (Dehaene, 2011; Halberda & Feigenson, 2008). However, differences in nonsymbolic
magnitude abilities is not the only explanatory factor involved in symbolic magnitude abilities, and
other factors (e.g., VSWM, number naming ability) play a significant role.

Profile versus age-based patterns of nonsymbolic–symbolic magnitude ability

Nonsymbolic–symbolic magnitude profile membership only partially overlapped with grade.

Although there was a trend for older children to belong to Profile 2, and for younger children to belong
to Profiles 1 and 3, there were exceptions in all profiles. Some older children were found in Profiles 1
and 3, and some younger children were found in Profiles 2 and 4. Our findings are consistent with
recent studies showing that identifying profiles of competence provided insights into the mechanisms
underlying cognitive phenomena, which may be masked by focusing on age differences (Gray & Reeve,
2014; Paul & Reeve, 2016). The current results caution against using age as a proxy for development
per se, particularly when many variables are changing at this age. Using LPA to identify subgroups is
valuable for theoretical reasons (i.e., children who share similar nonsymbolic–symbolic magnitude
abilities) and statistical power (i.e., minimizing within-group differences while maximizing
between-group differences). By allowing LPA to characterize age variability, we were able to examine
how age is related to the nonsymbolic–symbolic magnitude profiles a posteriori and the factors
related to profiles after taking age into account.
Our findings suggest that there is no one single nonsymbolic–symbolic magnitude ability pattern
linked to math abilities per se. Instead, different cognitive factors (VSWM, naming number, and basic
RT abilities) contribute differentially to the four profiles after age is taken into account (e.g., Profile 2 is
 C.S. Chew et al. / Journal of Experimental Child Psychology 152 (2016) 173–191 187

characterized by poorer VSWM, Profile 3 is characterized by poorer naming number and basic RT abil-
ities). Profiles and their differentially associated factors in combination predict math abilities. In addi-
tion, performance differences on cognitive indexes are often taken to directly reflect the underlying
construct directly (e.g., nonsymbolic and symbolic magnitude representations). However, this
assumption should be treated with caution because it is not always clear whether performance
reflects the construct, stimulus properties, or is biased by other factors separate from the construct
itself (see Karolis, Iuculano, & Butterworth, 2011, for examples of response biases examining the sub-
jective scale of numerical representation). We suggest that profiles of relative nonsymbolic–symbolic
magnitude strengths/weaknesses provide a useful way of characterizing heterogeneous group distri-
butions of nonsymbolic–symbolic magnitude abilities.

Implications and directions for future research

Although some researchers have suggested that different nonsymbolic–symbolic magnitude rela-
tionships and their respective theoretical positions might not be incompatible (Mussolin et al.,
2010), this is the first study to use latent profile analysis to examine this possibility. The finding of four
profiles supports the view that inter-individual variability—which may be obscured by the variable-
centered method—can be identified. Although latent profile analysis provided a way of characterizing
nonsymbolic–symbolic magnitude abilities, the findings should be treated with some caution and pro-
file validation is recommended (see Lanza & Cooper, 2016, for a discussion about replication and sam-
ple size issues in LPA research). Nevertheless, the profiles identified here were associated with
cognitive and math abilities in expected ways, emphasizing their validity.
Other studies have also used latent class analysis to identify math ability profiles. Reeve and
colleagues (2012), for example, showed that dot enumeration profiles (i.e., the ability to accurately
and quickly enumerate small sets of dots) were differentially related to early math abilities (see also
Gray & Reeve, 2016). It is possible that nonsymbolic–symbolic magnitude profiles found here may be
related to the dot enumeration profiles. Insofar as this is correct, the profile overlap may provide
important information about patterns of early math development.
Although different profiles of nonsymbolic–symbolic magnitude abilities (and cognitive and math
factors associated with them) were identified, research is needed to determine change and/or stability
over time in the relative numbers of children in the different profiles. (We expect that the numbers of
children in the profiles may vary as a function of sampling factors; however, the profiles themselves
will not change.) For example, Profile 2 (good nonsymbolic/good symbolic magnitude ability) com-
prised 26% of the sample. The numbers of children assigned to this profile would likely increase as
symbolic magnitude abilities increased. Although differences in the profiles identified suggest differ-
ent nonsymbolic–symbolic magnitude developmental pathways, they may represent different routes
to nonsymbolic–symbolic magnitude competence (equifinality). However, they may also be indicators
of difficulties that persist over time. For instance, children in Profile 3 (i.e., relatively poorer symbolic
magnitude abilities compared with nonsymbolic abilities) would likely improve in their abilities to
make nonsymbolic judgments, which in turn may support their symbolic development. For children
in Profile 1, it is possible that they may continue to develop symbolic magnitude abilities, which in
turn may refine nonsymbolic abilities; what this might mean for math abilities is unclear
(Mussolin, Nys, Content, & Leybaert, 2014; Noël & Rousselle, 2011). Nevertheless, the possibility of
an iterative influence between nonsymbolic and symbolic magnitude ability needs to be clarified
(Rittle-Johnson, Siegler, & Alibali, 2001). Moreover, the relationships among changes in profile
membership, cognitive mechanisms, and math abilities will need explication.

Conclusion

Mixed findings from studies examining the way(s) in which nonsymbolic and symbolic magnitude
abilities are related to each other raise questions about the relationship and how it is related to cog-
nitive and early math abilities. Using latent class analysis, we identified four different nonsymbolic–
symbolic magnitude ability profiles, which allowed us to construct a comprehensive framework of the
relationships among these profiles, cognitive abilities, and math abilities. Although nonsymbolic
188 C.S. Chew et al. / Journal of Experimental Child Psychology 152 (2016) 173–191

magnitude ability was important for symbolic magnitude ability, VSWM, naming numbers, and basic
RT ability also played a role in predicting young children’s math ability. The patterns of findings sug-
gest that different developmental math pathways may exist; whether these pathways are different
routes to the same point is a matter for future research.

Acknowledgments

We thank the editor and three reviewers for their valuable comments. We also thank our col-
leagues in the Cognitive Neuropsychological Development Lab at the University of Melbourne for com-
ments on earlier drafts of the manuscript.

Appendix A. Supplementary data

Supplementary data associated with this article can be found, in the online version, at http://dx.doi.
org/10.1016/j.jecp.2016.07.001.

References

Ansari, D. (2010). Neurocognitive approaches to developmental disorders of numerical and mathematical cognition: The perils
of neglecting the role of development. Learning and Individual Differences, 20, 123–129.
Bailey, D. H., Littlefield, A., & Geary, D. C. (2012). The codevelopment of skill at and preference for use of retrieval-based
processes for solving addition problems: Individual and sex differences from first to sixth grades. Journal of Experimental
Child Psychology, 113, 78–92.
Bakk, Z., Tekle, F. B., & Vermunt, J. K. (2013). Estimating the association between latent class membership and external variables
using bias-adjusted three-step approaches. Sociological Methodology, 43, 272–311.
Bartelet, D., Ansari, D., Vaessen, A., & Blomert, L. (2014). Cognitive subtypes of mathematics learning difficulties in primary
education. Research in Developmental Disabilities, 35, 657–670.
Berteletti, I., Lucangeli, D., Piazza, M., Dehaene, S., & Zorzi, M. (2010). Numerical estimation in preschoolers. Developmental
Psychology, 46, 545–551.
Bonny, J. W., & Lourenco, S. F. (2013). The approximate number system and its relation to early math achievement: Evidence
from the preschool years. Journal of Experimental Child Psychology, 114, 375–388.
Bouwmeester, S., & Verkoeijen, P. P. J. L. (2012). Multiple representations in number line estimation: A developmental shift or
classes of representations? Cognition and Instruction, 30, 246–260.
Brown, A. L., Campione, J. C., Webber, L. S., & McGilly, K. (1992). Interactive learning environments: A new look at assessment
and instruction. In B. R. Gifford & M. C. O’Connor (Eds.), Changing assessments (pp. 121–212). Boston: Kluwer Academic.
Bugden, S., & Ansari, D. (2011). Individual differences in children’s mathematical competence are related to the intentional but
not automatic processing of Arabic numerals. Cognition, 118, 35–47.
Bull, R., Espy, K. A., & Wiebe, S. A. (2008). Short-term memory, working memory, and executive functioning in preschoolers:
Longitudinal predictors of mathematical achievement at age 7 years. Developmental Neuropsychology, 33, 205–228.
Bull, R., & Johnston, R. S. (1997). Children’s arithmetical difficulties: Contributions from processing speed, item identification,
and short-term memory. Journal of Experimental Child Psychology, 65, 1–24.
Butterworth, B. (2010). Foundational numerical capacities and the origins of dyscalculia. Trends in Cognitive Sciences, 14,
534–541.
Butterworth, B., & Reeve, R. (2012). Counting words and a principles-after account of the development of number concepts. In
M. Siegal & L. Surian (Eds.), Access to language and cognitive development (pp. 162–175). Oxford, UK: Oxford University Press.
Chen, Q., & Li, J. (2014). Association between individual differences in non-symbolic number acuity and math performance: A
meta-analysis. Acta Psychologica, 148, 163–172.
Clark, S., & Muthen, B. (2009). Relating latent class analysis results to variables not included in the analysis. Retrieved from: https://
www.statmodel.com/download/relatinglca.pdf.
Collins, L. M., & Lanza, S. T. (2013). Latent class and latent transition analysis: With applications in the social, behavioral, and health
sciences. Hoboken, NJ: John Wiley.
Cotton, S. M., Kiely, P. M., Crewther, D. P., Thomson, B., Laycock, R., & Crewther, S. G. (2005). A normative and reliability study for
the Raven’s Coloured Progressive Matrices for primary school aged children from Victoria, Australia. Personality and
Individual Differences, 39, 647–659.
de Hevia, M. D., Vallar, G., & Girelli, L. (2008). Visualizing numbers in the mind’s eye: The role of visuo-spatial processes in
numerical abilities. Neuroscience and Biobehavioral Reviews, 32, 1361–1372.
De Smedt, B., & Gilmore, C. K. (2011). Defective number module or impaired access? Numerical magnitude processing in first
graders with mathematical difficulties. Journal of Experimental Child Psychology, 108, 278–292.
De Smedt, B., Noël, M.-P., Gilmore, C., & Ansari, D. (2013). How do symbolic and non-symbolic numerical magnitude processing
skills relate to individual differences in children’s mathematical skills? A review of evidence from brain and behavior. Trends
in Neuroscience and Education, 2(2), 48–55.
De Smedt, B., Verschaffel, L., & Ghesquière, P. (2009). The predictive value of numerical magnitude comparison for individual
differences in mathematics achievement. Journal of Experimental Child Psychology, 103, 469–479.
Dehaene, S. (1992). Varieties of numerical abilities. Cognition, 44, 1–42.
 C.S. Chew et al. / Journal of Experimental Child Psychology 152 (2016) 173–191 189

Dehaene, S. (2007). Symbols and quantities in parietal cortex: Elements of a mathematical theory of number representation and
manipulation. In P. Haggard (Ed.), Sensorimotor foundations of higher cognition (pp. 527–574). Oxford, UK: Oxford University
Press.
Dehaene, S. (2011). The number sense: How the mind creates mathematics (rev. ed.). Oxford, UK: Oxford University Press.
Dehaene, S., & Brannon, E. M. (2011). Space, time, and number in the brain: Searching for the foundations of mathematical thought.
London: Academic Press.
Dehaene, S., & Cohen, L. (1997). Cerebral pathways for calculation: Double dissociation between rote verbal and quantitative
knowledge of arithmetic. Cortex, 33, 219–250.
Dehaene, S., Izard, V., & Piazza, M. (2005). Control over non-numerical parameters in numerosity experiments. Unpublished
manuscript. Retrieved from: http://www.unicog.org.
Dehaene, S., Piazza, M., Pinel, P., & Cohen, L. (2003). Three parietal circuits for number processing. Cognitive Neuropsychology, 20,
487–506.
Deloche, G., & Seron, X. (1987). Numerical transcoding: A general production model. In G. Deloche & X. Seron (Eds.),
Mathematical disabilities: A cognitive neuropsychological perspective (pp. 137–170). Hillsdale, NJ: Lawrence Erlbaum.
Dowker, A. (2008). Individual differences in numerical abilities in preschoolers. Developmental Science, 11, 650–654.
Fazio, L. K., Bailey, D. H., Thompson, C. A., & Siegler, R. S. (2014). Relations of different types of numerical magnitude
representations to each other and to mathematics achievement. Journal of Experimental Child Psychology, 123, 53–72.
Feigenson, L., Dehaene, S., & Spelke, E. (2004). Core systems of number. Trends in Cognitive Sciences, 8, 307–314.
Feigenson, L., Libertus, M. E., & Halberda, J. (2013). Links between the intuitive sense of number and formal mathematics ability.
Child Development Perspectives, 7(2), 74–79.
Fuchs, L. S., Geary, D. C., Compton, D. L., Fuchs, D., Hamlett, C. L., Seethaler, P. M., et al (2010). Do different types of school
mathematics development depend on different constellations of numerical versus general cognitive abilities?
Developmental Psychology, 46, 1731–1746.
Geary, D. C. (2000). From infancy to adulthood: The development of numerical abilities. European Child and Adolescent
Psychiatry, 9(Suppl 2), 11–16.
Geary, D. C. (2004). Mathematics and learning disabilities. Journal of Learning Disabilities, 37, 4–15.
Geary, D. C. (2011). Cognitive predictors of achievement growth in mathematics: A 5-year longitudinal study. Developmental
Psychology, 47, 1539–1552.
Geary, D. C., Hoard, M. K., & Bailey, D. H. (2012). Fact retrieval deficits in low achieving children and children with mathematical
learning disability. Journal of Learning Disabilities, 45, 291–307.
Gebuis, T., Cohen Kadosh, R., de Haan, E., & Henik, A. (2009). Automatic quantity processing in 5-year-olds and adults. Cognitive
Processing, 10, 133–142.
Gelman, R., & Gallistel, C. (1978). The child’s understanding of number. Cambridge, MA: Harvard University Press.
Gersten, R., Jordan, N. C., & Flojo, J. R. (2005). Early identification and mathematics difficulties. Journal of Learning Disabilities, 38,
293–304.
Gilmore, C., Attridge, N., De Smedt, B., & Inglis, M. (2014). Measuring the approximate number system in children: Exploring the
relationships among different tasks. Learning and Individual Differences, 29, 50–58.
Gilmore, C., Attridge, N., & Inglis, M. (2011). Measuring the approximate number system. Quarterly Journal of Experimental
Psychology, 64, 2099–2109.
Gray, S. A., & Reeve, R. A. (2014). Preschoolers’ dot enumeration abilities are markers of their arithmetic competence. PLoS ONE, 9
(4), e94428.
Gray, S. A., & Reeve, R. A. (2016). Number-specific and general cognitive markers of preschoolers’ math ability profiles. Journal of
Experimental Child Psychology, 147, 1–21.
Halberda, J., & Feigenson, L. (2008). Developmental change in the acuity of the ‘‘number sense”: The approximate number
system in 3-, 4-, 5-, and 6-year-olds and adults. Developmental Psychology, 44, 1457–1465.
Halberda, J., Mazzocco, M. M. M., & Feigenson, L. (2008). Individual differences in non-verbal number acuity correlate with
maths achievement. Nature, 455, 665–668.
Hecht, S. A., Torgesen, J. K., Wagner, R. K., & Rashotte, C. A. (2001). The relations between phonological processing abilities and
emerging individual differences in mathematical computation skills: A longitudinal study from second to fifth grades.
Journal of Experimental Child Psychology, 79, 192–227.
Holloway, I. D., & Ansari, D. (2009). Mapping numerical magnitudes onto symbols: The numerical distance effect and individual
differences in children’s mathematics achievement. Journal of Experimental Child Psychology, 103, 17–29.
Holmes, J., & Adams, J. W. (2006). Working memory and children’s mathematical skills: Implications for mathematical
development and mathematics curricula. Educational Psychology, 26, 339–366.
Inglis, M., Attridge, N., Batchelor, S., & Gilmore, C. (2011). Non-verbal number acuity correlates with symbolic mathematics
achievement: But only in children. Psychonomic Bulletin & Review, 18, 1222–1229.
Iuculano, T., Tang, J., Hall, C. W. B., & Butterworth, B. (2008). Core information processing deficits in developmental dyscalculia
and low numeracy. Developmental Science, 11, 669–680.
Jordan, N. C., Hansen, N., Fuchs, L. S., Siegler, R. S., Gersten, R., & Micklos, D. (2013). Developmental predictors of fraction
concepts and procedures. Journal of Experimental Child Psychology, 116, 45–58.
Jordan, J. A., Mulhern, G., & Wylie, J. (2009). Individual differences in trajectories of arithmetical development in typically
achieving 5- to 7-year-olds. Journal of Experimental Child Psychology, 103, 455–468.
Kail, R. V., & Ferrer, E. (2007). Processing speed in childhood and adolescence: Longitudinal models for examining
developmental change. Child Development, 78, 1760–1770.
Karolis, V., Iuculano, T., & Butterworth, B. (2011). Mapping numerical magnitudes along the right lines: Differentiating between
scale and bias. Journal of Experimental Psychology: General, 140, 693–706.
Kessels, R. P. C., van Zandvoort, M. J. E., Postma, A., Kappelle, L. J., & de Haan, E. H. F. (2000). The Corsi block-tapping task:
Standardization and normative data. Applied Neuropsychology, 7, 252–258.
Landerl, K., & Kölle, C. (2009). Typical and atypical development of basic numerical skills in elementary school. Journal of
Experimental Child Psychology, 103, 546–565.
190 C.S. Chew et al. / Journal of Experimental Child Psychology 152 (2016) 173–191

Lanza, S. T., Collins, L. M., Lemmon, D. R., & Schafer, J. L. (2007). PROC LCA: A SAS procedure for latent class analysis. Structural
Equation Modeling, 14, 671–694.
Lanza, S. T., & Cooper, B. R. (2016). Latent class analysis for developmental research. Child Development Perspectives, 10(1),
59–64.
Le Corre, M., & Carey, S. (2007). One, two, three, four, nothing more: An investigation of the conceptual sources of the verbal
counting principles. Cognition, 105, 395–438.
Luwel, K., Foustana, A., Onghena, P., & Verschaffel, L. (2013). The role of verbal and performance intelligence in children’s
strategy selection and execution. Learning and Individual Differences, 24, 134–138.
Lyons, I. M., Ansari, D., & Beilock, S. L. (2012). Symbolic estrangement: Evidence against a strong association between numerical
symbols and the quantities they represent. Journal of Experimental Psychology: General, 141, 635–641.
Lyons, I. M., Price, G. R., Vaessen, A., Blomert, L., & Ansari, D. (2014). Numerical predictors of arithmetic success in Grades 1–6.
Developmental Science, 17, 714–726.
Magidson, J., & Vermunt, J. K. (2005). A nontechnical introduction to latent class models. DMA Research Council Journal. Retrieved
from: http://www.statisticalinnovations.com/articles/lcmodels2.pdf.
Magnusson, D. (2003, Fall). The person approach: Concepts, measurement models, and research strategy. New Directions for
Child and Adolescent Development, pp. 3–23. Retrieved from: http://doi.org/10.1002/cd.79.
Mammarella, I. C., Lucangeli, D., & Cornoldi, C. (2010). Spatial working memory and arithmetic deficits in children with
nonverbal learning difficulties. Journal of Learning Disabilities, 43, 455–468.
Mazzocco, M., Feigenson, L., & Halberda, J. (2011a). Impaired acuity of the approximate number system underlies mathematical
learning disability (dyscalculia). Child Development, 82, 1224–1237.
Mazzocco, M. M., Feigenson, L., & Halberda, J. (2011b). Preschoolers’ precision of the approximate number system predicts later
school mathematics performance. PLoS ONE, 6(9), e23749.
Moyer, R. S., & Landauer, T. K. (1967). Time required for judgements of numerical inequality. Nature, 215, 1519–1520.
Mundy, E., & Gilmore, C. K. (2009). Children’s mapping between symbolic and nonsymbolic representations of number. Journal
of Experimental Child Psychology, 103, 490–502.
Mussolin, C., Mejias, S., & Noël, M.-P. (2010). Symbolic and nonsymbolic number comparison in children with and without
dyscalculia. Cognition, 115, 10–25.
Mussolin, C., Nys, J., Content, A., & Leybaert, J. (2014). Symbolic number abilities predict later approximate number system
acuity in preschool children. PLoS ONE, 9(3), e91839.
Mussolin, C., Nys, J., Leybaert, J., & Content, A. (2012). Relationships between approximate number system acuity and early
symbolic number abilities. Trends in Neuroscience and Education, 1(1), 21–31.
Namkung, J. M., & Fuchs, L. S. (2012). Early numerical competencies of students with different forms of mathematics difficulty.
Learning Disabilities Research & Practice, 27, 2–11.
Noël, M.-P., & Rousselle, L. (2011). Developmental changes in the profiles of dyscalculia: An explanation based on a double
exact-and-approximate number representation model. Frontiers in Human Neuroscience, 5. http://dx.doi.org/10.3389/
fnhum.2011.00165.
Organization for Economic Cooperation and Development (2012). Education at a glance 2012: OECD indicators. Paris: OECD
Publishing.
Passolunghi, M. C., & Mammarella, I. C. (2012). Selective spatial working memory impairment in a group of children with
mathematics learning disabilities and poor problem-solving skills. Journal of Learning Disabilities, 45, 341–350.
Paul, J. M., & Reeve, R. A. (2016). Relationship between single digit addition strategies and working memory reflects general
reasoning sophistication. Learning and Instruction, 42, 113–122.
Piazza, M. (2010). Neurocognitive start-up tools for symbolic number representations. Trends in Cognitive Sciences, 14, 542–551.
Piazza, M., Facoetti, A., Trussardi, A. N., Berteletti, I., Conte, S., Lucangeli, D., et al (2010). Developmental trajectory of number
acuity reveals a severe impairment in developmental dyscalculia. Cognition, 116, 33–41.
Price, G. R., Palmer, D., Battista, C., & Ansari, D. (2012). Nonsymbolic numerical magnitude comparison: Reliability and validity
of different task variants and outcome measures, and their relationship to arithmetic achievement in adults. Acta
Psychologica, 140, 50–57.
Ratcliff, R., Thompson, C. A., & McKoon, G. (2015). Modeling individual differences in response time and accuracy in numeracy.
Cognition, 137, 115–136.
Raven, J., Court, J. H., & Raven, J. C. (1995). Coloured Progressive Matrices. Oxford, UK: Oxford Psychologists Press.
Reeve, R., Reynolds, F., Humberstone, J., & Butterworth, B. (2012). Stability and change in markers of core numerical
competencies. Journal of Experimental Psychology: General, 141, 649–666.
Rittle-Johnson, B., Fyfe, E. R., Loehr, A. M., & Miller, M. R. (2015). Beyond numeracy in preschool: Adding patterns to the
equation. Early Childhood Research Quarterly, 31, 101–112.
Rittle-Johnson, B., Siegler, R. S., & Alibali, M. W. (2001). Developing conceptual understanding and procedural skill in
mathematics: An iterative process. Journal of Educational Psychology, 93, 346–362.
Rousselle, L., & Noël, M.-P. (2007). Basic numerical skills in children with mathematics learning disabilities: A comparison of
symbolic vs non-symbolic number magnitude processing. Cognition, 102, 361–395.
Rousselle, L., & Noël, M.-P. (2008). The development of automatic numerosity processing in preschoolers: Evidence for
numerosity–perceptual interference. Developmental Psychology, 44, 544–560.
Sasanguie, D., De Smedt, B., Defever, E., & Reynvoet, B. (2012). Association between basic numerical abilities and mathematics
achievement. British Journal of Developmental Psychology, 30, 344–357.
Sasanguie, D., Defever, E., Maertens, B., & Reynvoet, B. (2014). The approximate number system is not predictive for symbolic
number processing in kindergarteners. Quarterly Journal of Experimental Psychology, 67, 271–280.
Schneider, M., Beeres, K., Coban, L., Merz, S., Susan Schmidt, S., Stricker, J., et al (2016). Associations of non-symbolic and
symbolic numerical magnitude processing with mathematical competence: A meta-analysis. Developmental Science. http://
dx.doi.org/10.1111/desc.12372.
Siegler, R. S. (1987). The perils of averaging data over strategies: An example from children’s addition. Journal of Experimental
Psychology: General, 116, 250–264.
 C.S. Chew et al. / Journal of Experimental Child Psychology 152 (2016) 173–191 191

Siegler, R. S. (1988). Individual differences in strategy choices: Good students, not-so-good students, and perfectionists. Child
Development, 59, 833–851.
Szucs, D., Devine, A., Soltesz, F., Nobes, A., & Gabriel, F. (2013). Developmental dyscalculia is related to visuo-spatial memory and
inhibition impairment. Cortex, 49, 2674–2688.
Träff, U. (2013). The contribution of general cognitive abilities and number abilities to different aspects of mathematics in
children. Journal of Experimental Child Psychology, 116, 139–156.
Vanbinst, K., Ceulemans, E., Ghesquière, P., & De Smedt, B. (2015). Profiles of children’s arithmetic fact development: A model-
based clustering approach. Journal of Experimental Child Psychology, 133, 29–46.
Vanbinst, K., Ghesquière, P., & De Smedt, B. (2012). Numerical magnitude representations and individual differences in
children’s arithmetic strategy use. Mind, Brain, and Education, 6(3), 129–136.
Vanbinst, K., Ghesquière, P., & De Smedt, B. (2015). Does numerical processing uniquely predict first graders’ future
development of single-digit arithmetic? Learning and Individual Differences, 37, 153–160.
Vermunt, J. K. (2010). Latent class modeling with covariates: Two improved three-step approaches. Political Analysis, 18,
450–469.
Vermunt, J. K., & Magidson, J. (2013). Technical guide for Latent GOLD 5.0: Basic, advanced, and syntax. Belmont, CA: Statistical
Innovations.
Weinert, F., & Helmke, A. (1998). The neglected role of individual differences in theoretical models of cognitive development.
Learning and Instruction, 8, 309–323.
Wohlwill, J. F. (1973). The study of behavioral development. New York: Academic Press.
Zorzi, M., Priftis, K., & Umiltà, C. (2002). Brain damage: Neglect disrupts the mental number line. Nature, 417, 138–139.