LEAIND-01131; No of Pages 6

Learning and Individual Differences xxx (2015) xxx–xxx

Contents lists available at ScienceDirect

Learning and Individual Differences

journal homepage: www.elsevier.com/locate/lindif

Differences in the reading–mathematics relationship: A multi-grade,

multi-year statewide examination☆
Jill L. Adelson a,⁎, Emily R. Dickinson b, Brittany C. Cunningham c
a
Department of Counseling and Human Development, University of Louisville, United States
b
HumRRO, United States
c
Insight Policy Research, United States

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

Article history: In this study, we explored the relationship between mathematics and reading achievement using statewide as-
Received 20 February 2014 sessment data from all students (Grades 3–11) in multiple years to examine how that relationship differs
Received in revised form 16 June 2015 based on student, school, and district characteristics. In modeling this relationship, the researchers found that al-
Accepted 13 August 2015
though reading and mathematics ability explain a significant amount of variability in an individual year's
Available online xxxx
achievement, there is substantial variability remaining to be explained beyond that, particularly at the middle
Keywords:
school and high school levels. Thus, the need to look at individual characteristics is strongly warranted. The re-
Reading achievement sults indicated that reading and mathematics achievement were positively related at the student level and
Mathematics achievement more strongly at the school level, but the relationship was not as strong among females and non-White students.
State assessment © 2015 Elsevier Inc. All rights reserved.
Hierarchical linear modeling
Structural equation modeling

1. Introduction and high school) and across multiple years and were able to link stu-
dents, schools and districts, thus giving a very broad and comprehensive
Reading ability is key for both educational progress and achieve- examination of what individual as well as organizational characteristics
ment. However, approximately 40% of children in the U.S. struggle might be related to the connection between mathematics and reading.
with reading (Walsh, Glaser, & Wilcox, 2006). Reading difficulties at
young ages can affect adolescents and adults (Lundetrae, Gabrielsen, &
Mykletun, 2010) in both their educational attainment and achievement 1.1. The relationship between reading and mathematics achievement
level and their employment (Bynner & Parsons, 2001; Carnevale, 2001;
Kamil, 2003; Lundetrae et al., 2010; Rychen & Salganik, 2003; Snow & In the American education system, there exists a strong emphasis on
Biancarosa, 2003). Moreover, a student's literacy ability has important reading as literacy skills have been substantiated as critical for overall
consequences for achievement in other content areas, including mathe- academic success. Students' literacy has important consequences for
matics. Although the relationship between reading and mathematics achievement in other content areas, including the development of
has been well established (Babaresi, Katusic, Colligan, Weaver, & mathematics skills and achievement. Students who have difficulty in
Jacobsen, 2005; Fuchs et al., 2006; Hooper, Roberts, Sideris, Burchinal, reading have a high likelihood of experiencing difficulty in mathematics
& Zeisel, 2010; Jordan, Hanich, & Kaplan, 2003; Lewis & Mayer, 1987), (Babaresi et al., 2005). Researchers have found that these two content
very little is known about how that relationship may differ across sub- areas are related as early as preschool (Duncan et al., 2007;
groups. In this study, we examined patterns in the mathematics- McClelland et al., 2007; Purpura, Hume, Sims, & Lonigan, 2011; Welsh,
reading relationship and how individual student background character- Nix, Blair, Bierman, & Nelson, 2010). Studies have reported correlations
istics, school characteristics, and district characteristics moderated that between various reading and mathematics skills as moderate to high
relationship. We used statewide data across all assessed grades (3–8 (Fuchs et al., 2006; Hecht, Torgesen, Wagner, & Rashotte, 2001;
Purpura et al., 2011; Swanson & Beebe-Frankenberger, 2004). Reading
and mathematics skills have been documented as being related over
☆ This research was supported in part by a grant from the Collaborative Center for time, as early as kindergarten and as late as high school (Duncan et al.,
Literacy Development (CCLD). The findings and opinions expressed here do not 2007; Hooper et al., 2010). In Kentucky, researchers have found moder-
necessarily reflect the position or policies of CCLD.
⁎ Corresponding author at: Department of Counseling and Human Development,
ate, positive correlations between the state assessment reading and
University of Louisville, Louisville, KY 40292, United States. mathematics scores at the high school level (Bacci, Koger, Hoffman, &
E-mail address: jill.adelson@louisville.edu (J.L. Adelson). Thacker, 2003).

http://dx.doi.org/10.1016/j.lindif.2015.08.006
1041-6080/© 2015 Elsevier Inc. All rights reserved.

Please cite this article as: Adelson, J.L., et al., Differences in the reading–mathematics relationship: A multi-grade, multi-year statewide
examination, Learning and Individual Differences (2015), http://dx.doi.org/10.1016/j.lindif.2015.08.006
2 J.L. Adelson et al. / Learning and Individual Differences xxx (2015) xxx–xxx

Many studies have documented that skills in each content area influ- lunch decreased slightly over the grade levels, ranging from 55.5% (third
ence the development of skills in the others (Fuchs, Fuchs, & Prentice, grade) to 46.7% (high school). The percent of students with an individ-
2004; Jordan, Kaplan, Olah, & Locuniak, 2006). Specifically, in mathe- ual education plan (IEP) and identified as English Language Learners
matics the ability to decipher and comprehend language used in math- (ELL) also decreased over the grade levels, ranging from 9.5% to 15.3%
ematical problems connects to a student's ability to solve such problems for IEP and 1.0 to 2.5% for ELL.
(Lewis & Mayer, 1987). Students who fail to comprehend what a math- Just as the percent of students receiving free- or reduced-price lunch
ematics problem is asking them to do might fail to apply a computation- across the sample decreased slightly over the grade levels, so did the
al technique that they have otherwise mastered. The wording and school percent free- or reduced-price lunch, ranging from 57.0 to
structure of mathematics word problems affects students' ability to 65.4%. The percent of Title I schools ranged from 72.0 to 89.0%. The
comprehend and ultimately solve the problems (Clarkson & Williams, schools were primarily White, averaging 10.0 to 15.2% non-White.
1994). Perhaps as a result, students who experience both reading and Grades 3 through 5 all had school enrollments in that grade of about
mathematics difficulties often develop mathematics skills at slower 445 students, while grades 6 through 8 ranged from an average of 486
rates compared to students who experience difficulties in mathematics to 533 students and high school had about 886 students in a grade level.
only Jordan et al., 2003). The districts were primarily White, with an average of 8.7 to 13.0%
Potential contributions to the significant relationship between read- non-White. The districts had similar percentages of students receiving
ing and mathematics achievement have been attributed to genetics, free- or reduced-price lunch (61.3 to 62.4%). District enrollment at
general cognitive ability, and environmental influences (Hart, Petrill, each grade level ranged from 3934 to 4118, on average.
Thompson, & Plomin, 2009). However, the specific nature of the rela-
tionship between reading and mathematics across subgroups remains 2.2. Measures
unclear, and much is to be learned. The aspects we take into consider-
ation in the current study are several that have not been widely 2.2.1. Outcome
researched in the literature: the interaction effect of reading achieve- We examined the relationship between KCCT reading and mathe-
ment and student, school, and district characteristics on mathematics matics scores, which are reported on a scale ranging from 0 to 80 scales.
achievement within grade levels and across years. Table 1 presents the average achievement scores for the current and
previous year for both reading and mathematics. The KCCT was used
1.2. The current study for state and federal accountability purposes but is not used for
student-level decision making. It included both selected- and open-
In the current study, we examined moderators of the reading- response items, and was subject to several validation studies
mathematics relationship in Kentucky, considering individual student (e.g., Kentucky Department of Education (KDE), 2007; Sinclair,
background characteristics as well as school and district characteristics. Thacker, Koger, & Dickinson, 2008). The test scores have exhibited ade-
The Kentucky Department of Education (KDE) collects statewide data at quate reliability, with alphas ranging from .87 to .90 for reading and .86
the student, school, and district levels and maintains identifiers to .90 for mathematics for the 2007 KCCT (Kentucky Department of
connecting these three levels. Therefore, we were able to examine pat- Education (KDE), 2007).
terns across each tested grade level and across multiple years and could
examine cross-level relationships (i.e., the relationship between school 2.2.2. Moderating variables
and district variables and the interaction between student mathematics KDE provided all demographic data on each assessed student,
and reading achievement). The purpose of this study was to examine school, and district. Table 2 includes the moderator variables at each
the relationship between reading achievement and mathematics level and their coding.
achievement and, more importantly, examine the patterns of moderat-
ing effects on that relationship (i.e., what student, school, and district 2.3. Analyses
characteristics consistently moderate the relationships across multiple
grades and years). These data violated the assumption of independence of observations
that most traditional statistics make as students were nested within
2. Materials and methods schools and schools were nested within districts. Additionally, we
were interested in examining the interactions among variables that
2.1. Participants occur at multiple levels (i.e., student, school, and district level). As rec-
ommended by McCoach and Adelson (2010), we used hierarchical line-
We analyzed data provided by the KDE Office of Assessment and ar modeling (HLM) to account for the clustered data and appropriately
Accountability and for all students in public, non-alternative schools model the predictors at multiple levels.
who took the Kentucky Core Content Test (KCCT). Tested grade levels To test the relationship between reading and mathematics achieve-
were 3 through 8 for both subjects, 10 for reading, and 11 for mathemat- ment, mathematics achievement was the outcome variable. We includ-
ics. Because we were interested in patterns that might indicate individ- ed the prior years' mathematics achievement as a control variable and
ual differences, we looked for trends versus cohort effects using data included statistically significant student, school, and district characteris-
from 2007 to 2010. tics as predictors of mathematics achievement to appropriately model
Average sample sizes per year for elementary school (Grades 3–5), the intercept. Then, we added reading achievement scores to the
middle school (Grades 6–8), and high school (Grades 10 and11) were model to determine what proportion of the variability in mathematics
41,613, 40,981, and 41,731, respectively. There was an average of 635 achievement was explained by reading achievement, above and beyond
elementary schools, 314 middle schools, and 214 high schools per prior mathematics achievement and student, school, and district
year. Those schools were in an average of 159, 157, and 145 districts, characteristics. Then, to examine the moderators of that relationship,
respectively. we tested interactions with reading achievement to see if the variables
In the assessed grades, averaged across 2007 to 2010, Kentucky had at any of the three levels (e.g., gender, non-White, Title I school) moder-
nearly equivalent proportions of male and female students (48.7% to ated the relationship between reading and mathematics achievement.
49.2% female). The majority of students (82.5% to 86.8%) were White, In building our models, we used three-level HLMs, with students
with 10.0 to 10.7% of students identifying as Black, 2.0% to 3.2% identify- nested in schools nested in districts, and we used a general analytic
ing as Hispanic, 1.0% to 1.3% identifying as Asian, and 1.5% to 2.2% iden- strategy that followed the guidelines suggested by Raudenbush and
tifying as Other. The percent of students receiving free- or reduced-price Bryk (2002). We built the models separately for each grade level and

Please cite this article as: Adelson, J.L., et al., Differences in the reading–mathematics relationship: A multi-grade, multi-year statewide
examination, Learning and Individual Differences (2015), http://dx.doi.org/10.1016/j.lindif.2015.08.006
 J.L. Adelson et al. / Learning and Individual Differences xxx (2015) xxx–xxx 3

Table 1
Average current and prior year achievement scores (averaged from 2007 to 2010) for students.

Grade 3 Grade 4 Grade 5 Grade 6 Grade 7 Grade 8 High school

Reading scale score 53.5 52.4 50.4 49.9 47.5 48.0 45.1
Prior reading scale score⁎ NA 53.9 51.8 50.4 49.7 47.7 48.9
Mathematics scale score 53.1 51.3 47.9 46.1 45.1 42.1 37.7^
Prior mathematics scale score⁎ NA 52.9 50.5 46.7 45.9 44.6 43.7^
⁎ Averaged from 2008 to 2010 because 2006 data could not be linked to 2007 data and were not comparable.
^
Data available only for 2010.

each year. We entered the previous years' achievement scores, reading of about 46 students per school, and between 11.5% and 15.9% of the
achievement scores, and student-level demographic variables as well variability in the achievement measures was between schools (with
as the previous years' achievement scores at level 1, school-level vari- the majority being between students within schools). Because we
ables at level 2, and district-level variables at level 3. The intercept and used maximum likelihood estimation, students who were tested in
reading achievement slopes were allowed to vary across schools and Kentucky for some but not all of the measurements (i.e., had missing
districts. data) were still included in the model. For this model, 93% of students
had third-grade data, 95% had fourth-grade data, and 89% had fifth-
3. Results grade data. The model fit the data extremely well (CFI = 1.000, TLI =
0.999, RMSEA = 0.008, SRMRwithin = 0.003, SRMRbetween = 0.007).
3.1. Preliminary analyses Reading achievement and mathematics achievement were strongly cor-
related at the student level, r = .864 (p b .001). School-level reading and
Before conducting the HLM analyses, we first conducted preliminary mathematics achievement were even more strongly correlated, r =
multilevel structural equation modeling (ML-SEM) analyses to examine .932, p b .001). As shown in Table 3, the amount of variation explained
the extent to which reading and mathematics achievement are related. in each achievement measurement (R2) ranged from 60% to 77% at
To account for measurement error in the achievement tests, we created the student level and from 56% to 86% at the school level, indicating
two latent constructs – reading achievement and mathematics achieve- that there is substantial variability in achievement left to be explained
ment, each of which was measured by three years of achievement beyond the construct of achievement.
scores in the content area. We conducted this analysis with students At the middle school level, there were 36,440 students who
who were in third grade in 2008, fourth grade in 2009, and fifth grade remained in the same school during the years they were tested in Ken-
in 2010 and in the same school each year they were tested in Kentucky. tucky. Those students were nested in 447 schools, with an average of
This allowed us to model elementary school achievement and account about 82 students per school, and between 12.0% and 14.6% of the var-
for school attended. Similarly, we conducted this analysis with students iability in the achievement measures were between schools (with the
who were in sixth grade in 2008, seventh grade in 2009, and eighth majority being between students within schools). Again, students test-
grade in 2010 and in the same school each year they were tested in ed in a Kentucky public school for some but not all three of the grade
Kentucky. This allowed us to model middle school achievement and ac- levels (i.e., had missing data) were included due to the use of maximum
count for school attended. As shown in Fig. 1, we allowed the errors likelihood estimation. For this model, 93% of students had sixth-grade
within a year to correlate, reflecting that those measures were obtained
at the same time and may have been affected by similar influences out-
side student achievement/ability.
At the elementary school level, there were 35,245 students who
remained in the same school during the years they were tested in
Kentucky. Those students were nested in 771 schools, with an average

Table 2
Moderating variables at each level.

Level Variable Coding

Student Gender 0 = male, 1 = female

Race/ethnicity Reference group = White
Free/reduced-price lunch status 0 = not eligible, 1 = eligible
IEP status 0 = no IEP, 1 = IEP
ELL status 0 = not identified as ELL, 1 =
identified as ELL
Prior achievement⁎ Grand-mean centered
School School enrollment size Grand-mean centered
Percent non-White students Grand-mean centered
Percent of students receiving Grand-mean centered
free/reduced-price lunch
Title I status 0 = not Title I, 1 = Title I
School-mean prior achievement⁎ Grand-mean centered
District Enrollment size Grand-mean centered
Percent non-White students Grand-mean centered
Percent of students receiving Grand-mean centered
free/reduced-price lunch
Region Dummy coded
District-mean prior achievement⁎ Grand-mean centered
⁎ We were unable to use prior achievement data for 2007 because KDE did not use Fig. 1. Model to examine the relationship between reading and mathematics achievement
consistent identifiers to link student data from 2006 to later years given significant chang- in elementary students (year 1 = Grade 3, 2008; year 2 = Grade 4, 2009; year 3 = Grade
es in the assessment. Additionally, we did not have prior achievement data for Grade 3 be- 5, 2010) and in middle school students (year 1 = Grade 6, 2008; years 2 = Grade 7, 2009;
cause it was the first grade level tested. year 3 = Grade 8, 2010).

Please cite this article as: Adelson, J.L., et al., Differences in the reading–mathematics relationship: A multi-grade, multi-year statewide
examination, Learning and Individual Differences (2015), http://dx.doi.org/10.1016/j.lindif.2015.08.006
4 J.L. Adelson et al. / Learning and Individual Differences xxx (2015) xxx–xxx

Table 3 proportion of variance explained ranged from 37% to 69%, and the
Proportion of variance explained (R2) in achievement measures by student achievement between-district proportion of variance explained ranged from 60%
construct in elementary school and middle school models.
to 84%.
Level Observed variable Elementary school Middle school Reading achievement scores were next entered into the model as
R2 S.E. p R2 S.E. p predictors of mathematics achievement. Because reading and mathe-
matics are not tested in the same grade at the high school level, prior
Within Reading Year 1 .598 0.006 b.001 .690 0.006 b.001
Reading Year 2 .741 0.005 b.001 .745 0.006 b.001 year reading achievement was entered into the model for Grade 11.
Reading Year 3 .626 0.006 b.001 .722 0.006 b.001 This also established temporal precedence, with reading achievement
Mathematics Year 1 .674 0.006 b.001 .751 0.006 b.001 predicting mathematics achievement. Table 4 presents the coefficients
Mathematics Year 2 .772 0.005 b.001 .834 0.005 b.001 and proportions of variance explained at each level by the reading
Mathematics Year 3 .735 0.006 b.001 .793 0.006 b.001
Between Reading Year 1 .674 0.040 b.001 .807 0.035 b.001
model.
Reading Year 2 .852 0.031 b.001 .942 0.018 b.001 As shown in Table 4, reading and mathematics achievement are con-
Reading Year 3 .628 0.036 b.001 .878 0.029 b.001 sistently positive and statistically significantly related. As a student's
Mathematics Year 1 .557 0.044 b.001 .824 0.034 b.001 reading achievement increases one unit, their mathematics achieve-
Mathematics Year 2 .862 0.037 b.001 .955 0.019 b.001
ment is expected to increase, on average, between 0.39 and 0.46 units,
Mathematics Year 3 .648 0.041 b.001 .868 0.028 b.001
across grade levels, after controlling for student-, school-, and district-
Note. For the elementary and middle school models, respectively, Year 1 = Grades 3 and 6,
level variables. Above and beyond the student, school, and district char-
Year 2 = Grades 4 and 7, and Year 3 = Grades 5 and 8.
acteristics, adding reading to the model explains between 12% and 45%
of the variance in mathematics achievement between students, with a
data, 93% had seventh-grade data, and 75% had eighth-grade data great deal being explained in Grade 4 and a moderate amount being ex-
(possibly due to some schools only having grades six and seven housed plained at other grade levels; between 17% and 57% of variance between
within the middle school). The model fit the data very well (CFI = 0.999, schools, with a great deal being explained at the elementary school level
TLI = 0.997, RMSEA = 0.016, SRMRwithin = 0.005, SRMRbetween = 0.015). and the least amount being explained at the Grade 8 and high school
Reading achievement and mathematics achievement were strongly cor- levels; and between 0% and 40% of variance between districts.
related at the student level, r = .861(p b .001). Again, school-level read- What is of greater interest for this study was whether or not there
ing and mathematics achievement were even more strongly correlated, were differences in the relationship between prior reading achievement
r = .907 (p b .001). As shown in Table 3, the amount of variation ex- and mathematics achievement among the various student subgroups.
plained in each achievement measurement (R2) ranged from 69% to Table 5 presents the coefficients for the interactions between reading
83% at the student level and from 81% to 96% at the school level, with and gender, non-White, and free/reduced-price lunch status. With the
the greatest amount of variability being explained in seventh-grade exception of Grade 4, the interaction between gender and reading is
achievement. These proportions indicate that there is variability in statistically significant. Across all the grade levels, the coefficients
achievement left to be explained beyond the construct of achievement are negative, indicating that the relationship between prior reading
in the middle grades, particularly at the student level, although not as achievement and mathematics achievement is lower for females than
much as at the elementary school level. for males. The interaction between free/reduced-price lunch status
and reading is not consistently statistically significant at any of the
3.2. Differences in the relationship between reading and mathematics grade levels (with the exception of high school in which only one year
achievement is reported), nor is there consistency in the direction of the relationship.
Though not consistently statistically significant across all grade levels,
Given the strong relationship between reading and mathematics the interaction between non-White and reading is consistently negative
achievement shown in the preliminary analyses, we next used HLM to (with the exception of Grade 5 in 2008 and Grade 7 in 2009). With the
conduct cross-sectional analyses of the data to examine differences in exception of Grade 7 in 2009, the non-White/reading interaction is sta-
the relationship between reading and mathematics achievement. First, tistically significant across the middle school grades. The negative coef-
we analyzed a null model with no predictors to calculate intraclass cor- ficients can be interpreted as the relationship between prior reading
relations (ICCs). Similar to the models with reading achievement as the achievement and mathematics achievement being smaller for non-
outcome, the unconditional model revealed that across all grade levels, White students than for White students.
the majority of variance in student mathematics achievement was be- School- and district-level variables were subsequently added into
tween students within schools (87%–92%), while an average of 8% to the model as predictors of the relationship between prior reading
11% of variance was between schools within districts and between 2% achievement and mathematics achievement. None of these variables
and 3% was between districts. (Note that as with the reading achieve- had consistently statistically significant relationships with the reading-
ment models, because several districts had only one high school and on-mathematics slope, and the majority of them were not statistically
there was not statistically significant variability in achievement at the
district level, Grade 10 was modeled as a 2-level model. Even so, 92% Table 4
of variability was between students and 8% was between schools.) Average coefficient for the reading slope (predicting mathematics achievement) and the
These ICCs were consistent across all four waves of data with standard proportion of variance in mathematics explained by reading above and beyond control
deviations around the average grade level ICC being between 0.002 variables.

and 0.011. Proportion of variance explained

Next, in a series of models, we added student-level predictors, Coefficient Between Between Between
then school-level predictors, and finally district-level predictors of students schools districts
mathematics achievement, removing those that were not statistically
Mean SD Mean SD Mean SD Mean SD
significant at each step. This provided a model controlling for those
Grade 4 0.46 0.04 0.45 0.41 0.57 0.29 0.00 1.03
characteristics. Similar to the ranges of values observed for reading
Grade 5 0.42 0.02 0.17 0.01 0.54 0.35 0.40 0.16
achievement, the proportion of between-student variance was approx- Grade 6 0.39 0.01 0.18 0.00 0.31 0.04 0.24 0.01
imately 87% to 92%, between-school variance was 8% to 11%, and Grade 7 0.42 0.01 0.17 0.01 0.25 0.07 0.00 5.49
between-district variance was between 2% and 3%. The proportion of Grade 8 0.39 0.02 0.14 0.01 0.17 0.10 0.27 0.16
between-student variance in mathematics achievement explained Grade 11⁎ 0.42 NA 0.12 NA 0.17 NA NA NA

by the control variables ranged from 53% to 66%. The between-school ⁎ Grade 11 mathematics is predicted from Grade 10 reading.

Please cite this article as: Adelson, J.L., et al., Differences in the reading–mathematics relationship: A multi-grade, multi-year statewide
examination, Learning and Individual Differences (2015), http://dx.doi.org/10.1016/j.lindif.2015.08.006
 J.L. Adelson et al. / Learning and Individual Differences xxx (2015) xxx–xxx 5

Table 5
Student-level interactions predicting mathematics achievement.

Grade 4 Grade 5 Grade 6 Grade 7 Grade 8 High

2008 2009 2010 2008 2009 2010 2008 2009 2010 2008 2009 2010 2008 2009 2010 school
–0.014 –0.010 –0.012 –0.034 –0.039 –0.025 –0.038 –0.039 –0.042 –0.054 –0.026 –0.047 –0.045 –0.054 –0.033 –0.083
Female*RD

Free/reduced–
0.020 0.009 0.014 0.022 –0.003 0.016 0.007 0.008 0.002 –0.006 0.017 0.029 0.001 0.007 –0.022 –0.069
price lunch*
RD
Non–white* –0.006 –0.014 –0.027 0.002 –0.017 –0.021 –0.023 –0.028 –0.022 –0.034 0.006 –0.053 –0.034 –0.031 –0.046 –0.028
RD
Notes. Shaded cells reflect non-significant values (p N .05). RD = reading.
⁎High school is for 2010 only due to the availability of pretest data that was linked.

significant across grades and years. Also, for at least one year in each 4. Discussion
grade level (with the exception of Grade 11 in which only one year of
data was available), there was no statistically significant amount of In this study, we explored the relationship between reading and
between-district variance in the reading-on-mathematics slope to be mathematics achievement in Kentucky. After controlling for student,
explained. Therefore, we next discuss the proportion of variance in the school, and district characteristics related to mathematics achievement,
reading-on-mathematics slope explained by the student, school, and students' reading achievement had a positive relationship with their
district characteristics included in these analyses. mathematics achievement; students who scored highly on one test
Table 6 shows that essentially no between-student variance in the tended to score highly on the other. This is not surprising given prior re-
relationship between prior reading achievement and mathematics search results (e.g., Babaresi et al., 2005; Duncan et al., 2007; Fuchs et al.,
achievement is accounted for by the interactions between gender, 2006; Hecht et al., 2001; McClelland et al., 2007; Purpura et al., 2011;
free/reduced price lunch status, and non-White classification. These in- Swanson & Beebe-Frankenberger, 2004; Welsh et al., 2010). However,
teractions explain between 0.3% and 5% of the variance in the reading- what was interesting is that this effect appeared to be mitigated some-
on-mathematics slope between schools and between approximately what by a student's gender, as the relationship between prior reading
0% and 18% of between-district variance in the relationship between achievement and mathematics achievement was lower for females
prior reading achievement and mathematics achievement. Similarly, than for males. This relationship was also slightly lower for non-White
school and district characteristics, with a couple of exceptions (Grade students than for White students, but again the magnitude of the effect
6 and Grade 8 between-district variance), explain on average essentially was small. Student free/reduced price lunch status did not have a con-
none of the variance between schools and districts in the relationship sistent effect on the relationship between prior reading achievement
between reading achievement and mathematics achievement. and mathematics achievement, and school and district characteristics
added little or nothing to the explanation of variance between schools
and districts in the relationship between prior reading achievement
and mathematics achievement.
Females have traditionally experienced higher reading achievement
than their male counterparts (Chiu & McBride-Chang, 2006; Gambell &
Table 6 Hunter, 1999; Klein, 1977), and this was also found in this study of
Average Proportion of Variance in Reading-on-Mathematics Slope Explained by Student, achievement in Kentucky across multiple years. Thus, the fact that reading
School, and District Characteristics 2008–2010. achievement is more highly related to mathematics achievement for
Between student Between school Between district
males points to a further need to ensure students of both genders are
achieving at adequate levels in reading. Interestingly, although ethnic-
Mean SD Mean SD Mean SD
minority students tend to have lower reading achievement, this study
Student interactions found the relationship between reading and mathematics achievement
Grade 4 0 5.556 0.05 0.012 0.09 0.124
is not as strong for non-White students as it is for White students
Grade 5 0 0.005 0.05 0.042 0.05 0.037
Grade 6 0 0.000 0.04 0.018 0.03 0.087 in Kentucky. Thus, there may be some protective factors in place. For
Grade 7 0.01 0.007 0.05 0.034 0.18 0.220 instance, ELLs report higher academic self-concept in mathematics
Grade 8 0 0.006 0.05 0.050 0 0.632 in third grade (Niehaus & Adelson, 2013), which may serve as a protective
Grade 11 ⁎ ^
0 NA 0.003 NA NA NA factor in the relationship between reading and mathematics achievement.
School interactions Further, the study results show that although reading and mathe-
Grade 4 0.02 0.044 0 0.207 matics ability explain a significant amount of variability in an individual
Grade 5 0.03 0.024 0.02 0.058 year's achievement, there is substantial variability remaining to be ex-
Grade 6 0 0.027 0.31 0.072
Grade 7 0.04 0.049 0 0.435
plained beyond that, particularly at the middle school and high school
Grade 8 0 0.059 0 0.133 level. Some of this is a result of student and/or school characteristics
Grade 11⁎^ 0 NA NA NA on which the state regularly collects data (Adelson, Dickinson, &
District interactions
Cunningham, 2014). However, there are other factors that might be af-
Grade 4 0 1.200 fecting each year's achievement score. For instance, greater achieve-
Grade 5 0 3.170 ment motivation may affect achievement (Richardson & Abraham,
Grade 6 0 3.027 2009), and achievement motivation may vary from year to year. This
Grade 7 0 6.559
suggests a need for states to collect data beyond just achievement scores
Grade 8 0.34 0.550
Grade 11⁎^ NA NA and demographics. By administering some psychometrically sound,
theoretically chosen affective instruments throughout students' educa-
⁎ Because several districts had only one high school and there was not statistically sig-
nificant variability in achievement at the district level, Grade 11 was modeled as a 2-level
tional careers and collecting data on instructional practices and pro-
model. gramming, states can begin to look at other factors that are related to
^
Grade 11 based on 2010 only. student achievement.

Please cite this article as: Adelson, J.L., et al., Differences in the reading–mathematics relationship: A multi-grade, multi-year statewide
examination, Learning and Individual Differences (2015), http://dx.doi.org/10.1016/j.lindif.2015.08.006
6 J.L. Adelson et al. / Learning and Individual Differences xxx (2015) xxx–xxx

This study makes a unique contribution to the literature in that it ex- Fuchs, L. S., Fuchs, D., & Prentice, K. (2004). Responsiveness to mathematical problem-
solving instruction comparing students at risk of mathematics disability with and
amines patterns in the connection between reading and mathematics without risk of reading disability. Journal of Learning Disabilities, 37, 293–306.
achievement using a comprehensive statewide data set over several Gambell, T. J., & Hunter, D. M. (1999). Rethinking gender differences in literacy. Canadian
years. Because our goal was to get this broad, encompassing look at Journal of Education, 24, 1–16.
Hart, S. A., Petrill, S. A., Thompson, L. A., & Plomin, R. (2009). The ABC's of math: A genetic
the mathematics-reading relationship, the use of an existing statewide analysis of mathematics and its links with reading ability and general cognitive abil-
dataset limited the variables available for us to examine, and we were ity. Journal of Educational Psychology, 101, 388–402.
not able to examine other potential moderators, including other indi- Hecht, S. A., Torgesen, J. K., Wagner, R. K., & Rashotte, C. A. (2001). The relations between
phonological processing abilities and emerging individual differences in mathemati-
vidual differences between students. Additionally, using this dataset cal computation skills: A longitudinal study from second to fifth grades. Journal of
limited us to one measure of achievement; had we used other measures Experimental Child Psychology, 79, 192–227.
such as GPA, ACT scores, or other standardized test scores, we may have Hooper, S. R., Roberts, J., Sideris, J., Burchinal, M., & Zeisel, S. (2010). Longitudinal predic-
tors of reading and math trajectories through middle school for African American ver-
found different patterns of moderation due to the method effect in
sus Caucasian students across two samples. Developmental Psychology, 46, 1018–1029.
different scores (Dickinson & Adelson, 2013). Jordan, N. C., Hanich, B. L., & Kaplan, D. (2003). A longitudinal study of mathematical com-
Although this study is correlational and does not allow for causal in- petencies in children with specific mathematics difficulties versus children with co-
ference, the patterns found across multiple cohorts and multiple grades morbid mathematics and reading difficulties. Child Development, 74, 834–850.
Jordan, N. C., Kaplan, D., Olah, L. N., & Locuniak, M. N. (2006). Number sense growth in
do suggest some directions for research and practice. The study found a kindergarten: A longitudinal investigation of children at risk for mathematics difficul-
strong connection between reading achievement and mathematics ties. Child Development, 77, 153–175.
achievement, with individual differences in the strength of that rela- Kamil, M. L. (2003). Adolescents and literacy: Reading for the 21st century. Washington, DC:
Alliance for Excellent Education (Retrieved from http://web.all4ed.org/publications/
tionship. As such, reading strategies should be integrated into mathe- AdolescentsAndLiteracy.pdf).
matics classrooms. For students who have strong mathematics skills Kentucky Department of Education (KDE) (2007). Commonwealth accountability testing
and weak reading skills, this would allow them to address their weak- system: 2007 technical report.
Klein, H. (1977). Cross-cultural studies: What do they tell about f differences in reading?
ness through their strength. Additionally, given this relationship, The Reading Teacher, 30, 880–885.
by strengthening their reading achievement, they may also increase Lewis, A. B., & Mayer, R. E. (1987). Students' miscomprehension of relational statements
their ability to do well in mathematics. This study suggests this may in arithmetic word problems. Journal of Educational Psychology, 79(4), 363–371.
Lundetrae, K., Gabrielsen, E., & Mykletun, R. (2010). Do basic skills predict youth unem-
be even more important for some students than others, but all students
ployment (16 – to 24 year-olds) also when controlled for accomplished upper-
would benefit from a recognition of the connection between reading secondary schools? A cross-country comparison. Journal of Education and Work, 23,
and mathematics. 233–254.
McClelland, M. M., Cameron, C. E., Connor, C. M., Farris, C. L., Jewkes, A. M., & Morrison, F. J.
(2007). Links between behavioral regulation and preschoolers' literacy, vocabulary,
References and math skills. Developmental Psychology, 43, 947–959.
McCoach, D. B., & Adelson, J. L. (2010). Dealing with dependence (part I): Understanding
Adelson, J. L., Dickinson, E., & Cunningham, B. C. (2014, April). A multigrade, multiyear the effects of clustered data. Gifted Child Quarterly, 54, 152–155.
statewide examination of reading achievement. Philadelphia, PA: Paper presented at Niehaus, K., & Adelson, J. L. (2013). Self-concept and native language background: A study
the 2014 American Educational Research Association Annual Meeting and Exhibition. of measurement invariance and cross-group comparisons in third grade. Journal of
Babaresi, W. J., Katusic, S. K., Colligan, R. C., Weaver, A. L., & Jacobsen, S. J. (2005). Math Educational Psychology, 105, 226–240. http://dx.doi.org/10.1037/a0030556.
learning disorder: Incidence in a population-based birth cohort, 1976–1982, Roches- Purpura, D. J., Hume, L. E., Sims, D. M., & Lonigan, C. J. (2011). Early literacy and early nu-
ter, Minn. Ambulatory Pediatrics, 5, 281–289. meracy: The value of including early literacy skills in the prediction of numeracy de-
Bacci, E. D., Koger, M. E., Hoffman, R. G., & Thacker, A. A. (2003). Relationships among velopment. Journal of Experimental Child Psychology, 110, 647–658.
Kentucky's core content test, ACT scores, and students' self-reported high school grades Raudenbush, S. W., & Bryk, A. S. (2002). Hierarchical linear models: Applications and data
for the classes of 2000 through 2002 (FR-03–19). Alexandria, VA: Human Resources analysis methods (2nd ed.). Thousand Oaks, CA: Sage Publications.
Research Organization. Richardson, M., & Abraham, C. (2009). Conscientiousness and achievement motivation
Bynner, J., & Parsons, S. (2001). Qualifications, basic skills and accelerating social exclu- predict performance. European Journal of Personality, 23, 589–605.
sion. Journal of Education and Work, 14, 279–291. Rychen, D. S., & Salganik, L. H. (2003). Key competencies for successful life and a well-
Carnevale, A. P. (2001). Help wanted…college required. Washington, DC: Educational Test- functioning society. Gottingen: Hogrefe & Huber.
ing Service, Office for Public Leadership. Sinclair, A. L., Thacker, A. A., Koger, L. E., & Dickinson, E. R. (2008). Relations between
Chiu, M., & McBride-Chang, C. (2006). Gender, context, and reading: A comparison of stu- students' scores on the revised 2007 KCCT and the Prior KCCT. (FR-08–11). Alexandria,
dents in 43 countries. Scientific Studies of Reading, 10, 331–362. http://dx.doi.org/10. VA: Human Resources Research Organization.
1207/s1532799xssr1004_1. Snow, C. E., & Biancarosa, G. (2003). Adolescent literacy and the achievement gap: What do
Clarkson, S. P., & Williams, W. H. (1994). Are you assessing reading or mathematics? we know and where do we go from here? New York, NY: Carnegie Corporation.
Paper presented at the annual meeting of the American Mathematics Association of Swanson, H. L., & Beebe-Frankenberger, M. (2004). The relationship between working
Two-year Colleges, Tulsa, OK. Retrieved from ERIC database (ED393666). memory and mathematical problem solving in children at risk and not at risk for
Dickinson, E., R., & & Adelson, J. L. (2013, May). Using structural equation modeling to ex- serious math difficulties. Journal of Educational Psychology, 96, 471–491.
plore similarities and differences among measures of student achievement. San Francisco, Walsh, K., Glaser, D., & Wilcox, D. D. (2006). What education schools aren't teaching about
CA: Paper presented at the 2013 American Educational Research Association Annual reading and what elementary teachers aren't learning. National Council on Teacher
Meeting and Exhibition. Quality (Retrieved from http://www.nctq.org/nctq/publications/).
Duncan, G. J., Dowsett, C. J., Claessens, A., Magnuson, K., Huston, A. C., Klebanov, P., ... Japel, Welsh, J. A., Nix, R. L., Blair, C., Bierman, K. L., & Nelson, K. E. (2010). The development of
C. (2007). School readiness and later achievement. Developmental Psychology, 43, cognitive skills and gains in academic school readiness for children from low-income
1428–1446. families. Journal of Educational Psychology, 102, 43–53.
Fuchs, L. S., Fuchs, D., Compton, D. L., Powell, S. R., Seethaler, P. M., Capizzi, A. M., ...
Fletcher, J. M. (2006). The cognitive correlates of third-grade skill in arithmetic,
algorithmic computation, and arithmetic word problems. Journal of Educational
Psychology, 98, 29–43.

Please cite this article as: Adelson, J.L., et al., Differences in the reading–mathematics relationship: A multi-grade, multi-year statewide
examination, Learning and Individual Differences (2015), http://dx.doi.org/10.1016/j.lindif.2015.08.006