HHS Public Access

Author manuscript
J Exp Child Psychol. Author manuscript; available in PMC 2018 May 01.
Author Manuscript

Published in final edited form as:

J Exp Child Psychol. 2017 May ; 157: 125–143. doi:10.1016/j.jecp.2016.12.005.

Fluid reasoning predicts future mathematics among children

and adolescents
Chloe T. Green1, Silvia A. Bunge2,3, Victoria Briones Chiongbian4, Maia Barrow2, and
Emilio Ferrer5
1School Psychology Program, Department of Education, University of California, Berkeley
2Helen Wills Neuroscience Institute, University of California Berkeley
Author Manuscript

3Department of Psychology, University of California Berkeley

4MVSStats Consultants
5Department of Psychology, University of California Davis

Abstract
The aim of this longitudinal study was to determine whether fluid reasoning (FR) plays a
significant role in the acquisition of mathematics skills, above and beyond the effects of other
cognitive and numerical abilities. Using a longitudinal cohort sequential design, we examined how
FR measured at three assessment occasions, spaced approximately 1.5 years apart, predicted math
outcomes for a group of 69 participants between ages 6 and 21 across all three assessment
Author Manuscript

occasions. We used structural equation modeling (SEM) to examine the direct and indirect
relations between children's prior cognitive abilities and their future math achievement. A model
including age, FR, vocabulary, and spatial skills accounted for 90% of the variance in future math
achievement. In this model, FR was the only significant predictor of future math achievement;
neither age, vocabulary, nor spatial skills were significant predictors. Thus, FR was the only
predictor of future math achievement across a wide age range that spanned primary and secondary
school. These findings build on Cattell's conceptualization of FR (Cattell, 1987) as a scaffold for
learning, showing that this domain-general ability supports the acquisition of rudimentary math
skills as well as the ability to solve more complex mathematical problems.

Keywords
Author Manuscript

Children; Math; Cognitive Development; Fluid Reasoning; Working Memory; Problem Solving

American educators face the tall order of improving outcomes in science, technology,
engineering, and mathematics (STEM) (Lee, 2012). To generate solutions to some of the

Corresponding Author: Chloe Green ([email protected]).

Present/permanent address: 132 Barker Hall - Helen Wills Neuroscience Institute, University of California, Berkeley, Berkeley, CA
94720
Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our
customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of
the resulting proof before it is published in its final citable form. Please note that during the production process errors may be
discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
 Green et al. Page 2

globe's most pressing challenges, educators will need to teach children to become better
Author Manuscript

problem-solvers who can apply to their work the information learned in their STEM courses.
Courses in mathematics are especially challenging for many students, and these courses have
become a gatekeeper to higher education and job opportunities in technological fields
(Moses & Cobb, 2001). Because math instruction builds upon previously acquired
knowledge and skills, it is difficult for children who fall behind early to catch up with their
peers. In an effort to improve math and language outcomes across the nation, educators have
recently released new national standards for math and language arts education called the
Common Core State Standards (National Governors Association and Council of Chief State
School Offices, 2014). The new standards lay out progressions of math skill building
benchmarks that have opened up discussions about how teachers can provide better support
for students in bolstering their math proficiency skills.

Complementary lines of research in psychology and education aim to identify which

Author Manuscript

cognitive precursors lead to proficient acquisition of mathematics skills. A long-term aim of

this line of research is to inform educators about the precursors to math development, so that
they may create lesson plans that target not only specific math skills, but also underlying
domain-general cognitive processes. The cognitive abilities required to solve math problems
have been difficult to isolate because mathematics is a heterogeneous subject matter (e.g.,
arithmetic, fractions, geometry, statistics), and problems within the same topic area require
several different operations and computations (e.g., adding, subtracting, multiplying,
dividing). Nevertheless, researchers have begun to identify common key cognitive functions
that are critically important for disparate types of mathematical computations (Bisanz,
Sherman, Rasmussen, & Ho, 2005; Desoete & Gregoire, 2007; Krajewski & Schneider,
2009).
Author Manuscript

Relationships between math and cognitive abilities are often studied within the framework of
the Cattell-Horn-Carroll (CHC) theory, arguably the most comprehensive and empirically
supported theory of cognitive abilities derived from over 70 years of psychometric research
using factor analytic theory (Keith & Reynolds, 2010). The utility of the theory is in
clarifying the relations between cognitive and academic abilities to inform educational and
psychological practices. The most recent revision of this model, by Schneider and McGrew
(2012), includes 16 broad cognitive abilities that each contains more narrow cognitive
abilities within them. This model does not include a general intelligence g factor; rather, it is
based on accumulating evidence that broad and narrow CHC cognitive abilities explain more
variance in specific academic abilities than g alone, and that these specific relationships are
more informative to educational practice than general intelligence (e.g. Floyd, McGrew &
Evans, 2008; McGrew, Flanagan, Keith, & Vanderwood, 1997; Vanderwood, McGrew,
Author Manuscript

Flanagan, & Keith, 2002).

In a recent synthesis of studies investigating the concurrent relationships between CHC

cognitive abilities and achievement measures (CHC-ACH) by McGrew & Wendling (2010),
Fluid Reasoning (FR) was one of three broad cognitive abilities that was consistently related
to mathematical performance in calculation and problem solving at all age ranges
throughout development (the other two were Verbal Comprehension and Processing Speed).
FR was consistently related to future math achievement above and beyond the contribution

J Exp Child Psychol. Author manuscript; available in PMC 2018 May 01.
 Green et al. Page 3

of general intelligence. FR has been defined by contemporary CHC theory as the ability to
Author Manuscript

flexibly and deliberately solve novel problems without using prior information (Schneider &
McGrew; 2012). More specifically, it is the ability to analyze novel problems, identify
patterns and relationships that underpin these problems, and apply logic. On FR tests, one or
more of the following logic abilities is required: 1) induction, the ability to discover an
underlying characteristic (e.g. rule, concept, or trend) that governs a set of materials, 2)
general sequential reasoning (deduction), the ability to start with stated rules or premises and
engage in one or more steps to reach a solution to a novel problem (Schneider & McGrew,
2012). FR tests are commonly administered as part of IQ batteries that are administered to
children in schools or in clinical settings. While FR performance is strongly correlated to
general intelligence (g), as is verbal comprehension, there is unique shared variance among
tests of FR that cannot be accounted for by g alone (McGrew, Flanagan, Keith, &
Vanderwood, 1997).
Author Manuscript

FR Development
In typically developing children, FR begins to emerge during the first two years of life,
increases rapidly in early and middle childhood, continues to increase at a slower rate during
adolescence, and reaches asymptotic values around age 25, after which it begins to decline
(McArdle et al., 2002).

Analyses of longitudinal data from large samples that were used to create norms for the
standardized Woodcock-Johnson Cognitive Abilities testing battery (Schrank & Wendling,
2009; Woodcock, Mather, & McGrew, 2001) reveal that both FR performance (as measured
by Analysis Synthesis and Concept Formation tests) and Math Achievement increase rapidly
during childhood, peaking in late adolescence to age 24 and beginning to decline in
Author Manuscript

adulthood (Ferrer & McArdle, 2004). The trajectories of FR development and improvements
in math abilities parallel one other throughout development – and more so from ages 5-10 as
11-24 (Ferrer & McArdle, 2004). This observation hints at the possibility that FR plays a
bigger role in early math skill development in kindergarten and elementary school than in
higher levels of education. However, additional data are needed to examine the relationships
between these skills over development.

Fluid Reasoning and Math Achievement

Hypothesized link
One mechanism by which FR could support math skill acquisition is related to the fact that
both FR and math problems engage a common underlying cognitive process called relational
Author Manuscript

reasoning, or the ability to jointly consider multiple relationships between different

components of a problem (Halford, Wilson & Phillips, 1998; Miller Singley & Bunge,
2014). According to this framework, understanding mathematics requires the ability to form
abstract representations of quantitative and qualitative relations between variables (Halford,
Wilson & Phillips, 1998). For instance, when children first learn fractions, they must keep
several numerical relationships in mind: whole unit integers have to be understood as
subunits and they must learn to coordinate the value in the numerator and the value in the
denominator (Saxe, Taylor, McIntosh, & Gearhart, 2005).

J Exp Child Psychol. Author manuscript; available in PMC 2018 May 01.
 Green et al. Page 4

Furthermore, solving story word problems requires children to draw conceptual connections
Author Manuscript

between real-world situations and analogous numerical symbols and operations to solve the
problem (Clement, 1982). Another example of relational reasoning is evident in early
algebra, when students are asked to solve for one or more unknown numbers, and must keep
in mind the relationship between numbers on either side of the equal sign to determine
which operand is required to solve for the missing variable. Empirical support for a link
between FR and Math Achievement comes from both cross-sectional and longitudinal
research.

While multiple longitudinal studies have elucidated the importance of spatial skills in math
development (for a review see Mix & Cheng, 2012), only a few longitudinal studies have
explored the unique developmental role of prior FR. Further research is needed to
disentangle the role that each of these two cognitive abilities plays in math development
because, although FR and spatial abilities are highly correlated (Fry & Hale, 1996), they rely
Author Manuscript

on overlapping as well as separable cognitive processes and brain regions. FR tests (e.g.
Matrix Reasoning) not only require spatial skills (including visualization), but additionally
require relational reasoning, or the ability to consider relationships between multiple pieces
of information to detect the underlying conceptual relationship among visual objects, and to
use reasoning to identify and apply rules (Halford, Wilson & Phillips, 1998; Holyoak, 2012;
Bunge & Vendetti, 2015). In regards to the neural correlates, a visuospatial skill that is
commonly implicated in math achievement, visuospatial working memory, relies on the
intraparietal and superior frontal regions (for a review see Klingberg, 2006), whereas the
relational reasoning component of FR relies on the rostrolateral prefrontal coretex and lateral
parietal regions (for reviews see Krawczyk, 2010; Bunge & Vendetti, 2014). Therefore, it is
plausible that visuospatial abilities and FR make unique contributions to math achievement.
Author Manuscript

Longitudinal Precursors to Math Achievement

Fluid Reasoning
As mentioned, there are only a limited numbers of longitudinal studies that have examined
the extent to which FR skills uniquely contribute to the development of math proficiency in
childhood, separately from general IQ, and other domain general cognitive abilities. In one
such study (Fuchs et al., 2010), the authors compared the effect of basic numerical cognition
and other domain-general cognitive abilities measured at the beginning of the school year on
280 1st grade students' development of math problem solving over the course of that
academic year. They found that FR (measured by Matrix Reasoning) in the fall semester was
just as predictive of children's gains in word problem solving over the course of the year as
their basic numerical cognition skills. Primi, Ferrão, and Almeida (2010) showed that 7th
Author Manuscript

and 8th grade students' initial level of FR (measured by tests of numerical, verbal, spatial,
and abstract reasoning) was positively related to their subsequent growth in quantitative
abilities over the course of the next two academic years, such that children with higher FR
ability at the start of the year demonstrated more growth in math over the course of two
academic years than children with lower FR scores.

J Exp Child Psychol. Author manuscript; available in PMC 2018 May 01.
 Green et al. Page 5

Spatial skills
Author Manuscript

Several longitudinal studies have examined the robust role of spatial skills in the
development of math proficiency in childhood (for a review see Mix & Cheng). However,
studies in this literature rely on different operational definitions of spatial ability, such as
visualization and spatial working memory.

Visualization—Visualization is the most commonly studied spatial ability related to

mathematics (Mix & Cheng, 2012). Visualization is the ability to perceive visual patterns
and mentally manipulate them to simulate how they might nook when transformed (e.g.
rotated, changed in size, partially obscured) (Flanagan, Ortiz, & Alfonso, 2013).
Visualization is frequently measured using tests such as Block Design, which measures the
ability to use two-color cubes to construct replicas of two-dimensional, geometric patterns
under timed conditions. This test assesses the ability to mentally transform (or rotate)
Author Manuscript

blocks. One such study by Zhang et al. (2014) found that spatial skills in kindergarteners
(measured by spatial visualization), along with verbal skills, predicted level of arithmetic in
the 1st grade as well as arithmetic growth through the 3rd grade. Another such study by
Casey et al. (2015) examined the predictors in 1st grade of math problem solving in the 5th
grade, comparing the predictive power of performance on Block Design with the predictive
power of verbal and arithmetic skills. They found that Block Design performance in 1st
grade were just as predictive of 5th grade math problem solving as early arithmetic skills.

Visuospatial Working Memory: More recently, studies have also found that visuospatial
working memory, or the ability to temporarily store and process visual information to
complete a task, is a robust predictor of math achievement. For example, Li & Geary (2013)
showed that developmental gains in visuospatial working memory between 1st and 5th grade
Author Manuscript

was a strong predictor of math achievement at the end of 5th grade, as was general
intelligence (measured by WASI Matrix Reasoning, Block Design, Vocabulary, and
Similarities), and in-class attentive behavior. Similarly, LeFevre et al. (2010) advanced a
developmental theory that suggests that three key pathways contribute differentially to early
math development: quantitative, linguistic, and spatial pathways. They found that at age 4-5
years, early spatial attention (measured by spatial span) significantly predicted both number
naming and processing of numerical magnitude two years later.

FR tests may require visualization or spatial working memory, but they are distinguished
from these purely spatial ability tests because they require inductive or general sequential
(deductive) reasoning (Schneider & McGrew, 2012). Factor analysis contributing to CHC
theory has demonstrated that FR measures tap into a separable construct than spatial abilities
(Schneider & McGrew, 2012).
Author Manuscript

Quantitative Skills
Though multiple studies have demonstrated the strong predictive power of early math skills
on math achievement, over and above reading, attentive behavior, and domain general
cognitive predictors, many of these studies have been conducted in populations of primary
school-aged children and are consequently limited to more basic numerical competencies
(e.g. magnitude comparisons, number naming, arithmetic, fractions etc.), and do not

J Exp Child Psychol. Author manuscript; available in PMC 2018 May 01.
 Green et al. Page 6

incorporate measures of FR as a unique factor in their models (e.g., Duncan et al., 2007;
Author Manuscript

Fuchs et al., 2012, LeFevre et al., 2010; Watts et al., 2014). Therefore, more research is
needed to better understand the relative predictive power of FR, and other domain-general
cognitive skills, in relation to prior numerical skills in predicting math achievement across
primary and secondary school grades.

Study Goals
In the present study, we sought to expand upon previous research to evaluate the extent to
which prior FR predicts later math outcomes in children between 6 and 21 years old, above
and beyond other cognitive and numerical abilities that have previously been implicated in
math development. Our aims were threefold: 1) to test a latent model of FR that combines
three well-known psychometric tests, 2) to compare the contribution of prior FR to that of
prior math reasoning in predicting future Math Achievement at T3, 3) to compare the
Author Manuscript

relative contribution of prior FR to spatial skills, verbal skills, and age, in predicting future
Math Achievement at T3. Each of these cognitive abilities has been shown to be a strong
independent predictor of later Math Achievement (e.g. Primi, Ferrão, & Almeida, 2010).

To this end, we collected and analyzed data within the context of a larger longitudinal cohort
sequential design study examining the neurodevelopment of FR. We administered a battery
of age-normed neuropsychological tests to measure FR, as well as vocabulary, and spatial
skills, at three timepoints (∼1.5 years apart) in a group of 69 children who ranged in age
from 6 to 21 at the first assessment. At the second assessment (T2) we assessed participants
on a measure of math reasoning. At the final assessment (T3), we assessed participants on
three different math achievement measures: math problem solving, arithmetic fluency, and
math reasoning.
Author Manuscript

Methods
Participants
Participants were individuals in a longitudinal study designed to examine the cognitive and
neural factors that underlie the development of FR. All participants and their parents gave
their informed assent or consent to participate in the study approved by the Committee for
Protection of Human Subjects. Additionally, all participants were screened for neurological
impairment, psychiatric illness, and history of learning disabilities or developmental delays.

Understanding developmental processes requires longitudinal studies that focus on within-

person changes over time. This study design involved a cohort-sequential design in which
201 participants, ranging from 5 to 15 at the time of recruitment, were assessed at one to
Author Manuscript

three time points with an average delay of 1.5 years between time points. This cohort-
sequential design enabled us to examine both between-person differences and within-person
changes over a 5-year span – the five years of the funded research program – and with fewer
participants than a traditional longitudinal design. This approach provides insight into the
interplay of factors underlying such within-person changes over time i.e., improvements in
cognitive abilities over development (Bell, 1953; McArdle, Ferrer-Caja, & Woodcock, 2002;
Ferrer & McArdle, 2004).

J Exp Child Psychol. Author manuscript; available in PMC 2018 May 01.
 Green et al. Page 7

Parents completed the Child Behavioral Check List (Achenbach, 1991) on behalf of their
Author Manuscript

child. Participants who scored in the clinical range for either externalizing or internalizing
behaviors were excluded from further analyses. Of the 172 children and adolescents enrolled
in the study who scored in the normal range on the Child Behavior Check List, 69
participants successfully completed testing at three time points: T1, T2, and T3 – a
substantive time commitment, involving six long testing sessions (one behavioral and one
brain imaging session at each of the three time points). There was no statistical difference in
performance between children who participated at all three time points as compared with
those who did not.

The mean assessment ages for these 69 participants were 10.18 (SD = 3.32) at T1, 11.67
(SD = 3.35) at T2, and 13.45 (SD = 3.38) at T3. Across all of these participants and
timepoints, data were collected between ages 6 and 21 years. The ethnicity of the sample
reflects the ethnic and racial diversity found in the local population (7.4% Hispanic/Latino,
Author Manuscript

56.21% White, 12.43% Asian, 10.45% Black or African American, 18.4% multiple
ethnicities). Both genders were represented equally (48% males, 52% females). Most of
these children came from middle-class homes, and the majority of families (85%) reported
two adults living in the home. All mothers in the study had completed high school, and the
majority (84%) had completed some post-secondary education, in the form of a Bachelor's
or Associate's degree or a diploma from a vocational college. Most of the children spoke
English at home.

Measures
The behavioral measures selected for our longitudinal study were standardized measures
with very high internal consistency and test-retest reliability, ranging from .94 to .95
(McArdle et al., 2002; McGrew, Werder, & Woodcock, 1991).
Author Manuscript

Fluid Reasoning—FR ability was assessed using three standardized measures, including
the Matrix Reasoning subtest of the Wechsler Abbreviated Scale of Intelligence (WASI;
Wechsler, 1999), and the Analysis Synthesis and Concept Formation subtests of the
Woodcock-Johnson Tests of Cognitive Abilities-Revised (Woodcock, Mather, & McGrew,
2001). Though these three tests are quite different from one another, they were all designed
as measures of FR that rely one or more narrow FR abilities. As shown below, all three tests
loaded onto a single factor “FR” in our sample, which is consistent with prior factor analytic
work contributing to CHC theory (Schneider & McGrew; 2012). Thus, we used scores from
this FR factor in all subsequent analyses (for example of this approach see: Primi et al.,
2010).
Author Manuscript

Matrix Reasoning: This test was modeled after a traditional test of “fluid” or non-verbal
reasoning—Raven's Progressive Matrix Reasoning (Raven, 1938)—and required the
participants to examine an incomplete matrix, or geometric pattern, and then select the
missing piece from five response options arranged according to one or more progression
rules. The Matrix Reasoning subtest assesses FR induction skills, or the ability to identify an
underlying characteristic (e.g. rule or trend) that governs the existing pattern, and then to
choose a missing piece that contains this characteristic.

J Exp Child Psychol. Author manuscript; available in PMC 2018 May 01.
 Green et al. Page 8

Analysis Synthesis: On this test, participants are asked to analyze an incomplete logic
Author Manuscript

puzzle made up of colored squares and to use a key to determine the missing color in the
puzzle. To complete this task successfully, participants must use general sequential (or
deductive) reasoning skills to draw correct conclusions from a color combination key, with
more difficult items requiring a series of sequential steps.

Concept Formation: On this test, participants are asked to view a complete puzzle made up
of colored squares, and to identify and state the “rules” (color and shape) when shown
illustrations of both instances and non-instances of the concept (e.g. red square). The
Concept Formation test requires frequent switching from one rule to another. To complete
this task successfully, participants must use inductive reasoning skills to discover the rule
that governs the puzzle.

Vocabulary—We used the Wechsler Abbreviated Scale of Intelligence (WASI) Vocabulary

Author Manuscript

measure (Wechsler, 1999) to probe crystallized knowledge and, indirectly, semantic

memory. This test is a norm-referenced measure of expressive vocabulary. On this test, the
examiner presents stimulus words to participants and asks them to state each word's
meaning.

Spatial Skills—We administered two tests of Spatial Skills. Spatial Span is considered a
measure of visual memory, or the ability to remember visual images over short periods of
time (less than 30 seconds) (Schneider & McGrew, 2012). Block Design is considered a
measure of visualization, or the ability to mentally organize visual information by analyzing
part-whole relationships when information is presented spatially (Schneider & McGrew,
2012). Based on prior factor analytic work contributing to CHC theory demonstrating that
visual memory and visualization load onto a single factor, we created a factor score called
Author Manuscript

Spatial Skills using multiple imputation in AMOS (Schneider &McGrew, 2012). We used
scores from this Spatial Skills factor in all subsequent analyses.

Spatial Span: The Spatial Span test in the 4th edition of the Wechsler Intelligence Scale for
Children (WISC-IV) is a norm-referenced measure that requires participants to remember a
sequence of spatial locations on a grid in forward and reverse order. The Forward condition
measures spatial attention and short-term visuospatial memory, whereas the Backwards
condition additionally measures the ability to manipulate visuospatial representations in
working memory. Participants' scores on each of the conditions are summed into a Spatial
Span total score.

Block Design: The Block Design test in the Wechsler Abbreviated Scale of Intelligence
Author Manuscript

(WASI; Wechsler, 1999) is a norm-referenced measure that requires participants to perceive

patterns and mentally stimulate how they might look when transformed (e.g. rotated). On the
Block Design test, participants are asked to arrange a set of red-and-white blocks in such a
way as to reproduce a 2-dimensional visual pattern shown on a set of cards. The test is
timed, and scoring is based on both efficiency and accuracy of the pattern reproduction.

Math Achievement—All math measures came from the Woodcock-Johnson III Tests of
Achievement and Cognitive Abilities (WJ III ACH & COG), designed for use across the

J Exp Child Psychol. Author manuscript; available in PMC 2018 May 01.
 Green et al. Page 9

lifespan (Woodcock, Mather, & McGrew, 2001). A math reasoning test was administered at
Author Manuscript

the second timepoint, and three math measures were administered at the final timepoint.

Number Series: We administered the Number Series test from the WJ-III Cognitive
Abilities testing battery at the second and third timepoints as a measure of mathematical
reasoning. On this test, the examiner presents the participant with a page of numerical
sequences that contains a missing number. The participant is asked to complete each
sequence by identifying and applying the rule that applies to the other numbers in the
sequence. As the test advances, the underlying rules become more challenging (e.g., 2, 3, 4,
___? as compared with 15, 18, 21, ___?). Participants are awarded 1 point for each correct
answer and 0 points for each incorrect answer. The examiner discontinues the test when the
participant either finishes all items or misses six consecutive items.

Applied Problems: We administered this WJ-III subtest to measure participants' ability to

Author Manuscript

solve practical math word problems using simple counting, addition, or subtraction
operations at the third timepoint. On the Applied Problems test, a participant is presented
with a picture, (e.g., a group of mixed coins) and asked to listen to a problem (e.g., “How
much money is this?”). To solve a problem, the child must recognize the mathematical
procedure to be followed and perform the appropriate calculations. As the test advances, the
child must carry out more complex operations and have more advanced experience with
each particular concept, such as telling time or solving word problems. Participants are
awarded 1 point for each correct answer, and 0 for each incorrect answer. The examiner
discontinues the test when the child either finishes all items or missed six consecutive items
by the completion of the test page.

Math Fluency: To measure participants' ability to complete basic arithmetic problems, we

Author Manuscript

administered the Math Fluency test on the WJ-III Achievement testing battery at the third
timepoint. This test measures participants' ability to solve simple addition, subtraction, and
multiplication facts within a one-minute time limit. At the beginning of the test, the child is
presented with a worksheet composed of simple arithmetic problems and asked to solve as
many problems as he or she can in one minute.

Hypotheses
Latent Construct of Fluid Reasoning—To examine whether the three tests represent a
common construct, we used confirmatory factor analyses (CFA) to create a latent variable
‘Fluid Reasoning (FR)’ from participants' scores on three different tests at each time point:
Matrix Reasoning, Analysis Synthesis and Concept Formation. CFA procedures were
conducted to test the fit of the data to the FR construct for each time point (Figure 1).
Author Manuscript

Comparing FR and Math Reasoning as predictors of later Math Achievement—

Second, we tested a model including relations from FR skills and math reasoning skills to a
diverse set of math skills at a future time points (see Figure 2). To this end, we created a
math latent variable called ‘Math Achievement’ using three different math tests at T3, each
measuring different math skills: Math Reasoning, Applied Problems, and Math Fluency. As
shown in Figure 2, we hypothesized that FR at T1 and T2 would be the strongest predictors

J Exp Child Psychol. Author manuscript; available in PMC 2018 May 01.
 Green et al. Page 10

of Math Achievement at T3 after accounting for prior Math Reasoning at T2. In the next
Author Manuscript

analyses, we added age to the model.

Comparing FR to verbal and spatial abilities as predictors of later Math

Achievement—Third, we compared the relative contributions between FR and future
Math Achievement in relation to other cognitive abilities that have previously been
implicated in math development: Verbal Reasoning (Vocabulary) and Spatial Skills (Spatial
Span and Block Design). We tested the model hypothesizing that FR skills would remain a
strong predictor of future Math Achievement after accounting for verbal and spatial skills
(see Figure 3).

Results
Missing Value Analysis
Author Manuscript

Since the percentage of missing values for four of the variables was above five (refer to
Appendix A for the percentages of missing values per variable), the pattern of missingness
was assessed via Little's MCAR (Missing Completely at Random) procedure (Tabachnick &
Fidell, 2007). This procedure revealed that the data were missing at random (MCAR: χ2
(336) = 360.32, p = .173). Because of this, we used the Expectation Maximization algorithm
to estimate the model parameters (Tabachnick & Fidell, 2007).

Descriptive Statistics
Means and standard deviations for the study variables are shown in Table 1. Both the raw
and T-scores are presented. The T-scores are standardized scores wherein the mean is 50 and
the standard deviation is 10. Factor scores for FR and Spatial Skills were derived using
multiple imputation in AMOS. As shown in Table 1, all mean raw and T-scores increased
Author Manuscript

numerically over time, with the exception of the FR factor score, which decreases from T2
to T3 because the standardized loadings for the FR factor score in T3 are smaller than the
standardized loadings in T2. When composite (or average) FR scores are generated, it is
evident that performance increases across time. Pearson correlations between study variables
are shown in Table 2.

The Structure of Fluid Reasoning—To test whether Matrix Reasoning, Analysis

Synthesis, and Concept Formation could be combined into a latent factor of FR, we
conducted CFA for each of the time points using AMOS 23 software (Arbuckle, 2015). All
factor loadings were statistically significant (p < .001), with standardized loadings above .
65, thus, supporting a latent factor. Therefore, the FR construct was supported, and we
computed one latent factor for each time period using the three psychometric tests.
Author Manuscript

Comparing FR and Math Reasoning as predictors of later Math Achievement—

We employed structural equation modeling (SEM) to test our second hypothesis that prior
FR at T1 and T2 would be stronger predictors of T3 Math Achievement than T2 Math
Reasoning. This approach also allowed us to examine the effects of the predictor variables
simultaneously on a latent dependent measure. As shown in Table 3, our hypothesis was
supported: prior FR was the strongest predictor of Math Achievement at T3. Any model that

J Exp Child Psychol. Author manuscript; available in PMC 2018 May 01.
 Green et al. Page 11

did not involve FR as a predictor of T3 Math Achievement fit significantly worse and
Author Manuscript

decreased the amount of explained variance. In contrast, removing the path from T2 Math
Reasoning to T3 Math Achievement did not worsen the fit or decrease the amount of
explained variance of T3 Math Achievement. Results are reported in Table 4. This pattern
was similar when using T2 and T3 data only (i.e., eliminating FR T1 from the model).
Including age at T1 and T2 in the model depicted in Figure 2 did not change the results.
Indeed, removing all regression paths from age to the variables of interest (i.e., leaving age
in the model but eliminating its effects) did not worsen the fit.

Comparing prior FR to other cognitive abilities—To test our third hypothesis, we

carried out analyses to measure the relative predictive power of FR and other domain general
cognitive abilities, indexed by Vocabulary and Spatial Skills (Spatial Span and Block
Design), on Math Achievement measured at T3. We tested the model depicted in Figure 3.
Lines in Figure 3 represent hypothesized pathways for the longitudinal predictors of math
Author Manuscript

outcomes. As shown in the figure, all possible pathways between the three types of cognitive
skills and the three types of math outcomes were included in the model. This approach is
based on prior research supporting each of these early cognitive skills as possible predictors
of later Math Achievement (e.g. McGrew & Wendling, 2010; LeFevre et al., 2010).

Results from these analyses showed that both of the hypothesized models in Figure 2 fit the
data well (Tables 5 and 6). For the sake of simplicity, we feature here the results of the
structural model with age having a direct effect on Math Achievement at T3 (Figure 3).
Specifically, FR at T2 significantly predicted Math Achievement at T3, β = .52, p < .001. By
contrast, Spatial Skills at T2 did not significantly predict Math Achievement at T3, β = .19,
p = .205. Vocabulary at T2 also did not significantly predict Math Achievement at T3, β = .
15, p = .205. Similarly, age at T2 did not significantly predict Math Achievement at T3, β
Author Manuscript

= .16, p = .150. The T2 predictors (Age, Spatial Skills, Vocabulary, and FR) accounted for
90.2% of the variance in Math Achievement at T3. In summary, this analysis shows that FR
at T2 was a strong, unique predictor of Math Achievement approximately 1.5 years later.

These analyses also enabled us to test the mediating effect of FR between Age and Math
Achievement at T3. Age significantly predicted FR, and FR significantly predicted Math
Achievement at T3. Therefore, the first two criteria of mediation were met. As shown in
Table 7, the indirect effect was statistically significant, p < .001, but the direct effect was not,
p = .198. Therefore, the third and fourth criteria for mediation were met. As such, FR
significantly mediated the relationship between Age and Math Achievement at T3. By
contrast, Vocabulary and Spatial Skills did not significantly predict Math Achievement; thus,
these factors did not significantly mediate the relationship between age and Math
Author Manuscript

Achievement at T3.

Discussion
Summary of results
In this paper, we sought to test whether FR, or the ability to analyze novel problems, identify
patterns and relationships, and apply logic, contributes to future math achievement
throughout primary and secondary schooling. We were particularly interested in comparing

J Exp Child Psychol. Author manuscript; available in PMC 2018 May 01.
 Green et al. Page 12

FR to other cognitive precursors (verbal and spatial skills) that have been previously linked
Author Manuscript

to math development (e.g. McGrew & Wendling, 2010). Most prior developmental math
research studies have been conducted in populations of school-aged children in the primary
grades, and did not incorporate measures of FR in their predictive models. Therefore, the
current research provides a necessary extension to the existing developmental math literature
by examining the role of FR and its relation to other pertinent cognitive precursors in
predicting future math achievement across a wide age range of children, providing a more
comprehensive model of math development.

To this end, we first created a latent factor score of FR from three psychometric tests
designed to measure FR (Matrix Reasoning, Concept Formation, & Analysis Synthesis)
using confirmatory factor analysis. Second, we compared the strength of the associations
between prior FR and prior math reasoning on later math achievement at T3 (measured by
Applied Problem Solving, Math Reasoning, and Math Fluency) using Structural Equation
Author Manuscript

Modeling. Results showed that across all three-time points, prior FR was the strongest
predictor of later math achievement at T3, after accounting for prior math reasoning and age.
Once we had determined that FR was a better predictor of later math achievement than prior
numerical reasoning, we compared the relative contribution of prior FR to other important
cognitive abilities associated with math, indexed by verbal reasoning (measured by
Vocabulary), and spatial skills (measured by Spatial Span and Block Design), to future Math
Achievement at T3. This model accounted for over 90% of the variance in Math
Achievement. In this model, FR was the strongest cognitive predictor of future Math
Achievement measured approximately 1.5 years later. Notably, spatial skills, vocabulary, and
age were not significant predictors in this model.

Though some studies have shown that spatial skills and verbal comprehension are also
Author Manuscript

robust precursors to future math achievement (e.g. Li & Geary, 2013; LeFevre et al. 2010),
many prior studies have not incorporated measures of FR in their predictive models. Thus,
we interpret the current findings as support for the notion that FR is a foundational skill that
influences future development of numerical reasoning and potentiates math problem solving
skills. Thus, the findings indicate that FR should be incorporated into future developmental
models. These results support and extend Cattell's (1971; 1987) notion that FR development
is an important cognitive precursor for even the most basic math skill development,
including timed arithmetic, as well as more complex equations and word problems.

Study limitations
A limitation of the study is that we did not administer math measures at the first assessment
rendering us unable to control for the initial effect of these domain specific precursors on
Author Manuscript

future math outcomes. However, we were able to include math reasoning at T2 in our model,
which enabled us to compare the relative contribution of prior math reasoning to prior FR in
predicting future math achievement. Prior FR emerged as a better predictor of future math
achievement than prior math reasoning. This finding builds on prior studies showing that
domain general FR is as good a predictor of later math skills as prior numerical reasoning
skills (e.g. Fuchs et al., 2010). Another limitation of the study is the relatively small sample

J Exp Child Psychol. Author manuscript; available in PMC 2018 May 01.
 Green et al. Page 13

size. However, the results are statistically reliable, and the wide age range enables us to
Author Manuscript

make a novel contribution to the literature.

Theoretical implications
This work demonstrates that FR and mathematics achievement are linked throughout
development, and that FR supports mathematical thinking and reasoning throughout the
school years. One account for the strong relation between FR and math assessments is that
both engage a common underlying cognitive ability called relational reasoning, or the ability
to jointly consider multiple relations between different components of a problem (Halford,
Wilson & Phillips, 1998; Carpenter, Fennema, Franke, 2013; Miller Singley & Bunge, 2014;
Richland, Holyoak, Stigler, 2004; White, Alexander, Daugherty, 1998). The emerging ability
to reason relationally may form the foundation for mathematical conceptual development,
from the time children learn to compare the value of one number to another, to the time they
Author Manuscript

learn to extract the value of a fraction by comparing the value of the numerator to the value
of the denominator, to when they learn algebra and have to solve for an unknown variable by
keeping in mind the relationship between numbers on both sides of the equal sign, and so on
(Miller Singley & Bunge, 2014).

Demonstrating that FR predicts future math achievement across ages, above and beyond the
effects of age, math reasoning, and other cognitive factors correlated with math proficiency -
vocabulary and spatial skills – advances existing developmental theories of mathematics.
While many existing developmental theories were formed based on studies involving
younger children (approximately 4-9 years of age), (e.g. LeFevre et al., 2010), the current
sample spans a broader age range of 6-21 years. This work also replicates and extends the
findings in previous longitudinal research conducted by Fuchs et al. (2010) and Primi et al.
(2010), who found that FR was a robust cognitive predictor of future math achievement over
Author Manuscript

the course of one to two academic years in children in grades 1, 7, and 8 (respectively). Our
study included a wider age range of children and adolescents between 6 and 21 years old,
and adopted an analytic approach that enabled us to look at sequential influences of FR and
later math proficiency measured by three specific math achievement domains.

These findings expand upon an existing developmental framework proposed by LeFevre et

al. 2010, who hypothesize that there are three different pathways that contribute to early
math development in children 4 to 7 years old: prior linguistic (or verbal) skills, spatial
skills, and quantitative skills. In the current analyses, we have included these same pathways
as well as a fourth pathway, FR. Our findings indicate that FR is a robust pathway that may
be even more influential to math development than linguistic and spatial skills, though the
relative contribution of these predictors should be systematically compared in future
Author Manuscript

research.

Broader Implications
More generally, this line of research has possible relevance to school classroom settings.
Fluid reasoning is thought to support all forms of new learning for which an individual has
to problem solve (by integrating new information) without relying solely on prior
knowledge, and therefore fluid reasoning could be applicable to many subject areas.

J Exp Child Psychol. Author manuscript; available in PMC 2018 May 01.
 Green et al. Page 14

However, we posit that fluid reasoning is particularly helpful for learning math, which is
Author Manuscript

hierarchical in nature and requires individuals to solve novel problems as each new level
advances. Currently, educators often focus on building computational proficiency as a means
to improving mathematical achievement, without much consideration of the cognitive
precursors that underpin these skills or the students' strengths and weaknesses (Boaler,
1998). Theories such as CHC, as well as longitudinal studies such as this one provide
insights on the links between specific cognitive abilities and math achievement that can
inform educational practices. Though some new math curricula do incorporate spatial
rotation or block construction exercises, FR has not typically been emphasized in current
math curricula. However, even students with strong basic numerical skills and spatial skills
may not be proficient in applying logical reasoning techniques to solve novel problems.

We argue that math curriculum should incorporate opportunities for students to practice a
core aspect of FR known as relational thinking, or the ability to jointly consider several
Author Manuscript

relations among mental representations (Miller-Singley & Bunge, 2014). One example of a
curriculum that incorporates relational thinking practice into math exercises is called Early
Algebra (Carpenter, Franke & Levi, 2003). This curriculum involves teaching children as
young as 6 years old to view the equal sign as a form of equivalency using non-traditional
number sentences. For example, children solve equations such as “5+3 = 6 + _” and explain
their thinking aloud. By solving these types of equations and having students explain their
thinking, students come to understand the component relationship between numbers on
opposite sides of the equation, and can often identify the correct answer without doing any
calculations (Carpenter, Franke & Levi, 2003). Another effective approach involves
practicing early abstract reasoning skills with kindergartners and preschoolers to improve
early numeracy skills (Ciancio, Rojas, McMahon, & Pasnak, 2001; Kidd et al., 2008). Many
other approaches can be used to incorporate FR skill building opportunities into math
Author Manuscript

curriculum (e.g. Miller-Singley & Bunge, 2014).

Finally, the assessment of FR in elementary school could serve to identify students who are
likely to have difficulty-learning math. This information could help guide teachers to better
understand which interventions may be most fruitful for individual students at different
developmental levels of FR and math achievement skills.

Acknowledgments
This work was supported by a NIH/NINDS Grant R01 NS057146 to S.A.B. and E.F., and by a James S. McDonnell
Foundation Scholar Award to S.A.B. We thank Ori Ellis, Brian Johnson, Susanna Hill, Alexis Ellis, and all other
NORA study team members for invaluable assistance with data collection and management. We also thank Kirstie
Whitaker, Carter Wendelken, and Ariel Starr for their insightful guidance on data analyses or the manuscript. We
are especially grateful to the NORA study participants and their families for taking the time to complete the
Author Manuscript

extensive batteries of assessments over multiple years.

Appendix A
Frequencies and Percentages for Missing Data

Variables Frequency Percentage

Time 1

J Exp Child Psychol. Author manuscript; available in PMC 2018 May 01.
 Green et al. Page 15
Author Manuscript

Variables Frequency Percentage

Block design 1 1.4
Matrix 4 5.8
Analysis synthesis 35 50.7
Concept formation 2 2.9
Vocabulary 3 4.3
Digit span 38 55.1
Time 2
Block design 1 1.4
Matrix 0 .0
Analysis synthesis 1 1.4
Concept formation 0 .0
Author Manuscript

Vocabulary 10 14.5
Digit span 0 .0
Coding 16 23.2
Number series 0 .0

References
Achenbach, TM. Integrative guide for the 1991 CBCL/4-18, YSR, and TRF profiles. Burlington, VT:
University of Vermont, Department of Psychiatry; 1991.
Arbuckle JL. SPSS Amos 19.0 User's Guide. 2015
Adey P, Csapó B, Demetriou A, Hautamäki J, Shayer M. Can we be intelligent about intelligence?:
Why education needs the concept of plastic general ability. Educational Research Review. 2007;
2(2):75–97.
Author Manuscript

Bell RQ. Convergence: An accelerated longitudinal approach. Child Development. 1953; 24:145–152.
[PubMed: 13141335]
Bergman-Nutley S, Söderqvist S, Bryde S, Thorell LB, Humphreys K, Klingberg T. Gains in fluid
intelligence after training non-verbal reasoning in 4-year-old children: A controlled, randomized
study. Developmental Science. 2011; 14(3):591–601. DOI: 10.1111/j.1467-7687.2010.01022.x
[PubMed: 21477197]
Bisanz, J., Sherman, J., Rasmussen, C., Ho, E. Development of arithmetic skills and knowledge in
preschool children. In: Campbell, JID., editor. Handbook of mathematical cognition. New York,
NY: Psychology Press; 2005. p. 143-162.
Blair C. How similar are fluid cognition and general intelligence? A developmental neuroscience
perspective on fluid cognition as an aspect of human cognitive ability. Behavioral and Brain
Sciences. 2006; 29:109–125. [PubMed: 16606477]
Boaler J. Open and closed mathematics: student experiences. Journal of Research in Math Education.
1998; 29(1):41–62.
Cattell, RB. Abilities: Their structure, growth, and action. Boston, MA: Houghton-Mifflin; 1971.
Author Manuscript

Cattell, RB. Intelligence: Its structure, growth and action. Amsterdam: North-Holland: 1987.
Casey, BM., et al. A longitudinal analysis of early spatial skills compared to arithmetic and verbal
skills as predictors of fifth-grade girls' math reasoning. Learning and Individual Differences. 2015.
http://dx.doi.org/10.1016/j.lindif.2015.03.028
Carpenter TP, Fennema E, Franke ML. Cognitively guided instruction: a knowledge base for reform in
primary mathematics instruction. The Elementary School Journal. 2013; 97(1):3–20.
Carpenter, TP., Franke, ML., Levi, L. Thinking mathematically: integrating arithmetic and algebra in
elementary school. Portsmouth, NH: Heinemann; 2003.

J Exp Child Psychol. Author manuscript; available in PMC 2018 May 01.
 Green et al. Page 16

Cheng YL, Mix KS. Spatial training improves children's mathematics ability. Journal of Cognition and
Development. 2012; Advance online publication. doi: 10.1080/15248372.2012.725186
Author Manuscript

Ciancio DS, Rojas AC, McMahon K, Pasnak R. Teaching oddity and insertion to Head Start children:
An economical cognitive intervention. Journal of Applied Developmental Psychology. 2001;
22:603–621.
Clement, John J. Algebra word problem solutions: Thought processes underlying a common
misconception. Journal for Research in Mathematics Education. 1982; 13(1):16–30.
Desoete A, Grégoire J. Numerical competence in young children and in children with mathematics
learning disabilities. Learning and Individual Differences. 2007; 16:351–367.
Duncan GJ, Dowsett CJ, Claessens A, Magnuson K, Huston AC, Klebanov P, et al. Japel C. School
readiness and later achievement. Developmental Psychology. 2007; 43(6):1428–1446. DOI:
10.1037/0012-1649.43.6.1428 [PubMed: 18020822]
Ferrer E, McArdle JJ. An experimental analysis of dynamic hypotheses about cognitive abilities and
achievement from childhood to early adulthood. Developmental Psychology. 2004; 40:935–952.
[PubMed: 15535749]
Floyd RG, McGrew KS, Evans JJ. The Relative Contributions of the Cattell-Horn-Carroll Cognitive
Author Manuscript

Abilities in Explaining Writing Achievement during Childhood and Adolescence. Psychology in

the Schools. 2008; 45:132–144. http://dx.doi.org/10.1002/pits.20284.
Fuchs LS, Fuchs D, Compton DL, Powell SR, Seethaler PM, Capizzi AM, Schatschneider C, Fletcher
JM. The cognitive correlates of third-grade skill in arithmetic, algorithmic computation, and
arithmetic word problems. Journal of Educational Psychology. 2006; 98:29–43.
Fuchs LS, Geary DC, Compton DL, Fuchs D, Hamlett CL, Seethaler PM, et al. Schatschneider C. Do
different types of school mathematics development depend on different constellations of numerical
versus general cognitive abilities? Developmental Psychology. 2010; 46(6):1731–1746. DOI:
10.1037/a0020662 [PubMed: 20822213]
Fuchs LS, Compton DL, Fuchs D, Powell SR, Schumacher RF, Hamlett CL, et al. Vukovic RK.
Contributions of domain-general cognitive resources and different forms of arithmetic
development to pre-algebraic knowledge. Developmental Psychology. 2012; 48(5):1315–1326.
DOI: 10.1037/a0027475 [PubMed: 22409764]
Fry AF, Hale S. Processing speed, working memory, and fluid intelligence. Psychological Science.
1996; 7:237–241.
Author Manuscript

Heathcote D. The role of visuo-spatial working memory in the mental addition of multi-digit addends.
Current Psychology of Cognition. 1994; 13:207–245.
Holyoak, KJ. Analogy and relational reasoning. In: Holyoak, KJ., Morrison, RG., editors. The Oxford
handbook of thinking and reasoning. New York: Oxford University Press; 2012. p. 234-259.
Halford GS, Wilson WH, Phillips W. Processing capacity defined by relational complexity:
Implications for comparative, developmental and cognitive psychology. Behavioral Brain
Sciences. 1998; 21(6):803–831. [PubMed: 10191879]
Keith TZ. Effects of general and specific abilities on student achievement: Similarities and differences
across ethnic groups. School Psychology Quarterly. 1999; 14:239–262.
Keith TZ, Reynolds MR. Cattell–Horn–Carroll abilities and cognitive tests: What we've learned from
20 years of research. Psychology in the Schools. 2010; 47(7):635–650.
Kidd JK, Pasnak R, Gadzichowski M, Ferral-Like M, Gallington D. Enhancing early numeracy by
promoting the abstract thought involved in the oddity principle, seriation, and conservation.
Journal of Advanced Academics. 2008; 19(2):164–200.
Author Manuscript

Kline, RB. Principles and practice of structural equation modeling. 2nd. New York, NY: Guilford
Press; 2005.
Klingberg T. Development of a superior frontal-intraparietal network for visuo-spatial working
memory. Neuropsychologia. 2006; 44(11):2171–2177. DOI: 10.1016/j.neuropsychologia.
2005.11.019 [PubMed: 16405923]
Krawczyk DC. The cognition and neuroscience of relational reasoning. Brain Res. 2012; 1428:13–23.
DOI: 10.1016/j.brainres.2010.11.080 [PubMed: 21129363]

J Exp Child Psychol. Author manuscript; available in PMC 2018 May 01.
 Green et al. Page 17

LeFevre JA, Fast L, Skwarchuk SL, Smith-Chant BL, Bisanz J, Kamawar D, Penner Wilger M.
Pathways to mathematics: Longitudinal predictors of performance. Child Development. 2010;
Author Manuscript

81:1753–1767. DOI: 10.1111/j.1467-8624.2010.01508.x [PubMed: 21077862]

Li Y, Geary DC. Developmental gains in visuospatial memory predict gains in mathematics
achievement. PLoS ONE. 2013; 8:e70160.doi: 10.1371/journal.pone.0070160 [PubMed:
23936154]
Mackey AP, Hill SS, Stone SI, Bunge SA. Differential effects of reasoning and speed training in
children. Developmental Science. 2011; 14:582–590. [PubMed: 21477196]
McArdle JJ, Ferrer-Caja E, Hamagami F, Woodcock RW. Comparative longitudinal structural analyses
of the growth and decline of multiple intellectual abilities over the life span. Developmental
Psychology. 2002; 38(1):115–142. [PubMed: 11806695]
McGrew KS, Flanagan DP, Keith TZ, Vanderwood M. Beyond g: The impact of Gf-Gc specific
cognitive abilities research on the future use and interpretations intelligence test batteries in the
schools. School Psychology Review. 1997; 26:189–210.
McGrew KS, Wendling BJ. Cattell-Horn-Carroll cognitive-achievement relations: What we have
learned from the past 20 years of research? Psychology in the Schools. 2010; 47:651–675.
Author Manuscript

Miller Singley AT, Bunge SA. Neurodevelopment of relational reasoning: Implications for
mathematical pedagogy. Trends in Neuroscience and Education. 2014; 3:33–37.
Miyake A, Friedman NP, Rettinger DA, Shah P, Hegarty M. How are visuospatial working memory,
executive functioning, and spatial abilities related? Alatent-variable analysis. Journal of
Experimental Psychology General. 2001; 130:621–640. [PubMed: 11757872]
Moses, RP., Cobb, CE, Jr. Radical equations: Civil rights from Mississippi to the algebra project.
Boston, MA: Beacon Press; 2001.
National Governors Association and Council of Chief State School Offices. Common Core State
Standards. 2014. Retrieved from http://www.corestandards.org/
Primi R, Ferrão ME, Almeida LS. Fluid intelligence as a predictor of learning: A longitudinal
multilevel approach applied to math. Learning and Individual Differences. 2010; 20(5):446–451.
DOI: 10.1016/j.lindif.2010.05.001
Raven, JC. Progressive Matrix Reasoning: A perceptual test of intelligence. London: H. K. Lewis;
1938.
Author Manuscript

Richland LE, Holyoak KJ, Stigler JW. Analogy use in eighth-grade mathematics classrooms. Cognitive
Instruction. 2004; 22(1):37–60. http://dx.doi.org/10.1207/s1532690Xci2201.
Saxe GB, Taylor EV, McIntosh C, Gearhart M. Representing fractions with standard notation: A
developmental analysis. Journal for Research in Mathematics Education. 2005; 36(2):137–157.
Schrank, FA., Wendling, BJ. Woodcock-Johnson III Assessment Service Bulletin No 10. Rolling
Meadows, IL: Riverside Publishing; 2009. Educational interventions and accommodations related
to the Woodcock-Johnson® III Tests of Cognitive Abilities and the Woodcock-Johnson III
Diagnostic Supplement to the Tests of Cognitive Abilities.
Schneider, WJ., McGrew, K. The Cattell-Horn-Carroll model of intelligence. In: Flanagan, D.,
Harrison, P., editors. Contemporary Intellectual Assessment: Theories, Tests, and Issues. 3rd. New
York: Guilford; 2012. p. 99-144.
Tabachnick, BG., Fidell, LS. Using multivariate statistics. 5th. San Francisco, CA: Pearson, Inc; 2007.
Vanderwood ML, McGrew KS, Flanagan DP, Keith TZ. The contribution of general and specific
cognitive abilities to reading achievement. Learning and Individual Differences. 2002; 13:159–
188.
Author Manuscript

Vendetti MS, Bunge SA. Evolutionary and developmental changes in the lateral frontoparietal
network: a little goes a long way for higher-level cognition. Neuron. 2014; 84:906–917. [PubMed:
25475185]
Watts TW, Duncan GJ, Siegler RS, Davis-Kean PE. What's Past is Prologue: Relations Between Early
Mathematics Knowledge and High School Achievement. Education Research. 2014; 43(7):352–
360. DOI: 10.3102/0013189X14553660
White CS, Alexander PA, Daugherty M. The relationship between young children's analogical
reasoning and mathematical learning. Math Cognition. 1998; 4:103–23.

J Exp Child Psychol. Author manuscript; available in PMC 2018 May 01.
 Green et al. Page 18

Wechsler, D. Wechsler Abbreviated Scale of Intelligence. New York, NY: The Psychological
Corporation Harcourt Brace & Company; 1999.
Author Manuscript

Wechsler, D. Wechsler Intelligence Scale for Children–Fourth Edition. San Antonio, TX: The
Psychological Corporation; 2003a.
Woodcock, RW., McGrew, KS., Mather, N. Woodcock-Johnson III Tests of Cognitive Abilities. Itasca,
IL: Riverside Publishing; 2001.
Zhang X, Koponen T, Räsänen P, Aunola K, Larkkanen M, Nurmi J. Linguistic and spatial skills
predict early arithmetic development via counting sequence knowledge. Child Development. 2014;
85(3):1091–1107. http://dx.doi.org/10.1111/cdev.12173. [PubMed: 24148144]
Author Manuscript
Author Manuscript
Author Manuscript

J Exp Child Psychol. Author manuscript; available in PMC 2018 May 01.
 Green et al. Page 19

Highlights
Author Manuscript

The relative contribution of different cognitive abilities on future math achievement was
investigated in children 6 to 21 years old. Fluid Reasoning was the best predictor of later
math proficiency in multiple Math Achievement domains: math applied problem solving,
math reasoning, and math fluency 1.5 years later. This work demonstrates that Fluid
Reasoning and Mathematics Achievement are linked throughout development, and that
Fluid Reasoning supports mathematical thinking and reasoning throughout the school
years.
Author Manuscript
Author Manuscript
Author Manuscript

J Exp Child Psychol. Author manuscript; available in PMC 2018 May 01.
 Green et al. Page 20
Author Manuscript
Author Manuscript
Author Manuscript

Figure 1.
Standardized parameter estimates from the CFA of FR for each measurement occasion. All
three indicators loaded on significantly to the FR constructs at each time point.
Author Manuscript

J Exp Child Psychol. Author manuscript; available in PMC 2018 May 01.
 Green et al. Page 21
Author Manuscript
Author Manuscript
Author Manuscript

Figure 2.
Longitudinal models predicting Math Achievement from Prior FR and Math Reasoning. In
Model 1, Math achievement is predicted from previous assessments of FR and Math
Reasoning. In Model 2, Age is included at the two previous occasions. Circles represent
latent variables of FR comprised of three observed variables. One-headed arrows represent
regressions and two-headed arrows represent covariance or correlations.
Author Manuscript

J Exp Child Psychol. Author manuscript; available in PMC 2018 May 01.
 Green et al. Page 22
Author Manuscript
Author Manuscript

Figure 3.
Longitudinal model predicting Math Achievement from Prior FR, Spatial Skills, Vocabulary
and Age.
Author Manuscript
Author Manuscript

J Exp Child Psychol. Author manuscript; available in PMC 2018 May 01.
 Author Manuscript Author Manuscript Author Manuscript Author Manuscript

Table 1
Descriptive Statistics for the Study Variables (N = 69)

Time 1 Time 2 Time 3

Green et al.

Raw Score T-Score Raw Score T-Score Raw Score T-Score

Variables Range M SD M SD Range M SD M SD Range M SD M SD
Assessment age 5 to 16 10.18 3.32 -- -- 6 to 19 11.67 3.35 -- -- 8 to 21 13.45 3.85 -- --
Vocabulary 11 to 66 41.62 14.49 56.02 11.44 23 to 71 48.46 12.21 59.61 10.48 30 to 74 54.36 10.38 54.36 8.21
Matrix Reasoning 3 to 32 22.05 7.47 54.96 9.61 5 to 35 26.38 5.56 59.13 7.81 8 to 34 28.70 4.06 59.98 5.89
Analysis synthesis 1 to 37 23.60 8.35 60.90 13.17 12 to 34 27.41 4.56 65.43 12.16 20 to 34 29.49 3.22 68.64 12.26
Concept formation 1 to 35 23.63 10.03 58.51 10.42 7 to 35 28.49 7.32 64.39 10.67 7 to 35 31.69 4.86 69.37 8.56
Fluid reasoning 4 to 31 21.64 7.17 -- -- 11 to 39 32.31 5.91 -- -- 14 to 30 26.45 2.88 -- --
Number series -- -- -- -- -- 7 to 22 15.77 3.42 -- -- 11 to 23 17.43 2.81 -- --
Spatial Skills -- -- -- -- -- 21 to 83 56.28 17.81 -- -- 30 to 90 68.60 13.57 -- --
Applied problems -- -- -- -- -- -- -- -- 29 to 60 46.24 7.60 65.90 12.74
Math fluency -- -- -- -- -- -- -- 29 to 160 95.66 30.73 -- --

J Exp Child Psychol. Author manuscript; available in PMC 2018 May 01.
Page 23
 Author Manuscript Author Manuscript Author Manuscript Author Manuscript

Table 2
Pearson Correlations between the Study Variables

Variables 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
Green et al.

First assessment
1 Matrices
2 Analysis synthesis .68
3 Concept formation .71 .51
4 Vocabulary .81 .63 .69
5 Fluid reasoning .99 .72 .75 .84
Second assessment
6 Matrices .69 .77 .60 .77 .82
7 Analysis synthesis .60 .77 .59 .65 .75 .82
8 Concept formation .62 .60 .74 .73 .75 .76 .77
9 Spatial Skills .73 .68 .67 .80 .88 .77 .68 .78
10 Number series .68 .78 .55 .77 .80 .76 .77 .54 .75
11 Vocabulary .78 .66 .73 .87 .84 .77 .66 .64 .86 .74
12 Fluid reasoning .69 .79 .77 .77 .73 .94 .95 .88 .97 .82 .78
Third assessment
13 Matrices .60 .61 .50 .67 .76 .79 .71 .34 .64 .63 .58 .80
14 Analysis synthesis .55 .77 .48 .65 .70 .72 .75 .38 .65 .65 .62 .77 .65
15 Concept formation .57 .47 .63 .47 .62 .57 .54 .52 .53 .51 .49 .63 .52 .54
16 Spatial Skills .76 .75 .80 .88 .93 .89 .83 .87 .91 .83 .88 .96 .78 .78 .64
17 Number series .39 .53 .63 .55 .54 .57 .58 .48 .59 .59 .56 .65 .56 .71 .45 .73

J Exp Child Psychol. Author manuscript; available in PMC 2018 May 01.
18 Vocabulary .80 .65 .63 .82 .83 .73 .66 .56 .73 .64 .81 .75 .64 .56 .35 .81 .56
19 Applied problems .60 .66 .69 .79 .76 .72 .66 .54 .72 .74 .75 .77 .69 .74 .47 .86 .80 .71
20 Math fluency .67 .62 .65 .74 .77 .71 .67 .52 .86 .69 .81 .79 .61 .65 .37 .89 .60 .67 .71
21 Fluid reasoning .66 .76 .78 .72 .71 .83 .81 .78 .83 .72 .67 .87 .87 .91 .72 .90 .71 .72 .79 .70

Note. All correlations were statistically significant at .05.

Page 24
 Green et al. Page 25

Table 3
Fit statistics for the Structural Models Comparing FR to Math Reasoning in Predicting
Future Math Achievement
Author Manuscript

Variables χ2 df Δχ2/df R2
Model 1 Full (Figure 2 without Age) 61.02 19 .94
Model 1A (FRt1 → Matht3 = 0) 62.63 20 1.61/1 .90

Model 1B (FRt1 → Matht3 = 0) (FRt2 → Matht3 = 0) 85.03 21 22.40/1 *** .71

Model 1C (FRt1 → Matht3 = 0) (Matht2 → Matht3 = 0) 63.30 21 .67/1 .91

Model 2 Full (Figure 2 with Age) 139.71 31 .92

Model 2A (Aget1 → varst2 = 0) (Aget2 → varst3 = 0) 155.23 34 15.52/3 .90

*
p < .05.
**
p < .01.
Author Manuscript

***
p < .001.
Author Manuscript
Author Manuscript

J Exp Child Psychol. Author manuscript; available in PMC 2018 May 01.
 Green et al. Page 26

Table 4
Unstandardized and Standardized Path Coefficients for the Structural Models
Comparing prior FR to Math Reasoning in Predicting Future Math Achievement
Author Manuscript

Variables B SE β
Model 1C
FRt1 → FRt2 .77 .09 .98 ***
FRt1 → Matht2 .39 .05 .90 ***
FRt2 → Matht3 .99 .12 .96 ***
Model 2
FRt1 → FR2 .74 .09 .97 ***
FRt1 → Matht2 .34 .04 .86 ***
FRt1 → Matht3 -.10 .43 -.34
Author Manuscript

FRt2 → Matht3 .72 .49 .85

*
p < .05.
**
p < .01.
***
p < .001.
Author Manuscript
Author Manuscript

J Exp Child Psychol. Author manuscript; available in PMC 2018 May 01.
 Green et al. Page 27

Table 5
Fit Indices for the Structural Model Predicting Math Achievement at T3 from Prior FR,
Spatial Skills, Vocabulary and Age
Author Manuscript

Index Indirect Effects Direct Effect

Chi-square 67.68 65.69
Degrees of freedom 28 27
Probability level .00 .00
Normed chi-square 2.42 2.43
Goodness of fit index (GFI) .84 .85
Comparative fit index (CFI) .94 .94
Root mean square error of approximation (RMSEA) .14 .15
Lower bound 90% confidence interval .10 .10
Upper bound 90% confidence interval .19 .19
Author Manuscript

P-close .00 .00

Standardized root mean square residual (SRMR) .04 .04
Author Manuscript
Author Manuscript

J Exp Child Psychol. Author manuscript; available in PMC 2018 May 01.
 Green et al. Page 28

Table 6
Unstandardized and Standardized Path Coefficients for the Structural Model Predicting
Math Achievement at T3 (with Direct Effect from Age to Math Achievement)
Author Manuscript

Variables B SE β
Age to:

Vocabulary 2.96 .26 .81 ***

Spatial Skills 4.29 .45 .78 ***

Fluid Reasoning 1.37 .20 .71 ***

Math Achievement .32 .22 .16
Vocabulary →Math Achievement .08 .08 .15
Spatial Skills →Math Achievement .07 .05 .19

Fluid Reasoning → Math Achievement .52 .16 .52 ***

Author Manuscript

*
p < .05.
**
p < .01.
***
p < .001.
Author Manuscript
Author Manuscript

J Exp Child Psychol. Author manuscript; available in PMC 2018 May 01.
 Green et al. Page 29

Table 7
Standardized Direct and Indirect Effects of Age on Math Reasoning at Time 3
Author Manuscript

Effect Model w/o Direct Effect Model w/Direct Effect

Total effect .76 *** .80 ***

Direct effect -- .16

Indirect effect .76 *** .64 ***

*
p < .05.
**
p < .01.
***
p < .001.
Author Manuscript
Author Manuscript
Author Manuscript

J Exp Child Psychol. Author manuscript; available in PMC 2018 May 01.