0% found this document useful (0 votes)
40 views26 pagesThe Relation Between Students' Math and Reading Ability
Diagnóstico
Uploaded by
edesehegoCopyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF or read online on Scribd
0% found this document useful (0 votes)
40 views26 pagesThe Relation Between Students' Math and Reading Ability
Diagnóstico
Uploaded by
edesehegoCopyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF or read online on Scribd
/ 26
ResearchGate
THE RELATION BETWEEN STUDENTS’ MATH AND READING ABILITY AND
THEIR MATHEMATICS, PHYSICS, AND CHEMISTRY EXAMINATION GRADES
IN SECONDARY EDUCATION
TERS CRATE
hie publestion ae aloo
and secondary chook ie poe HANKE KORPERSHOEK, HANS KUYPER and GREETJE VAN DER WERF
THE RELATION BETWEEN STUDENTS’ MATH AND READING
ABILITY AND THEIR MATHEMATICS, PHYSICS, AND
CHEMISTRY EXAMINATION GRADES IN SECONDARY
EDUCATION
Received: 19 September 2013; Accepted: 12 March 2014
ABSTRACT. Word problems are math- or science-related problems presented in the
context of a story or real-life scenario. Literature suggests that, to solve these problems,
advanced reading skills are required, in addition to content-related skills in, for example,
‘mathematics. In the present study, we investigated the relation between students” reading
ability and their achievements in advanced mathematics, physics, and chemistry when
controlling for their mathematical ability. The study included 1,446 Dutch secondary
school students who had taken their final examinations in these subjects. Using
multivariate multilevel models, we found that both math and reading ability (the latter
only in the pre-university track) were positively related to the examination grades on
‘mathematies, physics, and chemistry. This relation was equally strong for boys and girls
This relation was neither moderated by students” sex, nor by students’ ethnicity (Dutch
versus non-Dutch origin).
KEYWORDS: math ability, multivariate multilevel models, reading ability, secondary
education, word problems
Word problems (or “story-problems”) are math- or science-related
problems presented in the context of a story or real-life scenario (Adams,
2003, p. 790). In secondary school books, they are commonly used in
addition to more traditional problems that have to be solved by correctly
applying mathematical, physics, or chemistry formulae. They generally
contain textual instructions or a situational context for finding a solution
(Chapman, 2006; Lantz-Andersson, Linderoth & Séljé, 2009). Word
problems require students to move between two symbolic codes: written
language and mathematical symbols (Lantz-Andersson et al., 2009). Word
problems largely differ in their reading difficulty, for instance, in terms of
number of words used, word familiarity, or word difficulty. They require the
identification of relevant details and relationships between concepts (ie.
reading comprehension), which are often hidden within the language. To
solve a word problem, the students must derive the relevant information from
the text to set up a calculation problem for retrieving the missing information
(Fuchs, Fuchs, Fletcher, Hamlett & Lambert, 2008).
International Journal of Science and Mathematics Education 2014
© Ministry of Science and Technology, Taiwan 2014
Published online: 10 April 2014 HANKE KORPERSHOEK ET AL.
Evidently, advanced reading skills are needed to deal successfully with
a number of leaming tasks, such as solving contextualized problems,
interpreting written assignments, and following the oral or written
instructions given by the teacher. Contextualized problems are frequently
used in mathematics, physics, and chemistry examination tests in
secondary education. It is therefore plausible that some students do not
perform well on these tests because their reading skills inhibit their ability
to solve these contextualized problems, even though they have a good
understanding of course content itself (see also Reusser & Stebler, 1997).
In this paper, we examined the significance of students’ reading ability
for their academic achievement in mathematics, physics, and chemistry in
secondary education. As it is well-known that students’ achievement in
these subjects is partly predicted by their math ability, we included a
measure of students’ math ability as well. Using both ability measures, we
sought to understand to what extent students’ reading ability is related to
their academic achievement in mathematics, physics, and chemistry when
taking students’ math ability into account. Moreover, we examined
whether this relationship was moderated by gender and ethnic back-
ground (Dutch versus non-Dutch origin). To our knowledge, the impact
of students’ reading ability on mathematics, physics, and chemistry
understanding has not been studied extensively.
THeorETICAL FRAMEWORK
In this theoretical framework, we first explain the use of word problems
in mathematics- and science-related courses in secondary education in
more detail. Second, the relation between reading and math/science is
clarified. The focus in most studies has been on the relation between
reading and mathematics rather than reading and physics or chemistry;
therefore, the general focus of this paragraph is limited to the former
relation. Following the reported findings in the literature, we finally
present the aims of the present paper.
Word problems in mathematics- and science-related courses in secondary
education
There is an ongoing concern that a large part of the test items in math-
and science-related examinations assess students’ reading comprehension
rather than (or next to) their subject understanding (Flick & Lederman,
2002). Reading literacy is a broader conceptualization of reading
comprehension; it is generally conceived as the ability to understand STUDENTS’ MATH AND READING ABILITY AND EXAMINATION GRADES
and use written documents containing both verbal and pictorial
information, for example texts, pictures, charts, and tables (Baumert,
Lidtke, Trautwein & Brunner, 2009, p. 167) while mathematical literacy
is the ability to mathematize real-world situations and appropriately use
mathematics in problem contexts (OECD, 1999). The conceptualization
of mathematical modeling in mathematics, physics, and chemistry courses
goes far beyond the mere computation and application of formulae
(Baumert et al., 2009). In a similar vein, Bruner (1986) distinguishes two
modes of knowing—paradigmatic and narrative—as two irreducible
approaches to cognitive functioning when solving word problems.
Paradigmatic knowing requires a focus on universal and context-free
mathematical models or structures while narrative knowing demands an
emphasis on the social context of the problem. This social context could
involve a cover story of the word problem based on which (among other
issues) the storyline, situations, and relationships can be understood, The
textbooks of mathematics, physics, and chemistry courses in secondary
education often use these contextualized problems as a tool for
introducing the ‘real world’ into the classroom (see also Ogborn, 1996).
Particularly with respect to narrative knowing, sufficient reading skills are
required to understand the word problem presented. However, Chapman
(2006) reports that secondary school teachers generally emphasize
paradigmatic knowing. Accordingly, she suggests that incorporating both
modes of instruction would be more consistent with the current
perspectives on learning mathematics (see also Kurth, Kidd, Gardner &
Smith, 2002 on students’ use of narrative and paradigmatic forms of
dialogue in science conversations).
The reading-math/science relation
The fact that most math and science textbooks require reading skills has
led several scholars to investigate the relation between students’ reading
ability! and their achievements in these courses. Some researchers
evidently suggest that students’ reading ability is correlated with both
their general school achievements and those in mathematics (Ni Riordain
& O'Donoghue, 2009; Reikeris, 2006). Reikerds (2006), for example,
found that low achievement in reading slightly interfered with students’
development of arithmetic skills. Walker, Zhang & Surber (2008) believe
that mathematics items designed to measure higher cognitive skills, such
as problem solving and mathematical reasoning, are two-dimensional in
that they measure both reading ability and mathematics skills (pp. 163—
164). In their study (Walker et al., 2008), they found that students” scores HANKE KORPERSHOEK ET AL.
on these contextualized items were indeed influenced by their level of
reading ability. Furthermore, Grimm (2008) reports that early reading
comprehension (third grade) relates to a conceptual understanding of
mathematics and the application of mathematics knowledge.
We did find several studies on this topic (mostly conducted in
secondary education) which describe how students’ comprehension of
word problems can be improved (Borasj & Siegel, 1998; Brown & Ryoo,
2008; Carter & Dean, 2006; Helwig & Almond, 1999; Nathan, Kintsch &
Young, 1992). This focus suggests the assumption that students’ reading
ability is related to their understanding of math/science. Brown & Ryoo
(2008) demonstrated that students, who were taught via a “content-first”
approach facilitating the transition from an everyday understanding of
general phenomena to the use of scientific language, significantly
improved their understanding of science. As regards the text comprehen-
sion in mathematics tests, Helwig & Almond (1999) have suggested a
video format by which the questions are read by an actor on a video
monitor, a method specifically meant for students with an above average
mathematical understanding but low reading skills. Additionally, Nathan
et al. (1992) proposed using computer-based tutors to help students
improve their situational understanding of algebra word problems. Borasj
& Siegel (1998) give examples of how transactional reading strategies can
support mathematics instruction. They argue that unfamiliarity with the
vocabulary necessary to understand word problems can affect students’
performance. In the extensive math/science vocabulary, everyday words
such as ‘product’ and ‘volume’ take on new meanings. Several scholars
have suggested that teachers should explicitly teach their students this
vocabulary. Whether all these methods can successfully improve
students’ understanding of word problems in chemistry and physics
courses as well is as yet unclear. In recent years, increased attention has
been paid to the role of vocabulary and, for example, analogies in
teaching chemistry in secondary education (e.g. Bellocchi & Ritchie,
2011). Mathematics teachers generally do guide their students in
improving their reading comprehension of mathematics texts (Carter &
Dean, 2006). Notwithstanding the importance of this kind of teacher
support, however, Chapman (2006) reports that most teachers have little
experience in dealing with context in their teaching (i.e. the narrative
mode of knowing; see also Bell, Matkins & Gansneder, 2011 on the
impact of contextual instruction on teachers’ own understanding of
science),
All in all, we found multiple indications in the literature that students’
reading ability and their academic achievement in mathematics (and STUDENTS’ MATH AND READING ABILITY AND EXAMINATION GRADES
presumably also their achievement in physics and chemistry) are
somehow related.
The present study
Based on the literature review above, we conclude the following. First,
whereas the general relationship between prior school achievement and
future school results has been addressed by many scholars in the field,
research on the presumed connection between reading ability and math
ability and academic achievement in mathematics, physics, and chemistry
courses in secondary education has until now remained scarce. Most
studies on the relationship between students’ reading ability and their
achievements in mathematics have been conducted in primary education
(e.g. Kintsch & Greeno, 1985; Lee, Deaktor, Hart, Cuevas & Enders,
2005; Powell, Fuchs, Fuchs, Cirino & Fletcher, 2009; Stem, 1993;
Vilenius-Tuohimaa, Aunola & Nurmi, 2008; Walker et al., 2008).
Second, to our knowledge there are not many studies that relate
reading ability to academic achievement in physics and chemistry. The
studies we found generally deal with mathematics achievement. Similar to
the mathematics textbooks, the course materials used in physies and
chemistry also include contextualized (word) problems.” Only few
studies, however, have paid attention to the reading difficulty of physics
and chemistry textbooks (e.g. Jacobson, 1965; Koch & Eckstein, 1995;
Koch & Gunstone, 2001; Yore, 1991) or the text comprehension required
for biology texts (Ozuru, Dempsy & McNamara, 2009). An exception is a
publication by Fagan (1997), who suggests that there may be an
important connection between students’ reading ability and their achieve-
ments in advanced level physics at the end of secondary school.
Third, the studies on the relation between reading ability and math
achievement usually concentrate on students with reading deficiencies
(e.g. Kyttila, 2008) or English language learners in English-speaking
countries (e.g. Beal, Adams & Cohen, 2010; Lee, 2004; Shaw, Bunch &
Geaney, 2010; Stoddard, Pinal, Latzke & Canaday, 2002; Turkan & Liu,
2012), whereas reading ability is evidently important for the academic
achievement of all students. It is as yet unclear to what extent reading
ability and math ability relate to students’ academic achievement in
mathematics, physics, and chemistry in general
The current study addressed these three issues. The present study aims
at increasing our understanding of the possible significance of students’
math and reading ability for their academic achievements in mathematics,
physics, and chemistry in a large student sample in secondary education, HANKE KORPERSHOEK ET AL.
consisting of 1,446 students. Korpershoek, Kuyper, Van der Werf &
Bosker (2011) have presented correlations of 0.18 (senior general
secondary education) and 0.24 (pre-university education) between the
students’ math ability and their average examination grades on advanced
mathematics, physics, and chemistry examinations, which is why we used
math ability as covariate in this study. Tests were administered in the first
3 years of secondary education (seventh to ninth grade) to assess students’
reading and math ability, and students’ exam scores in mathematics,
physics, and chemistry were collected at the end of secondary education
(11™ or 12" grade). Including measures of both physics and chemistry
achievement in addition to math achievement is, to our knowledge, new
in the field. Furthermore, we examined whether the stated relationship
was moderated by gender and ethnic background (Dutch versus non-
Dutch origin). Contextualized education might be unfavorable to some
students, for example, to those for whom the language of instruction is
not their first language (e.g. Dutch for ethnic minority groups in The
Netherlands). Several studies have shown that English Language
Learers' reading proficiency is related to their mathematics achievement
in secondary education (e.g. Beal et al. 2010; Turkan & Liu, 2012).
Furthermore, the Programme for International Student Assessment
(PISA) 2006 study (Organisation for Economic Co-operation and
Development [OECD], 2007) has revealed that in all OECD countries
girls on average outperform boys in reading (see also Mullis, Martin
& Foy, 2008). Presumably, boys are generally slightly disadvantaged
compared with girls as regards the use of word problems in
contextualized education. Whether contextualized education is associ-
ated with underperformance among particular student groups is as yet
unknown. Based on the literature review, we formulated the
following hypotheses:
Hypothesis 1: Students’ reading ability is positively related to their
examination grades on the subjects advanced mathematics, physies, and
chemistry when controlling for their math ability.
Hypothesis 2: The relation proposed is stronger for students from ethnic
minority groups than for native Dutch students and stronger for boys than
for girls.
We did not have clear expectations for which of the three school
subjects investigated (advanced mathematics, physics, or chemistry) the
relation was the strongest. Therefore, no hypothesis was included on this
topic. The three school subjects were separately analysed to find out
which of the courses seemed to be the most ‘language-oriented.’ STUDENTS’ MATH AND READING ABILITY AND EXAMINATION GRADES
MetHop
Participants
The data used formed part of a large-scale longitudinal study conducted in The
Netherlands, the “Cohort Studies in Secondary Education” (VOCL °99). The
complete sample is representative for all schools and students in Dutch
secondary education (Van Berkel, 1999). For more information on the
VOCL’99 study, we refer to Korpershock, Kuyper & Van der Werf (2006).
The present study concentrated on the tracks that prepare students for higher
education; these are senior general secondary education (track B; duration
5 years), which prepares for higher professional education, and pre-university
education (track A; duration 6 years), which prepares for university. The main
difference between the two tracks is the difficulty level of the school subjects,
with track A being the most difficult track in Dutch secondary education. The
participants in this study started their educational career in secondary education
in 1999 (usually at age I] or 12 years) and took their Final School Examinations
(ESE) in the spring of 2004 (track B) or 2005 (track A).
We selected the students who had taken their FSE in advanced
mathematics, physics, and chemistry (overall sample, N= 1,752). Based
on the availability of prior reading ability and math ability scores and the
students’ examination grades, we could include 1,446 students (83 %) in
the analyses. This response group consisted of 931 track A students (from
108 different schools) and 515 track B students (from 71 different
schools). The socioeconomic status (SES) of the students (based on
parental education) was relatively high. Most parents of the students
included had pursued higher education (60 %), while 33 % had finished
secondary education without further study. Only 7 % of the parents had
attended primary education and/or a few years of secondary education
(low SES students). For some students, the ability measures and/or final
examination grades were unavailable due to the non-participation of
schools and/or drop out of individual students during the data collection
over time. Some schools, for example, prematurely ended their
participation in the cohort study or did not administer all requested tests.
‘A small number of students were not included in all waves of the data
collection because they declined to participate or had dropped out of
school. The “xesutts” section also contains the non-response analysis.
Variables and instruments
EXAMINATION GRADES. A\ the end of the 11" (track B) or 12" (track
A) grade, students in The Netherlands take part in the national HANKE KORPERSHOEK ET AL.
examinations (designed by specialists at the Dutch National Institute for
Educational Testing [Cito]) and school examinations (developed by the
schools themselves). The national examinations are of interest here,
because these te: re the same across all schools. The examinations in
track A are more complex than those for track B. For each course,
students obtain a grade between 1 (lowest grade) and 10 (highest grade)
(a 6 is a pass). The grades on the national tests in advanced mathematics,
physics, and chemistry among our student sample (retrieved from
Statistics Netherlands) varied from 2.5 to 9.9 and had a normal
distribution (skewness and kurtosis<1). For 164 track A students and
94 track B students (both 15 %), the final examination grades were
unavailable.
MATH ABILITY. The operationalization of math ability was based on
three math-related tests. The combination of these three tests yielded a
highly reliable measure of the students’ math ability. The first test was an
arithmetic test administered in the seventh grade, developed by Cito. The
reliability (a) of the test was 0.83. It included 20 multiple choice items
that covered the following topics: (1) numbers and calculations, (2)
proportions, fractions, and percentages, and (3) measuring, geometry,
time, and money. Second, an intelligence test was used (the Groninger
Intelligentietest voor Voortgezet Onderwijs [the Groningen Intelligence
Test for Secondary Education}, Van Dijk & Tellegen, 1994). This test,
assessed as reliable and valid (Evers, Van Vliet-Mulder & Groot, 2000),
was administered in the eighth grade. It consisted of a verbal and a
symbolic intelligence part, of which we used the scores on the symbolic
component (a for the symbolic intelligence part was 0.93). The symbolic
intelligence part covered two topics: numbers (25 items) and filling in
symbols (20 items). Third, we used a mathematics test for the ninth grade
(also developed by Cito) with a reliability (a) of 0.78. This test contained
33 multiple choice items, covering four topics: (1) arithmetic, measuring,
and estimation, (2) algebra/relations/graphs/functions, (3) geometry, and
(4) statistics and probabilities. In the overall student group of VOCL’99,
the correlations among the three tests were 0.42 (arithmetic-symbolic
intelligence), 0.50 (mathematics-symbolic intelligence), and 0.52 (arith-
metic—mathematics), respectively. A combined math ability score was
calculated for students who had completed at least two of these three
tests. This was done as follows. A factor analysis on the three
(standardized) test scores resulted in a one factor model with an
Eigenvalue of 1,963 (65 % common variance) and communalities of
0.64 (arithmetic), 0.71 (symbolic intelligence), and 0.62 (mathematics).
The factor loadings were 0.80 (arithmetic), 0.84 (symbolic intelligence), STUDENTS’ MATH AND READING ABILITY AND EXAMINATION GRADES
and 0.79 (mathematics), respectively. Based on these results and the
moderate correlations between the three tests, we concluded that the three
sts largely measured one common aspect of math ability but that they
also measured some other aspects of math ability (ie. 35 % unique
variance). Consequently, we used weighted coefficients rather than
unweighted coefficients to calculate students’ math ability, following this
procedure. To calculate the math ability scores for all students, we first
conducted a regression analysis for the full cases (i.e. that had valid scores
on all three tests). The resulting regression coefficients were then used in
regression equations to predict students’ math ability based on the
weighted test scores. For the students who had completed two out of three
tests, separate regression analyses were conducted for each combination,
each based on the weighted test scores of the test scores that were
available. Again, the resulting regression coefficients were used to predict
students’ math ability. By doing so, the relative weight of cach test that
the student had taken was taken into account when calculating students’
math ability. Finally, the scores were standardized again.
Throughout the overall student sample, the range of this (standardized)
math ability score ran from ~3.55 up to 4.44. Following recommendations
of Kamata, Turhan & Darandari (2003), we used stratified-alpha (as
proposed by Cronbach, Schénemann & McKie, 1965) to estimate the
reliability of the constructed variable, which was 0.92. The advantage of
using this combined score was that also students who had completed only
two of the three tests received a math ability score on the same scale as
the students who had completed all three tests, which strongly increased
the available sample size for the study. For 22 track A students (2 %) and
19 track B students (3 %), no math ability score was available.
READING ABILITY. The operationalization of reading ability was based
on three reading-related tests, following a procedure similar to that of
math ability. The first one was a Dutch language skills test administered
in the seventh grade, developed by Cito. This test of 20 items covered the
topics writing, spelling, text comprehension, and vocabulary. Its
reliability (@) was 0.74. Second, we used the students’ scores on the
verbal part of the intelligence test described above (a for the verbal
intelligence part was 0.92) which was administered in the eighth grade,
The verbal intelligence part covered three topics: synonyms (30 items),
analogies (25 items), and categories (30 items). Third, we employed a
Dutch comprehensive reading test (40 items) for the ninth grade (again
also developed by Cito) with a reliability of 0.75. The correlations among
the tests were 0.38 (language skills-verbal intelligence), 0.38 (compre-
hensive reading—verbal intelligence), and 0.29 (language skills-compre- HANKE KORPERSHOEK ET AL.
hensive reading). We calculated a combined reading ability score for
students who had completed at least two of these three tests. The factor
analysis on the three tests resulted in a one factor model with an
Eigenvalue of 1.719 (57 % common variance) and communalities of 0.54
(seventh-grade reading skills), 0.64 (verbal intelligence), and 0.53 (ninth-
grade reading skills). In general, they measured one common aspect of
reading ability, but the separate tests also measured other aspects of
reading ability (i.e. 43 % unique variance). Our ability measure is thus
based on a slightly broader range of aspects of reading ability (hence, the
higher unique variance of the separate tests and the lower correlations
among the tests) than was the case with the math ability measure
Throughout the overall student sample, the range of the (standardized)
reading ability score ran from —3.51 up to 4.55. The estimated reliability
(stratified-alpha) of the constructed variable was 0.83. For 40 track A
students (4%) and 29 track B students (5 %), no reading ability score was
available.
SEX. Data about student students’ gender were acquired by a question-
naire administered in the seventh grade. In our sample, 831 students were
boys (57 %) and 615 were girls (43 %).
ETHNICITY. The ethnic background of the students was determined by
questionnaire data collected among the seventh-grade students’ parents.
Based on the parents’ responses, the students were categorized as either
native Dutch students or as students from ethnic minority groups
(generally children with parents from non-Westem European countries).
The sample consisted of 1,286 native Dutch students (89 %) and 156
students from ethnic minority groups (11 %), which is representative of
the Dutch population, and four non-responders.
Analyses
We applied two three-level models (one for each track) containing the
students responses to our criterion variables nested within students, and
the students nested within the schools, to represent the hierarchical
structure of the data, The data were analyzed with multivariate multilevel
models, using MLwiN 2.23 software. Our criterion variables were the
examination grades on mathematics, physics, and chemistry for both
separate tracks. Next to unconditional models without predictors, several
models containing explanatory variables (math ability, reading ability,
sex, and ethnicity) were tested. Finally, we added two-way interaction
terms to the models between (1) math and reading ability, (2) the ability
measures and sex, and (3) the ability measures and ethnicity. STUDENTS’ MATH AND READING ABILITY AND EXAMINATION GRADES
Resutts
Non-response analysis
Table 1 gives an overview of the average scores of the response and the
non-response groups on the included ability and criterion variables
In track A, the response group scored on average somewhat higher on
reading ability than the non-response group, 1=—3.44, df 969, p<0.001
(Cohen’s d=0.56). This result also applied to the physics grades,
2.43, df 1,093, p<0.05 (d= 0.23), indicating that in the track A response
group was overrepresented by the higher performing students. The track
B response group scored somewhat lower on math ability than the non-
response group, f= 2.38, df 532, p<0.05 (d=—0.55), but other differences
between the two groups were not significant (p< 0.05).
Descriptive results
Table 2 presents the correlations among the variables included in the
study.
The highest correlations were established among the students’
examination grades (between 0.45 and 0.60). ‘There were moderate
associations between the math and reading ability scores (0.40 in track A
and 0.32 in track B) and small to moderate relations between both ability
scores and the students’ examination grades (correlations between 0.09
and 0.30). Furthermore, boys outperformed girls on math ability and
physics grades while girls outperformed boys on reading ability. Finally,
TABLE 1
Descriptive statistics response and non-response group (means and standard deviations)
Track A Track B
Response Non-response Response Non-response
Math ability 0.84 (0.86) 0.69 (1.12) 0.05 (0.83) 0.51 (0.86)
Reading ability 0,69 (0.92) 0.18 (0.85) -0.28 (0.84) -0.16 (0.79)
Examination grades*
Advanced mathematics 6.6 (1.3) 6.3 (1.4) 63 (1.1) 6.2 (Ll)
Physics 68 (1.2) 6.6 (1.4) 6.6 (1.0) 6.5 (0.9)
Chemistry 66(12) 64 (1.3) 6.4 (1.0) 6.2.(0.9)
N 931 189° SIS ur
“The grades range from I (lowest grade) to 10 (highest grade) (a 6 is a pass)
°N is the amount of students for which all the variables stated were missing, HANKE KORPERSHOEK ET AL.
100°0> dene "10'0> dee ‘SO0>4
(1 popoo sdnosB Sours omy wioyy siuspmis “9 papas swopms Ying] aAtvea) suOrE|aLI0D [eUDSI—IWIOd,
(1 popos sy “9 papo> s4oq) suorefauios yenDsiq-IuI0g,
I z0'0- L0°0- 80°0- 80°0- LE 0- 90°0- rena,
100 1 +010 sxs0T0 £00 +600 aah 1'0 2S
£0°0- ‘r0" 1 seaTSO, aeeSh'O «VO seeS10 Ansty
so'o- see TO- ¥409°0 1 wxx05°0 60° sxx8T'0 sorsattd
sonewayyeut
z0'°0- 90°0 wnt 0 409°0 1 010 w0x9T0 pooueapy
sxx 0- +100 sxx0f'0 wex870 44070 1 se760 Auge Beupeoy
so0- see L0- won ff0 wxx0£°0 waxtZ'0 wox0P'0 I Age eA,
£ 9 5 + £ z I
(s1}20 zamoy amp UF g_3pEN ‘S199 Jaddn om UE-Y yoRN) saqqeHTeA amp Waanaq suORTALI0)
Ta1av STUDENTS’ MATH AND READING ABILITY AND EXAMINATION GRADES
as regards students’ ethnic background, native Dutch students out-
performed students from ethnic minority groups on reading ability.
The multivariate multilevel models
Tables 3 (track A) and 4 (track B) show the results of the multivariate
multilevel models. Model 0 was the unconditional model without
predictors at the student level. Model 1 contained the predictor math
ability. Subsequently, we added students’ reading ability in Model 2.
Finally, in Model 3, we also included the students’ sex, ethnicity, and all
two-way interaction terms. The asterisks indicate that the ftest statistics
for the within-group regressions were significant.
After discussing the random parts of the models, we will subsequently
deal with the fixed parts in order to test our hypotheses.
As regards both tracks, the random parts show variance in the students”
examination grades at the student level and the school level. For example,
as regards track A, Model 1 for advanced mathematics indicates an
intraclass correlation coefficient (ICC) of 19.12/(19.12+151.09)=0.11.
This coefficient signifies that in this example 11 % of the variance in the
track A students’ examination grades on advanced mathematics was
found at the school level (i.e. the fraction of total variability that was
located at the school level), Tables 3 and 4 include all ICCs. The variance
at the school level for track A varies from 10 % to 12 % across the school
subjects. For track B, the school level variance varies from 4 % to 11 %,
Next, we investigated the explained variances at the student level.
For track A, for instance, Model 3 explained 8.0 % of the variance in
the students’ grades on advanced mathematics at the student level.
This outcome was calculated as follows. The student level variance
was 19.12+151.09=170.21 in Model 0 and 17.47+139.07= 156.54 in
Model 3. Thus, the explained variance at the student level was 1-
156.54/170.21 =0.080 (8.0 % reduction). These variances are included
in the tables. They vary from 2.3 % to 14.7 %, which is considered
quite low for achievement measures. The explained variances in the
track A models were, at least for the physics and chemistry grades,
evidently higher than those in the track B models. We will get back
to these findings in the “concLusions AND piscussion” section of this
paper.
As regards the fixed parts of the unconditional models, we observed
that the intercepts differed slightly among the school subjects for both
tracks, where the highest mean was established for physics, followed by
chemistry and advanced mathematics. The intercept of 64.98 for HANKE KORPERSHOEK ET AL.
TABLE 3
Parameter estimates (and standard errors) of the unconditional models and the models
containing explanatory variables, indicated separately for advanced mathematics, physics,
and chemistry (Track A; N=931)
‘Advanced mathematics
Fixed effects
Intercept
Math ability
Reading ability
Ethnicity
Reading ability sex
Reading ability ethnicity
Math ability sex
Math ability ethnicity
Sexxethnicity
Random effects
Suudent level variance
School level variance
Intraclass correlation
coefficient
Explained variance
student level
Physics
Fixed effects
Intereept®
Math ability
Reading ability
Sex
Ethnicity
Reading ability xsex
Reading ability xethnicity
Math ability sex
Math ability
