Content uploaded by Rhonda G Craven
Author content
All content in this area was uploaded by Rhonda G Craven on Jan 05, 2015
Content may be subject to copyright.
Earning Its Place as a Pan-Human Theory: Universality of the
Big-Fish-Little-Pond Effect Across 41 Culturally and
Economically Diverse Countries
Marjorie Seaton
University of Western Sydney
Herbert W. Marsh
Oxford University
Rhonda G. Craven
University of Western Sydney
For more than 2 decades, big-fish-little-pond effect (BFLPE) research has demonstrated that students in
high-ability classes and schools have lower academic self-concepts than their equally able counterparts
in mixed-ability schools. However, cross-cultural BFLPE research has been limited to mostly developed
and individualist countries. Using the Program for International Student Assessment database (Organi-
sation for Economic Co-operation and Development, 2005a, 2005b), the present investigation assessed
the BFLPE in 41 culturally and economically diverse countries. In support of the BFLPE, the effect of
school-average self-concept was negative for the total sample (effect size ⫽⫺.49), negative for each of
the 41 countries considered separately, and statistically significant in 38 countries. In this large, culturally
diverse sample of countries, the BFLPE was evident in both collectivist and individualist cultures and in
economically developing and developed nations. Implications for BFLPE theory and educational practice
are discussed.
Keywords: big-fish-little-pond effect, cross-cultural research, academic ability, academic achievement
Self-concept—a person’s sense of self, shaped through interac-
tion with the environment and other people (Shavelson, Hubner, &
Stanton, 1976)— has been studied across many areas of psychol-
ogy. It has been studied from a developmental perspective (e.g.,
Keller, Ford, & Meacham, 1978), in educational settings (e.g.,
Craven, Marsh, & Print, 2000; Marsh, Chessor, Craven, & Roche,
1995; Marsh & Hau, 2003; Marsh, Ko¨ller, & Baumert, 2001), and
in sport and exercise settings (e.g., Fox & Corbin, 1989; Marsh,
1997), and has been perceived as a central ingredient in the
development of personality (Marsh, Trautwein, Lu¨dtke, Ko¨ller, &
Baumert, 2006; Rogers, 1951). It is little wonder, then, that Marsh
and Craven (2006) described self-concept as a “hot variable that
makes good things happen, facilitating the realization of full hu-
man potential in a range of settings” (p. 134).
A positive self-concept is a crucial construct in many areas of
human functioning, ranging from maintaining good mental health
(e.g., Ha, Marsh, & Halse, 2004) to winning gold medals in
swimming (Marsh & Perry, 2005). In the academic domain, a
positive self-concept of one’s academic abilities (known as aca-
demic self-concept) has been associated with higher occupational
and educational aspirations, university attendance, course selec-
tion, and educational attainment levels (e.g., Guay, Larose, &
Boivin, 2004; Marsh, 1991; Marsh & Yeung, 1997b). Addition-
ally, self-concept and academic achievement have been shown to
be reciprocally related, with gains in one leading to gains in the
other (e.g., Guay, Marsh, & Boivin, 2003). These results imply that
promoting positive academic self-concepts in students is crucial to
maximizing educational outcomes and that educational policy
makers and practitioners should be encouraged to endorse the
promotion of positive academic self-concepts in order that students
derive optimal benefits from their education.
Notwithstanding this evidence, a growing body of research
suggests that the self-concepts of high-ability students suffer when
students are grouped on the basis of their academic ability (Craven
et al., 2000; Davis, 1966; Marsh, 1991, 2005; Marsh & Hau, 2003;
Marsh & Parker, 1984; Marsh, Trautwein, Lu¨dtke, Baumert, &
Ko¨ller, 2007). At the forefront of this research is the work of
Marsh and his colleagues, who have shown that grouping students
together in high-ability environments has a detrimental effect on
students’ perceptions of their academic abilities—a construct re-
ferred to in the literature as academic self-concept. Known as the
big-fish-little-pond effect (BFLPE), this body of research has
demonstrated that students who are educated in high-ability classes
and schools have lower academic self-concepts than equally able
students who are educated in mixed- or low-ability environments
(for an overview, see Marsh, 2007; Marsh et al., 2008). Moreover,
the BFLPE appears to be evident across cultures (Marsh & Hau,
2003; Zeidner & Schleyer, 1998). However, to date, the countries
examined in cross-cultural BFLPE studies have been mostly de-
Marjorie Seaton and Rhonda G. Craven, Centre for Educational Re-
search, University of Western Sydney, Penrith South DC, New South
Wales, Australia; Herbert W. Marsh, Department of Education, Oxford
University, Oxford, England.
Correspondence concerning this article should be addressed to Marjorie
Seaton, Centre for Educational Research, College of Arts, University of
Western Sydney, Bankstown Campus, Locked Bag 1797, Penrith South
DC, New South Wales 1797, Australia. E-mail: M.Seaton@uws.edu.au
Journal of Educational Psychology © 2009 American Psychological Association
2009, Vol. 101, No. 2, 403–419 0022-0663/09/$12.00 DOI: 10.1037/a0013838
403
veloped and individualist countries. Hence, the principal focus of
the present investigation was to assess the universality of the
BFLPE by ascertaining whether it extends to collectivist and less
developed nations.
Research Support for the Importance of a Positive
Self-Concept
A positive self-concept has been linked to many desirable edu-
cational outcomes, such as coursework selection, educational as-
pirations, subsequent university attendance, and academic achieve-
ment (e.g., Guay et al., 2003; Marsh, 1991; Marsh & Craven, 2005,
2006). For example, Marsh and Yeung (1997b) found that for high
school students (Grades 8 and 10), academic self-concept in spe-
cific school subjects was a better predictor of coursework selection
than the corresponding school grades in these subjects. Indeed,
school grades made no significant improvement in predictions
beyond what could be explained in terms of academic self-
concepts. Moreover, Guay et al. (2004) conducted a 10-year lon-
gitudinal study to test the influence of academic self-concept,
academic achievement, and family variables on educational attain-
ment level. They found that a positive academic self-concept was
associated with better educational outcomes 10 years later and
concluded that their findings provided “good support for the long-
lasting effects of academic self-concept” (p. 64).
It is important to note that academic self-concept has been
demonstrated to have a causal impact on academic performance.
Marsh and Craven (2006) reviewed the relation between academic
self-concept and performance outcomes and demonstrated that
academic self-concept and academic achievement are reciprocally
related, such that improvements in one lead to improvements in the
other. They termed this relation the reciprocal effects model
(REM). For instance, in a longitudinal study, Marsh and Yeung
(1997a) measured students’ academic self-concepts and academic
achievement in math, English, and science. They found that prior
academic achievement had a statistically significant positive effect
on subsequent academic self-concepts for all three subjects. Con-
trolling for individual ability, they found that prior self-concepts
also had a statistically significant positive effect on subsequent
achievement for math and, to a lesser extent, for science and
English. Hence, this study provided clear support for the REM.
Further support for the REM was found in a recent meta-analysis
of relevant research, which concluded that academic achievement
was positively related to academic self-concept and vice versa
(Valentine & DuBois, 2005). Thus, the best option to improve
performance may be to improve academic self-concept and aca-
demic achievement simultaneously (see Marsh & Craven, 2006).
The development and maintenance of a positive self-concept has
also been regarded as one of the key objectives of education. For
example, an aim of Australian education is to instill in students
“qualities of self-confidence, optimism, high self-esteem, and a
commitment to personal excellence” (Ministerial Council on Ed-
ucation, Employment, Training and Youth Affairs, 1999, Goals, ¶
1). Similarly, one of the main conclusions emanating from a recent
Organisation for Economic Co-operation and Development
(OECD) study was that student engagement, broadly defined as
“students’ attitudes towards schooling and their participation in
school activities . . . is closely tied to students’ economic success
and long-term health and wellbeing and as such deserves to be
treated alongside academic achievement as an important schooling
outcome” (OECD, 2003, p. 9). However, positive self-concepts are
not always achieved when students are segregated according to
ability. Rather, research has consistently demonstrated that stu-
dents in high-ability classes and schools have lower academic
self-concepts than their equally able counterparts in low-ability
classes and schools—the BFLPE—and lower academic self-
concepts have been associated with poorer educational outcomes
(e.g., Craven et al., 2000; Marsh, 1991; Marsh et al., 2008).
The Big-Fish-Little-Pond Effect (BFLPE)
The BFLPE model posits that although individual ability is
positively related to academic self-concept, a high school-average
ability is negatively associated with academic self-concept. It is
this negative effect of school-average ability that characterizes the
BFLPE. According to this model, one’s academic self-concept
partly depends on one’s own ability and on the ability of other
students in one’s school. This positive relation between individual
ability and academic self-concept and the negative relation be-
tween school-average ability and academic self-concept are de-
picted in Figure 1. Marsh and his colleagues have proposed that the
theoretical underpinnings of the BFLPE derive from research on
psychophysical judgment (e.g., Helson, 1964; Marsh, 1974; Par-
ducci, 1995; Wedell & Parducci, 2000), social judgment (e.g.,
Morse & Gergen, 1970; Upshaw, 1969), sociology (Alwin & Otto,
1977; Hyman, 1942; Meyer, 1970), the theory of relative depriva-
tion (Davis, 1966; Stouffer, Suchman, DeVinney, Star, & Wil-
liams, 1949), and social comparison theory (Festinger, 1954; see
Marsh et al., 2008, for a detailed account of the theoretical basis of
the BFLPE).
School/Class
Average Ability
Individual Ability Individual Student
Academic Self-Concept
++
++ _
Figure 1. The big-fish-little-pond effect (Marsh & Hau, 2003).
404 SEATON, MARSH, AND CRAVEN
In explaining the development of the BFLPE, Marsh and his
colleagues have suggested that students form their academic self-
concepts partly by comparing their academic performance with
that of their peers. In academically selective settings these peers
are other highly intelligent students, and so comparisons can lead
to students feeling less able, with a resulting lowering of academic
self-concepts. In nonselective academic environments where the
school-average ability level is lower, academic self-concepts tend
to be higher. So, for example, an average student in a high-ability
school may be performing academically below the average
achievement level for that school. That student may then compare
his or her academic ability with other students in that school, and
this may result in a lower academic self-concept. However, if this
same student were attending an average-achievement school, the
student may well perform above the average for that school, with
a resulting increase in academic self-concept (Marsh & Hau,
2003). In making these comparisons, the environment imposes a
comparison on students, as they will use the only frame of refer-
ence available to them to evaluate their ability— other students in
their school.
The preceding discussion suggests that the BFLPE is not just an
effect; it is a theory. Not only does the BFLPE describe how
students in high-ability schools feel about their academic compe-
tencies compared with similar-ability students in low-ability
schools, but it also explains the processes underpinning the effect,
and it predicts that students in high-ability classes and schools will
have lower academic self-concepts than their equally able coun-
terparts in low-ability schools. Moreover, the BFLPE is extremely
testable. The following section describes some of the research that
has been conducted to test this theory (see Marsh et al., 2008).
Empirical Support for the BFLPE
The BFLPE has found widespread support in a myriad of studies
(e.g., Marsh, 2005; Marsh & Craven, 2002; Marsh & Parker, 1984;
Marsh & Rowe, 1996; Marsh et al., 2007; Mulkey, Catsambis,
Steelman, & Crain, 2005; Zeidner & Schleyer, 1998; see also
Marsh et al., 2008). For example, Craven et al. (2000) demon-
strated that the academic self-concepts of students in special gifted
and talented classes showed a greater decline over time than those
of students in streamed or mixed-ability groups.
In relation to educational outcomes, research has demonstrated
that compared with their counterparts in low-ability schools, stu-
dents in high-ability schools tend to have lower educational and
occupational aspirations, are less likely to undertake higher levels
of study in English and math, are more likely to be on lower
academic tracks, and have lower grade point averages (Marsh,
1991). The BFLPE also appears to be persistent. Results of two
large longitudinal studies indicated that school-average achieve-
ment had a negative association with math self-concept for up to 4
years after graduation from high school (Marsh et al., 2007).
Whereas Coleman and Fults (1985) noted that the BFLPE was
more pronounced for students of lower ability in academically
selective schools, Reuman (1989) suggested that between-class
ability grouping results in higher academic self-concepts for stu-
dents of low ability and lower academic self-concepts for students
of high ability. Others have found that the BFLPE affects all levels
of ability (Marsh et al., 1995; Marsh & Hau, 2003). These con-
tradictions in the literature led Marsh and Craven (2002) to con-
clude that the BFLPE is reasonably invariant across ability levels,
as the Class- and School-Average Ability ⫻Individual Student
Ability interactions are usually small and not even consistent in
direction (see also Marsh, 2005; Marsh et al., 2008). Hence, the
BFLPE appears to be consistent across ability levels.
Generalizability of the BFLPE: Cross-Cultural
Comparisons
One means of testing the generalizability of research findings
and theories is to establish whether they can be supported in
another cultural setting. Matsumoto (2002) suggested that cross-
cultural research can inform us whether our psychological con-
structs can be universally applied. Marsh and Hau (2003) empha-
sized the purpose of cross-cultural research by highlighting its
three complementary goals as outlined by Segall, Lonner, and
Berry (1998): (a) testing psychological knowledge in another
cultural setting, (b) finding new dimensions of the construct under
investigation within different local cultures, and (c) incorporating
information from Goals 1 and 2 to form universal principles.
Consequently, if psychological constructs can be applied univer-
sally across different cultures, then their generalizability is greatly
strengthened.
An interesting experimental study by McFarland and Buehler
(1995, Study 3) provided evidence that the BFLPE may not be as
pronounced in collectivist countries. In their study, McFarland and
Buehler classified their participants (45 undergraduate students) as
having either an individualist cultural heritage or a collectivist
cultural heritage on the basis of Hofstede’s (1980) classification
system. After completing a bogus test, half of the participants were
told that their performance was high in relation to their group but
that their group as a whole had performed poorly. The other half
of the participants were told that they had performed poorly but
that their group had performed well. Those who possessed an
individualistic background reported more favorable reactions to
this performance feedback when they thought they were high
performers in an unsuccessful group than when they thought they
were poor performers in a successful group. The type of feedback
had no influence on those from a collectivist heritage. The authors
concluded that their findings demonstrated that the BFLPE was
reduced for those participants who hailed from collectivist back-
grounds, as they placed a higher value on their social groups and
were less likely to consider their own individual achievements
within the group than were their individualist counterparts.
Intuitive perspectives, too, would suggest that the BFLPE might be
reduced in collectivist countries. In collectivist countries the social
group to which one belongs is more important than the individual,
and the interests, rights, and responsibilities of the group take
precedence over those of the individual. Children in collectivist
societies grow up to revere and respect the members of their
extended family and to think of themselves as part of this group.
As Hofstede and Hofstede (2005) observed, the group to which
one belongs in a collectivist society (usually the extended family)
“is the major source of one’s identity” (p. 75). In individualist
countries the opposite tends to be true. In these countries the
interests, rights, and responsibilities of the individual take prece-
dence over those of the social group to which one belongs. Chil-
dren in individualist societies tend to grow up in nuclear families,
often do not know the members of their extended family, and learn
405
A PAN-HUMAN THEORY
to be independent from a young age (Hofstede & Hofstede, 2005).
People in individualist societies think of themselves as individuals,
and their identity is of their own making (Hofstede & Hofstede,
2005). We acknowledge, however, that there will be differences
between individuals within any given collectivist or individualist
culture. For example, whereas some individuals in a collectivist
culture may be more individualistic (termed idiocentric) in nature,
others in an individualistic culture may be more collective (termed
allocentric) in nature (see Chen, Wasti, & Triandis, 2007; Triandis,
Bontempo, Villareal, Asai, & Lucca, 1988). Nevertheless, in the
present investigation we consider the individualism– collectivism
dimension to “represent the general attributes of a given culture”
(Chen et al., 2007, p. 261) and thus to be a country-level variable.
Hence, we suspect that students in collectivist cultures may place
a higher value on the group’s achievements than on their own. If
this were the case, then students in high-ability schools in collec-
tivist cultures may consider the ability level of the school as an
asset and assimilate it into their self-concepts, thus eliminating the
BFLPE.
Perspectives regarding the BFLPE in developing countries are
less clear. For example, compared with developed countries, de-
veloping countries tend to have a lower standard of living, they
rely more on an agricultural base than an industrial base, poverty
is comparably higher, and levels of literacy, education, and life
expectancy are also lower. In developing countries, where educa-
tional standards are lower, there may be less opportunity for the
BFLPE to flourish, as there may be less emphasis on education,
and, perhaps, the academic differences between schools may be
smaller. Alternatively, although the education standards on aver-
age in these developing countries may be lower than in developed
countries, there may be large differences in the education of rich
and poor, and as such the education of the rich may well be more
reflective of that of developed countries. In this instance, a BFLPE
may well be evident in developing countries.
Although cross-cultural BFLPE research has been conducted, it
is limited due to its reliance on mainly developed countries with
highly individualist cultural orientations, and, hence, studies ex-
amining the BFLPE in economically developing or collectivist
countries are few. Most BFLPE research has been undertaken in
Australia (e.g., Craven et al., 2000; Marsh, 2004; Marsh & Parker,
1984; Marsh et al., 1995), the United States (e.g., Marsh, 1987,
1991; Mulkey et al., 2005), Germany (Marsh et al., 2001), Israel
(Zeidner & Schleyer, 1998), and the United Kingdom (Ireson,
Hallam, & Plewis, 2001; Tymms, 2001; see also Marsh & O’Mara,
2008b). In regard to the few BFLPE studies that have been
conducted with collectivist and developed countries, results have
been similar to those in developed and individualist countries. For
example, Marsh, Kong, and Hau (2000) examined the BFLPE in
Hong Kong, which is known for its collectivist culture. Consistent
with BFLPE theory and research, they found academic self-
concept was lower for students who attended high-ability schools
compared with equally able students attending lower ability
schools. However, although the cultural orientation of Hong Kong
is collectivist, it is highly industrialized with an education system
that is highly segregated on the basis of ability.
Support for the BFLPE has been found in other countries. In a
large UK study (Tymms, 2001; N⫽21,000 Year 2 students from
628 schools representing 44 Local Education Authorities), aca-
demic self-concept was positively related to academic achieve-
ment (i.e., the brighter I am, the better my academic self-concept)
but negatively related to the class-average achievement (i.e., the
brighter my classmates are, the lower my academic self-concept).
Also in the United Kingdom, Ireson et al. (2001) examined 3,000
Year 9 pupils from 45 schools grouped into three ability levels.
They found a BFLPE for English self-concept, although there was
no significant effect on math or science self-concepts.
These cross-cultural findings were greatly extended by Marsh
and Hau (2003), who undertook a study of the BFLPE across 26
countries. Using the Program of International Student Assessment
(PISA) database administered by the OECD in 2000 (see OECD,
2001; Marsh, Hau, Artelt, Baumert, & Peschar, 2006), Marsh and
Hau (2003) sampled 103,558 fifteen-year-olds from 26 countries.
Participants completed achievement tests and a questionnaire that
included self-concept items. Results for the entire sample indicated
that whereas the relation between individual achievement and
academic self-concept was significantly positive, the relation be-
tween school-average achievement and academic self-concept was
significantly negative. Hence, the BFLPE was evident in the total
sample of 26 countries. Marsh and Hau noted that there was
variation between the countries, and although this variation was
small, it was significant. For this reason, they continued by exam-
ining the countries individually. As in the entire sample, in each of
the 26 countries the relation between individual achievement and
academic self-concept was significantly positive. Conversely,
school-average achievement had a negative association with aca-
demic self-concept in all 26 countries and was significantly neg-
ative in 24.
However, there were few economically developing or collectiv-
ist countries included in the Marsh and Hau (2003) study. In their
sample only five countries could be regarded as collectivist coun-
tries (Brazil, Korea, Mexico, Portugal, and Russia; for details of
country classification, see Hofstede & Hofstede, 2005) and only
six as developing (Brazil, Czech Republic, Hungary, Latvia, Mex-
ico, and Russia; see World Bank, 2007). Furthermore, Marsh and
Hau did not focus on these country-level characteristics to deter-
mine whether the country-to-country level variation that they
found could be explained in terms of them. Hence, as very few
BFLPE studies have been conducted in relation to collectivist or
economically developing countries, were support for the BFLPE to
be found in these countries, the universality of the BFLPE would
be greatly enhanced. Conversely, a significant influence of these
country characteristics would provide new insights into under-
standing the BFLPE— even in industrialized, individualist coun-
tries where there is consistent support for it. This limited focus of
BFLPE research is also problematic given that McFarland and
Buehler’s (1995) results suggest that the BFLPE may be reduced
for collectivist cultures. Although McFarland and Buehler’s re-
search suggests that culture plays an important part in the BFLPE,
it was limited to an experimental study with an ad hoc sample of
participants from one country (the United States), and the sample
size was particularly small. As cultural orientation is a country-
level variable, a more adequate test of the cross-cultural general-
izability of the BFLPE would be to use a large diverse sample of
countries and representative samples of students from each coun-
try, which would allow a multilevel perspective to be taken in
regard to analyses, with country being a critical unit of analysis.
Thus, an important issue for BFLPE theory and research to
address is whether individuals in collectivist and economically
406 SEATON, MARSH, AND CRAVEN
developing countries suffer the negative effects of the BFLPE. To
deal with this issue adequately, it is necessary to use a large sample
containing a diverse number of natural settings. These settings
should span both collectivist and individualist cultures and eco-
nomically developed and developing countries. Additionally, no
study to date has explored whether the BFLPE can be moderated
by a country’s cultural orientation or its stage of economic devel-
opment. The present investigation addressed all of these issues,
thereby attending to some of the limitations of previous research.
The Present Investigation
In 1990 Schwartz and Bilsky observed, “Theories that aspire to
universality . . . must be tested in numerous, culturally diverse
samples” (p. 878). Hence, to adequately test the universality of the
BFLPE, the present investigation replicated the Marsh and Hau
(2003) study but also extended it in valuable and important ways
by (a) increasing the number of students considered from 103,558
to 265,180, (b) expanding the sample to 41 nations, (c) increasing
the number of collectivist and economically developing countries,
and (d) testing whether the BFLPE differentially operates in col-
lectivist, developing, individualist, and developed nations. We
tested the universality of the BFLPE by
1. assessing whether the BFLPE is evident in a cross-
national sample that is representative of collectivist, in-
dividualist, economically developing, and developed
countries. If this analysis demonstrates that there is
country-to-country variation in the size of the BFLPE,
then is the direction of the effect consistently negative
across all 41 countries?
2. examining whether country level variables can explain
country-to-country differences in the size of the BFLPE.
Previous research (e.g., McFarland & Buehler, 1995)
suggests that more collectivist countries may have
smaller BFLPEs, but this hypothesis has never been
tested. We were additionally interested in assessing
whether stage of economic development could modify
the BFLPE.
Method
Participants
Every 3 years the OECD administers the PISA tests to 15-year-
old students worldwide. In the 2003 administration more than a
quarter of a million students (N⫽276,165) from 41 countries
participated. These students completed paper-and-pencil tests in
reading, math, science, and problem solving. Additionally, stu-
dents completed a background questionnaire that assessed various
aspects of their home and school life. Although performances in
reading, math, science, and problem solving were assessed, only
self-concept items for math were included in this questionnaire
(see OECD, 2005b). However, these math self-concept items were
not completed by all students. As one focus of the present inves-
tigation was to replicate and extend Marsh and Hau’s (2003) study,
key methodological decisions from the earlier study were used
here to enhance the comparability of the findings. Hence, for
example, students in the present investigation who did not com-
plete the math self-concept items were removed from further
analyses. Additionally, Marsh and Hau deleted schools that were
considered too small to be included in multilevel analyses. Thus, to
be comparable with that study, schools with 10 participating stu-
dents or fewer were removed, resulting in a sample of 265,180
students in 10,221 schools across 41 countries.
Measures
Math ability. To prevent biased population estimates from
being obtained, the PISA database does not contain a single
measure of math ability but rather estimates it using five plausible
values. The PISA documentation quotes Wu and Adams’s (2002)
description of plausible values, referring to them as representing a
student’s range of abilities. Using just one measure of a student’s
ability can result in that student’s ability being measured with
error. Hence, a set of plausible values was used to estimate a
student’s ability to more appropriately represent this measurement
error (see OECD, 2005a). The PISA documentation specifically
warns against averaging these plausible values or using a single
plausible value to estimate an individual’s ability. Rather, re-
searchers are advised that statistics should be based on each
plausible value separately and then those statistics should be
averaged. In the current study, each analysis was conducted with
all five plausible values separately, and all resulting parameters
were averaged (see OECD, 2005a). Standard errors were calcu-
lated to accommodate variance between and within plausible val-
ues. This was accomplished in steps outlined by Raudenbush,
Bryk, and Congdon (2005). Both linear and quadratic components
of math ability were measured.
Math self-concept. Math self-concept was measured in the
PISA (2003) database with five items. These were “I get good
marks in mathematics,” “I learn mathematics quickly,” “I am just
not good at mathematics,” “I have always believed that mathemat-
ics is one of my best subjects,” and “In my mathematics class, I
understand even the most difficult work.” These items were scored
on a 4-point Likert scale ranging from 1 (strongly agree)to4
(strongly disagree). Four items were inverted for scoring so that a
high score was indicative of a higher math self-concept. The
reliability of this scale based on responses from the current sample
was high, with a Cronbach’s alpha of .88. This variable was
standardized (M⫽0, SD ⫽1) across the entire sample, and scores
ranged from ⫺2.24 to 2.48. The items were based in part on the
Academic Self-Description Questionnaire–II (Marsh, 1990) and
were also demonstrated to be psychometrically sound through
confirmatory factor analysis by the PISA administrators (see
OECD, 2005b). For full details concerning the construction and
validation of this scale, refer to the PISA 2003 technical report
(OECD, 2005b).
Individualist– collectivist scale. The Individualism Index scale
(Hofstede & Hofstede, 2005) was used to classify countries.
Higher scores on this bipolar scale denote that the country is higher
in individualism, and lower scores indicate that the country’s
culture is more collectivist in nature. As indicated in Table 1,
scores were spread reasonably consistently across the entire range
of this scale. In the present investigation, scores on this scale
ranged from 14 to 91 (M⫽57.05, SD ⫽23.55). The 41 countries
of the present investigation include almost the entire range of
values considered by Hofstede and Hofstede (2005), who classi-
407
A PAN-HUMAN THEORY
fied 74 countries, with individualism scores ranging from 6 to 91.
Thus, both individualist and collectivist countries were represented
in the current sample. Five of the countries in the present
investigation, as noted in Table 1, were not represented in
Hofstede and Hofstede’s scale. These 5 countries were not
included in tests of the moderating analyses of cultural orien-
tation on the BFLPE. Scale scores were subsequently standard-
ized for multilevel analyses.
Economically developed and developing. The World Bank
categorizes countries into four economic groups according to their
gross national income per capita. These categories are low income,
lower middle income, upper middle income, and high income.
Table 1 lists the countries included in these analyses and indicates
the World Bank classification. According to the World Bank, low-
and middle-income economies may also be referred to as devel-
oping economies. For purposes of the present investigation, coun-
tries were coded as either developing (0) or developed (1), which
resulted in 27 countries being classified as developed and 14 as
developing.
Data Analysis
Weights. The database documentation (OECD, 2005b) recom-
mends that sample weights be used when conducting analyses to
prevent estimates of population parameters from being biased. The
PISA database contained three weights, two country weights, and
one individual student weight. Marsh and Hau (2003) set each
country’s effective sample size equal to the number of that coun-
try’s cases, prior to weighting. To appropriately extend their find-
ings, we followed the same procedure in the current study. Hence,
the individual student weight and the country weight were com-
bined, as recommended in the database documentation. This en-
sured that countries had “weights according to their sample sizes
so that the sum of weights in each country is equal to the number
of students in the database” (OECD, 2005b, p. 327) and resulted in
the weighted and unweighted sample sizes being similar. This
combined weight was used to calculate all inferential statistics.
Standardization. The five plausible values for math ability
were standardized separately across the entire sample (M⫽0,
SD ⫽1). A similar transformation was performed for math self-
concept and the Individualism Index scale scores. Quadratic ability
variables for each of the five plausible values were created by
squaring each standardized linear plausible value. For each plau-
sible value a school-average ability variable was calculated by
averaging each plausible value separately within each school. To
ensure that variables were kept in the same metric, neither the
quadratic components of ability nor the school-average ability
variables were restandardized. Interaction terms were created for
school-average ability with each linear and quadratic plausible
value, the Individualism Index scale score, and the economic
development classification, but these interaction terms were not
restandardized. By including the quadratic component of math
ability and the interactions of linear and quadratic ability with
school-average ability, it was possible to determine whether the
BFLPE (the effect of school-average achievement) was larger for
high-ability students, low-ability students, or average-ability stu-
dents (see Marsh & Hau, 2003; Marsh & Rowe, 1996).
Overview of analyses. There is a three-level hierarchical struc-
ture in the PISA (2003) data. Individual students are at the bottom
of the hierarchy, schools are next, and countries are at the top of
the hierarchy. Ignoring such a multilevel structure, by using tra-
Table 1
Individualism Score (Based on Hofstede & Hofstede, 2005) and Economic Classification (Based on World Bank, 2007, Categories)
by Country
Country
Individualism
score Economic classification Country
Individualism
score Economic classification
United States 91 High (1) Slovak Republic 52 Upper middle (0)
Australia 90 High (1) Spain 51 High (1)
United Kingdom 89 High (1) Japan 46 High (1)
Canada 80 High (1) Russian Federation 39 Upper middle (0)
Hungary 80 Upper middle (0) Brazil 38 Lower middle (0)
Netherlands 80 High (1) Turkey 37 Upper middle (0)
New Zealand 79 High (1) Uruguay 36 Upper middle (0)
Italy 76 High (1) Greece 35 High (1)
Belgium 75 High (1) Mexico 30 Upper middle (0)
Denmark 74 High (1) Portugal 27 High (1)
France 71 High (1) Hong Kong (China) 25 High (1)
Sweden 71 High (1) Macao (China) 20 High (1)
Ireland 70 High (1) Thailand 20 Lower middle (0)
Norway 69 High (1) Korea 18 High (1)
Germany 67 High (1) Indonesia 14 Lower middle (0)
Switzerland 67 High (1) Iceland No score High (1)
Finland 63 High (1) Latvia No score Upper middle (0)
Luxembourg 60 High (1) Liechtenstein No score High (1)
Poland 60 Upper middle (0) Serbia and Montenegro No score Lower middle (0)
Czech Republic 58 Upper middle (0) Tunisia No score Lower middle (0)
Austria 55 High (1)
Note. Countries with an upper or lower middle economic classification are denoted as developing (coded as 0). Countries with a high economic
classification are denoted as developed (coded as 1).
408 SEATON, MARSH, AND CRAVEN
ditional single-level statistical methods, can lead to problems such
as violations of assumptions of independence, aggregation bias,
ecological fallacy, and heterogeneity of regression and result in
spurious significant results (Hox, 2002; Raudenbush & Bryk,
2002; Rowe, 2005). Consequently, the present investigation used
multilevel modeling to analyze these data (MLwiN; see Rasbash et
al., 2002, for details). A multilevel regression equation comprises
fixed and random effects. In the current study, the combination of
these fixed and random effects and the treatment of fixed effects
were different depending on the research question being tested.
Differences in the treatment of fixed effects were especially evi-
dent when the generalizability of the BFLPE was investigated,
whereby tests were conducted at a cross-national level and at an
individual country level.
Three sets of multilevel regression analyses were conducted.
The first set of analyses was a test of the generalizability of the
BFLPE across the entire cross-national sample. Math self-concept
was the outcome variable, and individual ability (linear and qua-
dratic), school-average math ability, and interactions of school-
average ability with linear and quadratic ability were the predictor
variables (see Appendix for the equation used in these analyses).
To be consistent with Marsh and Hau (2003), linear and quadratic
individual ability, school-average ability, their interaction terms,
and a constant were fixed effects. The intercepts, variances, and
covariances of the levels used in the model were random effects.
Variation in the country (Level 3), school (Level 2), and student
(Level 1) intercepts demonstrated the extent of the variation that
existed in the intercepts of the regression equations between coun-
tries, between schools, and between students. Additionally, multi-
level modeling can accommodate variations in the slopes of the
different levels that comprise multilevel regression equations.
Thus, at the school and country level, the effect of individual
ability (both linear and quadratic components) was allowed to vary
(sometimes called “allowing to be random”) in order to demon-
strate how much the effect of individual ability on math self-
concept varied between countries and between schools. The effect
of school-average ability was also allowed to be random at the
country level in order to demonstrate how much the effect of
school-average ability on math self-concept varied between
countries.
The second set of multilevel regression analyses tested the BFLPE
separately in the 41 individual countries that composed the cross-
national sample. Math self-concept was the outcome variable and
individual ability (linear and quadratic) and school-average math
ability were the predictor variables (see equation presented in
Appendix). The fixed components were individual ability, school-
average ability, and a constant. The random components were the
intercepts and variances of the two levels used in these models:
Level 1 (student) and Level 2 (school). With the exception of the
constant (which represents the intercepts), none of the effects for
the fixed components was allowed to vary.
The third set of analyses tested the moderating effect of cultural
orientation and economic development on the BFLPE across the
entire cross-national sample. Math self-concept was the outcome
variable, and individual ability (both linear and quadratic), school-
average math ability, the moderator (either cultural orientation or
economic development), and the interaction of school-average
ability with the respective moderator were the predictor variables.
To determine whether country characteristics could explain
country-to-country differences in the size of the BFLPE, we tested
seven models, starting with a base model and adding predictors in
subsequent models. In doing so, we expected to be able to deter-
mine the contribution of the country-level variables by examining
the significance of these predictors and by evaluating the extent of
changes in the variance components.
The first model was the base model with no predictors. In the
second model, we added linear ability, quadratic ability, and
school-average ability. All these predictors were kept fixed. In the
third model, we allowed the effect of the predictors to vary across
countries. In subsequent models the country-level predictors and
their interactions with school-average ability were added to the
previous model. Hence, in the fourth model, individualism–
collectivism was added and left as a fixed effect. The interaction of
school-average ability and individualism– collectivism was added
in the fifth model. In the sixth model, economic development was
added as a fixed effect, and, finally, in the seventh model, the
interaction between school-average ability and economic develop-
ment was added (see Appendix for details of the equations used to
test these models).
Results
The BFLPE in the Cross-National Sample
Results indicated evidence of the BFLPE in a cross-national
sample of 41 economically and culturally diverse countries. As
indicated in Table 2, the relation between school-average math
ability and math self-concept was significantly negative (⫺0.300).
Students in schools with an average ability level one standard
Table 2
The Effects of Individual Achievement and School-Average
Achievement on Math Self-Concept for the Entire
Cross-National Sample
Effect SE
Fixed effects
Main effects
Individual achievement (linear) 0.520
ⴱ
0.020
Individual achievement (quadratic) 0.118 0.005
School-average achievement ⫺0.300
ⴱ
0.023
Interaction effects
Linear Achievement ⫻School-Average
Achievement ⫺0.048
ⴱ
0.008
Quadratic Achievement ⫻School-Average
Achievement ⫺0.016
ⴱ
0.005
Constant ⫺0.124
ⴱ
0.036
Random effects
Level 3 country intercept 0.052
ⴱ
0.015
Level 3 individual achievement (linear) 0.014
ⴱ
0.003
Level 3 individual achievement (quadratic) 0.001
ⴱ
0.000
Level 3 school-average achievement 0.016
ⴱ
0.006
Level 2 school intercept 0.032
ⴱ
0.005
Level 2 individual achievement (linear) 0.007
ⴱ
0.003
Level 2 individual achievement (quadratic) 0.000 0.000
Level 1 student intercept 0.777
ⴱ
0.035
Note. All parameter estimates are significant at 0.05 level (denoted by
asterisk) when they differ from zero by more than two standard errors.
409
A PAN-HUMAN THEORY
deviation above the mean had math self-concepts that were 0.3 of
a standard deviation below the average self-concept level.
Both the linear and quadratic components of individual math
ability had a statistically significant positive relation with math
self-concept (0.520 and 0.118, respectively). The nature of these
positive relations is depicted in Figure 2. After controlling for
school-average math ability, math self-concept increased as ability
increased, but this increase was much more gradual for those at the
lower end of the ability spectrum. Ratings of math self-concept did
not increase as steadily for students whose ability levels were
below average compared with students with above-average ability.
BFLPE studies are inherently multilevel studies. However, to
date, there are no commonly defined measures of effect sizes for
group-level constructs in multilevel studies. To rectify this,
Tymms (2004; see also Trautwein, Gerlach, & Lu¨dtke, in press)
has proposed an effect size measure comparable to Cohen’s d
(Cohen, 1988) that is calculated with the following formula:
⌬⫽2⫻B⫻SDpredictor /e.
In this formula, B is the unstandardized regression coefficient in
the multilevel model, SD
predictor
is the standard deviation of the
predictor variable at the class level, and
e
is the residual standard
deviation at the student level. When this formula is applied to the
current data, the effect size for the total sample is .49 (⌬⫽2⫻
⫺.300 ⫻.72/.881), clearly large enough to be of theoretical and
practical importance.
Additionally, the effect of the Individual Math Ability (linear) ⫻
School-Average Math Ability interaction on math self-concept
was statistically negatively significant (⫺0.048). Although
smaller, the quadratic Individual Math Ability ⫻School-Average
Math Ability interaction was also statistically negatively signifi-
cant (⫺0.016). The pattern of the Individual Ability ⫻School-
Average interaction is depicted in Figure 3. The figure indicates
that the BFLPE was somewhat stronger (i.e., the effect of school-
average ability is more negative) for high-ability students attend-
ing high-ability schools, as reflected in the somewhat steeper
slopes for students whose ability was one to two standard devia-
tions above the mean. However, students of all ability levels
suffered the BFLPE, whereby in high-ability schools students of
all ability levels had lower math self-concepts than their counter-
parts in low-ability schools. Nevertheless, although the Individual
Ability ⫻School-Average Ability interaction was significantly
negative, it was small, especially given the power engendered by
the large sample size. Moreover, although there was a significantly
negative quadratic interaction, it was so small that it is not even
visually evident in the graph of the interaction equation.
The BFLPE in 41 Individual Countries
The analysis of the cross-national sample demonstrated a statisti-
cally significant, but small, random effect of school-average math
ability at the country level (0.016). Although the size of this country
variation was small, it was statistically significant, indicating that
the size of the BFLPE was not entirely consistent across countries.
To determine whether the direction of the effect was consistently
negative across all 41 countries, we continued by evaluating the
BFLPE on a country-by-country basis.
Table 3 presents the results of multilevel modeling analyses ac-
cording to individual countries. In all 41 countries, individual ability
(linear) was a significantly positive predictor of math self-concept,
ranging from 0.223 in Indonesia to 0.786 in Tunisia. Hence, consis-
tent with Marsh and Hau (2003), students with higher math ability
tended to have higher math self-concepts in all countries. Even in
Indonesia, where the effect was smallest, students whose ability levels
were one standard deviation above the mean had math self-concepts
that were 0.223 of a standard deviation above the average math
self-concept score. The quadratic component of individual ability was
a significantly positive predictor of math self-concept in all countries
except Japan, where its effect was not statistically significant. The
relation between quadratic individual ability and math self-concept
ranged from 0.033 in Japan to 0.175 in Poland, indicating a nonlinear
relation between math ability and math self-concept, as depicted in
Figure 2 for the entire sample.
As expected from the random effects in the cross-national
sample analysis, the size of the BFLPE varied across countries.
Results demonstrated that school-average math ability negatively
predicted math self-concept in all 41 countries, ranging from
⫺0.014 in Korea to ⫺0.713 in Germany (see Table 3). This effect
was significantly negative in 38 of these countries and not statis-
tically significant in 3 (Iceland, Ireland, and Korea). Controlling
for individual ability, we found that students in higher ability
schools in 38 of the 41 countries had lower math self-concepts than
equally able students in lower ability schools. The smallest signif-
icant negative association was in Tunisia, where students in
schools with an average ability level one standard deviation above
the mean had math self-concepts that were 0.161 of a standard
deviation below the average self-concept level. However, in Ger-
many, where the relation was largest, students in schools with an
average ability level one standard deviation above the mean had
math self-concepts that were 0.713 of a standard deviation below
the average self-concept level.
The Moderating Effect of Cultural Orientation and
Economic Development on the BFLPE
Table 4 displays the results of the seven models tested. In Model
2, with the individual variables only as fixed effects, a BFLPE was
present (main effects of linear ability and quadratic ability on math
self-concept were significantly positive, 0.523 and 0.097, respec-
tively, and the main effect of school-average ability on math
self-concept was significantly negative, ⫺0.363). When these ef-
-1
-0.5
0
0.5
1
1.5
-2SD -1SD Mean +1SD +2SD
Individual Ability
Math Self-Concept
Figure 2. The relation between individual ability and math self-concept
while controlling for school-average ability, based on predicted values.
410 SEATON, MARSH, AND CRAVEN
fects were allowed to vary across countries (see Model 3), the main
effect of linear ability and school-average ability on math self-
concept decreased (0.511 and ⫺0.313, respectively), whereas that
of quadratic ability increased (0.104). When cultural rating, an
indication of a country’s individualist– collectivist cultural orien-
tation, was added to the model (Model 4), its effect on math
self-concept was not statistically significant (0.025). However,
when the interaction of Cultural Rating ⫻School-Average Ability
was added (Model 5), it had a statistically significant negative
effect on math self-concept (⫺0.053). As seen in Figure 4, the
BFLPE was slightly larger for students from individualist coun-
tries than for those from collectivist countries, as indicated by the
slightly steeper slope for students from more individualist coun-
tries. Nevertheless, although our a priori hypothesis was supported,
such a small interaction effect found in such a large sample, though
statistically significant, is not of substantive or practical impor-
tance. The BFLPE is clearly evident in both collectivist and
individualist countries, supporting the generalizability of the
BFLPE. Irrespective of whether a country had an individualist or
a collectivist culture, compared with being an equally able student
in a low-ability school, being in a high-ability school was associ-
ated with lower math self-concepts.
When added, the main effect of stage of economic development
was statistically negatively significant (⫺0.177; see Model 6),
indicating that students in economically developed countries had
math self-concepts that were 0.177 of a standard deviation signif-
icantly lower than students in developing countries. However, the
Economic Development ⫻School-Average Math Ability interac-
tion on math self-concept (⫺0.058; see Model 7) was not statis-
tically significant, indicating that the BFLPE was similar across
both economically developing and developed countries. In sum-
mary, being in a high-ability class had a similar negative relation
with math self-concept, irrespective of the economic development
of the country.
The addition of the country-level variables produced very little
difference in the variance components for linear ability, quadratic
ability, and school-average ability, indicating that the slopes did
not change with the addition of these new variables. Hence,
although there was significant variation between countries regard-
ing the size of these effects, this variation was not explained by the
country-level variables we added.
Discussion
The current study, the largest cross-cultural BFLPE study un-
dertaken, tested the universality of the BFLPE and investigated
whether the size of the BFLPE varied across different levels of
individual student ability. In doing so, the current study used
nationally representative samples of students from each of 41
economically and culturally diverse countries, thus providing an
appropriate vehicle for adequately testing the universality of the
BFLPE (see Schwartz & Bilsky, 1990).
Although some studies have demonstrated the existence of the
BFLPE in collectivist countries (Marsh et al., 2000; Zeidner &
Schleyer, 1998) and developing countries (Marsh & Hau, 2003),
these have been few, and most previous BFLPE research has been
based primarily on developed and individualist countries (e.g.,
Craven et al., 2000; Marsh & Craven, 2002; Marsh et al., 2001;
Marsh & Rowe, 1996; Mulkey et al., 2005; Marsh et al., 2007).
Marsh and Hau’s (2003) study was the most impressive in terms of
the number of different countries considered and is the study on
which the current investigation was based. However, the Marsh
and Hau study was based on PISA 2000 data that were less
representative of collectivist and economically disadvantaged
countries and did not specifically test whether country-to-country
variation in the size of the BFLPE could be explained by country-
level cultural characteristics. In this respect, the current study
extends Marsh and Hau’s work by (a) increasing the number of
-2
-1
0
1
2
3
-2SD -1SD M ean +1SD +2SD
School-Average Ability
Mat h Self-C onc ept
-2SD Ab ility
-1SD Ab ility
Mean Ability
+1SD A bility
+2SD A bility
Figure 3. Math self-concept as a function of school-average ability and individual ability, based on predicted
values.
411
A PAN-HUMAN THEORY
students considered from 103,558 to 265,180, (b) increasing the
sample of countries from 26 to 41, (c) expanding the number of
collectivist and developing countries tested, and (d) assessing
whether the BFLPE differentially operated in collectivist, devel-
oping, individualist, and developed nations.
The present investigation demonstrated that the BFLPE was
evident not only in developed and individualist countries but also
in economically developing countries and collectivist nations. This
was demonstrated in the cross-national sample and on a country-
by-country basis. The cross-national sample of 41 countries com-
prised developed countries, individualist countries, economically
developing countries, and collectivist countries. Analyses using
this sample demonstrated that individual ability had a positive
association with math self-concept and that, after controlling for
individual ability, school-average math ability had a negative
association with math self-concept. This negative association is
characteristic of the BFLPE. The size of this effect was .49, thus
demonstrating that the BFLPE is both theoretically and practically
important. Thus, in the cross-national sample, students in high-
ability schools had lower math self-concepts than their equally
able counterparts in mixed- and low-ability schools. These find-
ings are consistent with previous research in individualist (e.g.,
Craven et al., 2000; Marsh & Craven, 2002; Marsh et al., 2001;
Marsh & Rowe, 1996; Marsh et al., 2007; Mulkey et al., 2005),
Table 3
The Effects of Individual Ability and School-Average Ability on Math Self-Concept, in Individual Countries
Country (n)
Fixed effects Random effects
Linear ability
(SE)
Quadratic ability
(SE)
School-average
ability (SE) Constant (SE)
School level
(SE)
Individual level
(SE)
Australia (12,383) 0.434
ⴱ
(0.014) 0.077
ⴱ
(0.008) ⫺0.281
ⴱ
(0.025) ⫺0.028
ⴱ
(0.014) 0.015
ⴱ
(0.003) 0.710
ⴱ
(0.011)
Austria (4,378) 0.599
ⴱ
(0.030) 0.103
ⴱ
(0.017) ⫺0.483
ⴱ
(0.053) ⫺0.076
ⴱ
(0.033) 0.067
ⴱ
(0.010) 0.984
ⴱ
(0.025)
Belgium (8,460) 0.382
ⴱ
(0.019) 0.081
ⴱ
(0.010) ⫺0.447
ⴱ
(0.028) ⫺0.120
ⴱ
(0.022) 0.025
ⴱ
(0.005) 0.886
ⴱ
(0.017)
Brazil (4,120) 0.538
ⴱ
(0.036) 0.072
ⴱ
(0.013) ⫺0.372
ⴱ
(0.039) 0.052 (0.041) 0.025
ⴱ
(0.008) 0.775
ⴱ
(0.018)
Canada (25,685) 0.624
ⴱ
(0.012) 0.118
ⴱ
(0.008) ⫺0.427
ⴱ
(0.025) ⫺0.008 (0.014) 0.030
ⴱ
(0.003) 0.942
ⴱ
(0.010)
Czech Republic (6,019) 0.576
ⴱ
(0.021) 0.068
ⴱ
(0.012) ⫺0.446
ⴱ
(0.029) ⫺0.245
ⴱ
(0.019) 0.021
ⴱ
(0.005) 0.698
ⴱ
(0.015)
Denmark (4,041) 0.657
ⴱ
(0.021) 0.078
ⴱ
(0.017) ⫺0.296
ⴱ
(0.055) 0.044 (0.027) 0.023
ⴱ
(0.007) 0.778
ⴱ
(0.017)
Finland (5,730) 0.643
ⴱ
(0.022) 0.149
ⴱ
(0.014) ⫺0.301
ⴱ
(0.065) ⫺0.337
ⴱ
(0.038) 0.017
ⴱ
(0.004) 0.762
ⴱ
(0.017)
France (4,154) 0.541
ⴱ
(0.034) 0.111
ⴱ
(0.021) ⫺0.383
ⴱ
(0.047) ⫺0.343
ⴱ
(0.031) 0.047
ⴱ
(0.009) 0.922
ⴱ
(0.023)
Germany (4,357) 0.689
ⴱ
(0.030) 0.090
ⴱ
(0.018) ⫺0.713
ⴱ
(0.042) 0.047 (0.027) 0.032
ⴱ
(0.008) 1.136
ⴱ
(0.024)
Greece (4,228) 0.523
ⴱ
(0.023) 0.071
ⴱ
(0.013) ⫺0.174
ⴱ
(0.047) 0.160
ⴱ
(0.022) 0.016
ⴱ
(0.004) 0.694
ⴱ
(0.019)
Hong
Kong (China) (4,463) 0.369
ⴱ
(0.023) 0.084
ⴱ
(0.013) ⫺0.200
ⴱ
(0.035) ⫺0.517
ⴱ
(0.026) 0.014
ⴱ
(0.004) 0.747
ⴱ
(0.018)
Hungary (4,356) 0.415
ⴱ
(0.025) 0.095
ⴱ
(0.016) ⫺0.323
ⴱ
(0.042) ⫺0.268
ⴱ
(0.025) 0.029
ⴱ
(0.006) 0.665
ⴱ
(0.018)
Iceland (3,043) 0.671
ⴱ
(0.030) 0.150
ⴱ
(0.019) ⫺0.209 (0.125) ⫺0.265
ⴱ
(0.046) 0.027
ⴱ
(0.009) 1.041
ⴱ
(0.030)
Indonesia (10,448) 0.223
ⴱ
(0.027) 0.063
ⴱ
(0.011) ⫺0.235
ⴱ
(0.030) ⫺0.042 (0.032) 0.037
ⴱ
(0.004) 0.374
ⴱ
(0.009)
Ireland (3,801) 0.434
ⴱ
(0.025) 0.120
ⴱ
(0.019) ⫺0.103 (0.057) ⫺0.205
ⴱ
(0.023) 0.020
ⴱ
(0.006) 0.758
ⴱ
(0.022)
Italy (11,436) 0.568
ⴱ
(0.016) 0.073
ⴱ
(0.010) ⫺0.409
ⴱ
(0.028) ⫺0.122
ⴱ
(0.018) 0.049
ⴱ
(0.006) 0.926
ⴱ
(0.015)
Japan (4,681) 0.355
ⴱ
(0.028) 0.033 (0.017) ⫺0.307
ⴱ
(0.040) ⫺0.641
ⴱ
(0.026) 0.029
ⴱ
(0.007) 0.896
ⴱ
(0.022)
Korea (5,380) 0.410
ⴱ
(0.025) 0.085
ⴱ
(0.015) ⫺0.014 (0.033) ⫺0.711
ⴱ
(0.024) 0.016
ⴱ
(0.004) 0.669
ⴱ
(0.014)
Latvia (4,522) 0.463
ⴱ
(0.017) 0.121
ⴱ
(0.014) ⫺0.221
ⴱ
(0.048) ⫺0.240
ⴱ
(0.020) 0.018
ⴱ
(0.005) 0.556
ⴱ
(0.015)
Liechtenstein (320) 0.492
ⴱ
(0.076) 0.129
ⴱ
(0.038) ⫺0.554
ⴱ
(0.061) 0.001 (0.063) 0.000 (0.000) 0.974
ⴱ
(0.125)
Luxembourg (3,870) 0.427
ⴱ
(0.040) 0.130
ⴱ
(0.023) ⫺0.428
ⴱ
(0.075) ⫺0.060 (0.036) 0.014
ⴱ
(0.006) 1.191
ⴱ
(0.035)
Macao (China) (1,250) 0.384
ⴱ
(0.037) 0.105
ⴱ
(0.025) ⫺0.330
ⴱ
(0.091) ⫺0.352
ⴱ
(0.050) 0.032
ⴱ
(0.013) 0.779
ⴱ
(0.030)
Mexico (28,408) 0.667
ⴱ
(0.014) 0.153
ⴱ
(0.007) ⫺0.357
ⴱ
(0.019) 0.172
ⴱ
(0.014) 0.029
ⴱ
(0.002) 0.621
ⴱ
(0.007)
Netherlands (3,860) 0.607
ⴱ
(0.046) 0.083
ⴱ
(0.021) ⫺0.696
ⴱ
(0.051) ⫺0.083
ⴱ
(0.035) 0.041
ⴱ
(0.009) 0.868
ⴱ
(0.024)
New Zealand (4,428) 0.394
ⴱ
(0.021) 0.114
ⴱ
(0.013) ⫺0.314
ⴱ
(0.042) ⫺0.021 (0.027) 0.019
ⴱ
(0.005) 0.670
ⴱ
(0.017)
Norway (3,904) 0.729
ⴱ
(0.022) 0.184
ⴱ
(0.016) ⫺0.168
ⴱ
(0.069) ⫺0.419
ⴱ
(0.024) 0.019
ⴱ
(0.006) 0.862
ⴱ
(0.023)
Poland (4,312) 0.534
ⴱ
(0.017) 0.175
ⴱ
(0.014) ⫺0.279
ⴱ
(0.046) ⫺0.146
ⴱ
(0.018) 0.013
ⴱ
(0.004) 0.644
ⴱ
(0.018)
Portugal (4,512) 0.579
ⴱ
(0.021) 0.139
ⴱ
(0.017) ⫺0.205
ⴱ
(0.045) ⫺0.259
ⴱ
(0.017) 0.010
ⴱ
(0.005) 0.758
ⴱ
(0.019)
Russian
Federation (5,822) 0.360
ⴱ
(0.016) 0.072
ⴱ
(0.011) ⫺0.187
ⴱ
(0.033) 0.069
ⴱ
(0.018) 0.024
ⴱ
(0.005) 0.540
ⴱ
(0.013)
Serbia and
Montenegro (4,164) 0.473
ⴱ
(0.022) 0.092
ⴱ
(0.017) ⫺0.181
ⴱ
(0.043) 0.047 (0.026) 0.021
ⴱ
(0.005) 0.719
ⴱ
(0.020)
Slovak Republic (7,118) 0.550
ⴱ
(0.019) 0.079
ⴱ
(0.013) ⫺0.411
ⴱ
(0.033) ⫺0.174
ⴱ
(0.017) 0.029
ⴱ
(0.004) 0.565
ⴱ
(0.012)
Spain (10,695) 0.517
ⴱ
(0.015) 0.119
ⴱ
(0.012) ⫺0.244
ⴱ
(0.044) ⫺0.325
ⴱ
(0.018) 0.048
ⴱ
(0.006) 0.857
ⴱ
(0.013)
Sweden (4,513) 0.548
ⴱ
(0.018) 0.137
ⴱ
(0.013) ⫺0.202
ⴱ
(0.055) ⫺0.098
ⴱ
(0.023) 0.020
ⴱ
(0.005) 0.756
ⴱ
(0.017)
Switzerland (8,023) 0.436
ⴱ
(0.019) 0.105
ⴱ
(0.012) ⫺0.446
ⴱ
(0.033) ⫺0.028
ⴱ
(0.018) 0.031
ⴱ
(0.005) 1.004
ⴱ
(0.018)
Thailand (5,120) 0.276
ⴱ
(0.023) 0.093
ⴱ
(0.012) ⫺0.194
ⴱ
(0.035) ⫺0.163
ⴱ
(0.024) 0.018
ⴱ
(0.005) 0.422
ⴱ
(0.013)
Tunisia (4,653) 0.786
ⴱ
(0.069) 0.138
ⴱ
(0.024) ⫺0.161
ⴱ
(0.062) 0.586
ⴱ
(0.057) 0.038
ⴱ
(0.008) 1.195
ⴱ
(0.027)
Turkey (4,652) 0.520
ⴱ
(0.032) 0.054
ⴱ
(0.016) ⫺0.252
ⴱ
(0.040) ⫺0.063
ⴱ
(0.030) 0.030
ⴱ
(0.006) 0.883
ⴱ
(0.024)
United Kingdom (9,262) 0.457
ⴱ
(0.016) 0.120
ⴱ
(0.012) ⫺0.344
ⴱ
(0.029) ⫺0.060
ⴱ
(0.015) 0.020
ⴱ
(0.004) 0.739
ⴱ
(0.013)
United States (5,202) 0.516
ⴱ
(0.018) 0.132
ⴱ
(0.013) ⫺0.230
ⴱ
(0.043) 0.131
ⴱ
(0.020) 0.023
ⴱ
(0.005) 0.929
ⴱ
(0.019)
Uruguay (5,407) 0.575
ⴱ
(0.026) 0.091
ⴱ
(0.013) ⫺0.240
ⴱ
(0.039) 0.080
ⴱ
(0.026) 0.041
ⴱ
(0.007) 0.911
ⴱ
(0.019)
Note. Parameter estimates are significant at 0.05 level (denoted by asterisk) when they differ from zero by more than two standard errors. Standard error
in parentheses.
412 SEATON, MARSH, AND CRAVEN
collectivist (Marsh et al., 2000; Zeidner & Schleyer, 1998), and
developing (Marsh & Hau, 2003) countries.
In the cross-national sample, consistent with the findings of
Marsh and Hau (2003), there was a significant but small variation
in the size of the BFLPE from country to country. To ascertain
whether a BFLPE was displayed in the individual countries that
formed the cross-national sample, separate multilevel regressions
evaluated the BFLPE on an individual country basis. Controlling
for individual ability, we found that school-average math ability
had a negative association with math self-concept (the BFLPE) in
all 41 countries, with this association being statistically significant
in 38 countries. These results replicate those of Marsh and Hau
(2003) but also extend them by including many more culturally
and economically diverse countries. Of the 3 countries displaying
nonsignificant associations, 1 could be regarded as individualist
(Ireland) and 1 as collectivist (Korea). The third, Iceland, did not
receive a score on the Individualism Index, which was used to
classify countries. All 3 countries are regarded as economically
developed (see World Bank, 2007).
To further strengthen claims concerning the generalizability of
the BFLPE, we conducted analyses to ascertain whether cultural
orientation or stage of economic development could modify the
BFLPE and explain country-to-country differences in the size of
the BFLPE. Results indicated that the BFLPE was present irrespec-
tive of a country’s economic development. However, although there
was a significant Cultural Orientation ⫻School-Average Ability
interaction, it was small and, given the large sample size, has to be
regarded as trivial. As Howell (1997) noted, large samples such as
this provide a great deal of power and can produce small signifi-
cant effects that would not otherwise reach significance in more
moderately sized samples. Additionally, the inclusion of the
country-level variables did not affect the variance components,
further suggesting that the interaction effect is not substantively or
practically important. Consequently, it appears that the BFLPE is
found not only in developed and individualist countries but also in
collectivist and economically developing countries. This conclu-
sion is also borne out when one considers the size of the BFLPE
in countries differing in cultural orientation and economic devel-
opment. For example, Mexico and the United Kingdom have
similarly high scores on school-average ability (ranked 15th and
16th, respectively, out of 41 countries, in terms of largest
Table 4
Models Assessing the Effect of Adding Country-Level Predictors
Effect Model 1 Model 2 Model 3 Model 4 Model 5 Model 6 Model 7
Main effects
Constant ⫺0.035 (0.026) ⫺0.137
ⴱ
(0.030) ⫺0.122
ⴱ
(0.035) ⫺0.139
ⴱ
(0.033) ⫺0.138
ⴱ
(0.033) ⫺0.005 (0.053) ⫺0.013
ⴱ
(0.050)
Linear ability 0.523
ⴱ
(0.021) 0.511
ⴱ
(0.019) 0.502
ⴱ
(0.019) 0.502
ⴱ
(0.019) 0.512
ⴱ
(0.019) 0.511
ⴱ
(0.019)
Quadratic ability 0.097
ⴱ
(0.009) 0.104
ⴱ
(0.005) 0.101
ⴱ
(0.006) 0.101
ⴱ
(0.006) 0.104
ⴱ
(0.005) 0.104
ⴱ
(0.005)
School-average ability ⫺0.363
ⴱ
(0.024) ⫺0.313
ⴱ
(0.022) ⫺0.321
ⴱ
(0.023) ⫺0.324
ⴱ
(0.022) ⫺0.313
ⴱ
(0.022) ⫺0.276
ⴱ
(0.023)
Individualism–collectivism 0.025 (0.038) 0.039 (0.039)
School-Average ⫻
Individualism–Collectivism ⫺0.053
ⴱ
(0.020)
Developed
a
⫺0.177
ⴱ
(0.064) ⫺0.164
ⴱ
(0.063)
School-Average ⫻Stage of
Economic Development ⫺0.058 (0.037)
Random effects
Level 3 country intercept 0.028
ⴱ
(0.008) 0.038
ⴱ
(0.011) 0.050
ⴱ
(0.014) 0.041
ⴱ
(0.010) 0.041
ⴱ
(0.010) 0.045
ⴱ
(0.011) 0.045
ⴱ
(0.011)
Level 3 linear ability 0.014
ⴱ
(0.003) 0.013
ⴱ
(0.003) 0.013
ⴱ
(0.003) 0.014
ⴱ
(0.003) 0.014
ⴱ
(0.003)
Level 3 quadratic ability 0.001
ⴱ
(0.000) 0.001
ⴱ
(0.000) 0.001
ⴱ
(0.000) 0.001
ⴱ
(0.000) 0.001
ⴱ
(0.000)
Level 3 school-average ability 0.018
ⴱ
(0.006) 0.018
ⴱ
(0.006) 0.015
ⴱ
(0.005) 0.018
ⴱ
(0.006) 0.017
ⴱ
(0.005)
Level 2 school intercept 0.036
ⴱ
(0.003) 0.034
ⴱ
(0.002) 0.029
ⴱ
(0.002) 0.027
ⴱ
(0.002) 0.027
ⴱ
(0.002) 0.029
ⴱ
(0.002) 0.029
ⴱ
(0.002)
Level 1 individual intercept 0.940
ⴱ
(0.051) 0.788
ⴱ
(0.035) 0.782
ⴱ
(0.035) 0.775
ⴱ
(0.036) 0.775
ⴱ
(0.036) 0.782
ⴱ
(0.035) 0.782
ⴱ
(0.035)
⫺2
ⴱ
log-likelihood 734026.000 696781.560
b
694090.560
b
648469.900
b
648464.060
b
694084.080
b
694082.200
Note. All parameter estimates are significant at 0.05 level (denoted by asterisk) when they differ from zero by more than two standard errors. Standard
errors are in parentheses.
a
Developing countries constitute the reference group.
b
The change in ⫺2
ⴱ
log-likelihood from previous model is significant at 0.05 level.
-0.4
-0.2
0
0.2
0.4
Lo w M ea n Hi g h
School-Average Ability
Math Self-Concept
More Collectivist
Mean Individualist
/Co lle ct ivis t
Mo re Ind ividualist
Figure 4. Math self-concept as a function of school-average ability and
individualism– collectivism dimension, based on predicted values. High
school-average ability ⫽1 standard deviation above the mean for school-
average math ability, and low school-average ability ⫽1 standard devia-
tion below the mean. Similarly, more collectivist ⫽1 standard deviation
below the mean for the individualism– collectivism dimension, and more
individualist ⫽1 standard deviation above the mean for the individualism–
collectivism dimension. Linear and quadratic abilities are held constant.
413
A PAN-HUMAN THEORY
BFLPEs), but Mexico is a collectivist and developing country and
the United Kingdom is individualist and developed. Conversely,
Indonesia and the United States display similarly moderate school-
average-ability effects (ranked 27th and 28th, respectively). How-
ever, Indonesia is the most collectivist country in the sample and
the United States is the most individualist. Additionally, Indonesia
is regarded as a developing nation, and the United States is one of
the most economically developed nations in the world. Hence, it
appears that the economic development or the cultural orientation
of a country is not indicative of the size of the BFLPE.
It is of interest that the BFLPE (the negative effect of school-
average ability) was not significant in Korea in both the present
investigation and the earlier Marsh and Hau (2003) study. One can
only speculate as to the reasons for this. Perhaps in Korea the
prestige of attending a high-ability school (a reflected glory effect)
is so great that it counterbalances the negative effect of the BFLPE.
For example, Marsh et al. (2000) found that a reflective effect
partially— but not completely— counteracted the BFLPE in Hong
Kong. Indeed, after controlling for the reflective glory effect
(based on a separate measure), the negative effect of school-
average ability (the BFLPE) was substantially more negative. This
speculation is supported by anecdotal evidence from a Korean
colleague. When asked why Korean students do not display a
BFLPE, he answered that in Korea “big fish should play in big
pond. . . . More and more people prefer to be a tail of dragon rather
than a head of snake because the title [how prestigious the school is]
matters and the title is more than just title” (J. Jeon, personal
communication, March 29, 2008). Future research in Korea could
explore this notion that reflected glory effects may prevent stu-
dents from experiencing the BFLPE. Hence, research in Korea
may be very productive in finding ways to overcome the BFLPE.
As with previous BFLPE research (e.g., Marsh & Parker, 1984;
Marsh et al., 2007; Marsh et al., 2008), in all instances, individual
ability positively predicted math self-concept, whereby students of
higher ability had higher math self-concepts. Additionally, consis-
tent with findings by Marsh and Hau (2003; see also related
discussion by Marsh & Rowe, 1996), there was evidence of a
quadratic association with math self-concept for individual ability.
Although math self-concept increased as ability increased, the
findings were suggestive of a gradual increase for students of
below-average ability and a stronger increase for students of
above-average ability.
Previous research assessing the moderating influence of indi-
vidual ability on the BFLPE has been inconsistent in direction and
in statistical significance (Coleman & Fults, 1985; Marsh et al.,
1995; Marsh & Hau, 2003; Marsh et al., 2008; Reuman, 1989). In
the present investigation across the span of ability levels, students
who attended high-ability schools had lower math self-concepts
than equally able students in low-ability schools. This decline in
math self-concept was slightly worse, however, for high-ability
students in high-ability schools. Students whose ability levels were
one to two standard deviations above the mean suffered the neg-
ative effects of the BFLPE slightly more than low-ability students.
Notwithstanding the significant moderating effect of individual
ability, the linear-ability interaction effect was small, and the
quadratic-ability interaction effect was even smaller, so small that
it was not visually evident in a graph of the interaction. It is
important to note that in the previously largest BFLPE study
(Marsh & Hau, 2003), this interaction was not even statistically
significant. Furthermore, other research has found small but sta-
tistically significant interactions whereby the BFLPE was largest
for the most able student (Seaton et al., 2008) or was largest for the
average-ability students (Marsh & Rowe, 1996). However, find-
ings from all these studies—including the present investigation—
are consistent with BFLPE theoretical predictions showing that the
BFLPE does not vary much with individual levels of achievement
(for further discussion, see Marsh et al., 2008). In summary, the
BFLPE generalizes over levels of individual student achievement.
In summary, the generalizability of the BFLPE was tested in the
current study both cross-nationally in 41 countries and on an
individual country basis. Moreover, its generalizability was further
assessed by evaluating the moderating effects on the BFLPE of
cultural orientation and status of economic development. Results
of this comprehensive test of the BFLPE indicate that the BFLPE
is not only a symptom of developed countries and individualist
societies but also evident in developing nations and collectivist
countries of the world.
Strengths and Limitations
The sample used in the current study was a major strength. As
it encompassed numerous countries differing not only in cultural
orientation but also in economic development, it was well suited to
test the BFLPE’s universality. The present investigation also ben-
efited from significantly extending Marsh and Hau’s (2003) study.
Whereas Marsh and Hau sampled 26 countries, the present inves-
tigation comprised 41 countries. The present investigation also
increased the number of collectivist and economically developing
countries tested and, as such, provided a stronger test of the
universality of the BFLPE than was possible in the Marsh and Hau
study. The current study also tested whether the BFLPE was
moderated by cultural orientation or economic development—an
issue not covered by Marsh and Hau. Additionally, math self-
concept, as used in the present investigation, was a stronger
measure of academic self-concept than that used in the Marsh and
Hau study, as it was based on five items rather than three. Hence,
the present investigation provided a considerable and valuable
extension to Marsh and Hau’s findings.
Additionally, multilevel modeling was used for analyses. As
previously noted, when single-level statistical techniques are used
to analyze data that comprise a multilevel structure, serious prob-
lems can occur. For example, independence of observations cannot
be assumed when samples comprised groups nested within one
another, as in the current study. Multilevel modeling can overcome
such problems by partitioning the variance from each group level
and thus providing more accurate statistical information.
As the present investigation was based on correlational data,
causality cannot be inferred. Although it is both impossible and
unethical to use random assignment in educational studies such as
this one, causality is still an important issue. Results of the current
study assume that individual and school-average ability precede
self-concept, but research has shown that academic achievement
and self-concept are reciprocally related (Marsh & Craven, 2006).
So, it is unknown whether students entered high-ability schools
with lower self-concepts or whether their self-concepts were low-
ered by virtue of attending such schools. For the causal relation
between self-concept and attendance at high-ability schools to be
explicated more fully, self-concept should be measured before,
414 SEATON, MARSH, AND CRAVEN
during, and after attending such a school. It is, however, important
to reiterate that previous longitudinal studies have shown that even
when ability levels of students are based on tests collected before
they actually attended the new school (e.g., Marsh et al., 2000),
there is clear support for the BFLPE. Similarly, longitudinal stud-
ies have shown that size of the BFLPE actually increases over
time. Hence, although causal conclusions must always be made
cautiously, the growing body of empirical research is consistent
with theoretical predictions that school-average ability has a neg-
ative effect on academic self-concept (for further discussion, see
Marsh et al., 2008).
Additionally, a caveat should be placed on our analyses of
cultural orientation (individualist– collectivist) and economic de-
velopment at the country level. Consistent with our emphasis on a
multilevel perspective, it is important that these findings not be
misinterpreted to mean that these variables do not interact with
similar variables at the level of the individual school or individual
student. However, other research does suggest that the BFLPE is
robust across a wide variety of individual student characteristics
(e.g., gender, socioeconomic status, achievement, motivational
orientation; see Chanal, Marsh, Sarrazin, & Bois, 2005; Marsh,
2007; Marsh et al., 1995; Marsh et al., 2008; Marsh & Parker,
1984; Seaton, 2007). Nevertheless, although our analyses were
concerned with country-level differences, evaluating individual
differences in corresponding variables measured at the individual
student level may be a fruitful area for future BFLPE research to
consider.
That five countries were missing cultural orientation scores is a
potential limitation of the present investigation. Although this
would have been problematic had we focused on the individual
level, we were interested in information at the country level.
Hence, it was considered to be a prudent strategy to conduct
moderating analyses without these countries. Additionally, as there
was no need to exclude these countries from other analyses where
they make a potentially important contribution, we feel that re-
taining them in the overall analyses was justified. A further limi-
tation is that schools with fewer than 10 students were excluded
from analyses. We used this strategy not only to be comparable
with the earlier Marsh and Hau (2003) study but also for valid
statistical reasons, as all estimates and standard deviations based
on small samples tend to be unreliable. Additionally, each school
in the PISA study had to sample at least 20 students (see OECD,
2005b), so the sample in schools with fewer than 10 students may
not be representative of the school as a whole. Hence, we feel that
our strategy was a reasonable course of action.
Implications and Future Directions for Theory, Practice,
and Research
By demonstrating that the BFLPE extends to economically
developing countries and collectivist countries, the current study
not only extends previous research, which has focused primarily
on developed and individualist nations, but also has important
theoretical implications for the BFLPE. In showing the BFLPE to
be evident in such a large sample of culturally and economically
diverse nations, the present investigation has demonstrated that the
BFLPE is generalizable and broadly applicable, thus placing it as
a universal theory, relevant to people everywhere.
Nevertheless, does it matter if the self-concepts of high-ability
students are reduced when they attend high-ability schools? Surely
if they are achieving to their potential, then it is of little conse-
quence if their self-concepts suffer. Two strands of research would
suggest that it does matter, as high-ability students may not be
achieving to their potential when they are segregated in high-
ability schools. The first strand of research concerns the REM.
When the current results are considered alongside previous results
from the REM, they suggest that segregating high-ability students
on the basis of academic ability may actually prevent them from
achieving their full academic potential. REM research, based on
sound longitudinal causal modeling studies along with results of
recent meta-analyses, demonstrates that self-concept and achieve-
ment share a mutually reinforcing causal relation, with increases in
one associated with increases in the other (Marsh & Craven, 2005,
2006; Marsh & Yeung, 1997a; Valentine & DuBois, 2005). Hence,
academic self-concept and achievement should be enhanced simul-
taneously to allow students to achieve their potential. However, as
evidenced by the current results, academic self-concepts suffer when
students attend high-ability schools, and according to the REM, their
performance must also suffer as a consequence. Accordingly, aca-
demically able students who are segregated on the basis of their
academic achievement may not be realizing their full potential.
The second strand of research concerns studies examining the
effects of ability tracking. Those in favor of tracking argue that not
only is it a valuable teaching approach to cater for different ability
levels, but it is also an effective means of raising achievement
levels (e.g., see Ireson & Hallam, 2001). However, in his compre-
hensive review of meta-analyses conducted on this subject, Hattie
(2002) concluded that ability tracking had very little effect on
achievement and that “the effects of tracking on self-esteem also
are near-zero overall” (p. 463; see also Marsh et al., 2008). In one
of the most comprehensive BFLPE studies in terms of diversity of
outcomes, Marsh (1991) evaluated the effects of school-average
ability on a set of academic outcomes intended to include most of
the important objectives of education, collected in Year 10, Year
12, and 2 years after graduation from high school as part of the
large, nationally (U.S.) representative High School and Beyond
study. After controlling for background and initial achievement,
the effects of school-average achievement were negative for al-
most all of the Year 10, Year 12, and postsecondary outcomes: Of
the 17 effects, 15 were significantly negative and 2 were nonsig-
nificant. There were no benefits associated with attending schools
where the average ability level of other students was high. School-
average achievement most negatively affected academic self-
concept (the BFLPE) and educational aspirations, but school-
average achievement also negatively affected general self-concept,
advanced coursework selection, school grades, academic effort,
standardized test scores, occupational aspirations, and subsequent
college attendance. The negative effects for educational aspirations
were clearly evident 2 years after graduation from high school. Con-
trolling for the negative effects of school-average achievement on
academic self-concept substantially reduced the size of negative ef-
fects on other outcomes, consistent with the proposal that these
negative effects of school-average ability were mediated—at least in
part— by academic self-concept. These results suggest that the neg-
ative effects of attending high-ability schools extend well beyond
those for academic self-concept that have been the focus of BFLPE
studies.
415
A PAN-HUMAN THEORY
Future research could build upon BFLPE theory by investigat-
ing the premise that social comparison processes are responsible
for the effect. Unfortunately, this was not possible in the current
investigation due to constraints of the database. Seaton et al.
(2008) made the first step in this investigation, although as this
study was a reanalysis of previously published data, it was limited
due to the lack of a standardized test and a psychometrically sound
measure of academic self-concept. Using both these measures,
future research should examine social comparison processes and
their relation with the BFLPE. Additionally, as suggested by
Marsh et al. (2008), BFLPE studies should examine microlevel
social comparison strategies that students use when they select an
individual classmate with whom to compare. Alternatively, social
comparison research should incorporate macrolevel social com-
parison strategies based on class-average information in addition
to the individual target that has been the focus of much social
comparison research. Moreover, a useful addition to both these
research methodologies would be to supplement quantitative re-
search with qualitative research with the intent of more fully
explicating the social comparison processes thought to underlie the
BFLPE.
Identifying individual differences between students could fur-
ther develop BFLPE theory. This knowledge would aid in devel-
oping policies to minimize the negative effects and maximize the
benefits of attending academically selective classes and schools.
However, to date, success in this area has been limited (e.g.,
Marsh, 1984, 1987, 1991; Marsh & Hau, 2003), as the sizes of
interactions between school-average ability and potential moder-
ators have been consistently small (or nonsignificant) and incon-
sistent across different studies (see Marsh et al., 2008). As such,
little is known about identifying strategies to counteract the effects
of the BFLPE for individual students. Hence, identifying con-
structs that may moderate the adverse effects of the BFLPE is an
important step for future research.
Methodological innovations are also important considerations
for future BFLPE research. In analyzing data, BFLPE research has
routinely used either structural equation modeling or multilevel
analyses. However, recent advances in statistical methodology
have been able to integrate both approaches, with the result that
multiple indicators of key constructs have been incorporated
within a multilevel framework. This was demonstrated by Marsh
and O’Mara (2008a), who used Mplus’s “complex design” option
in place of a multilevel model. They showed that multiple indica-
tors of both individual student and school-average variables could
be used while simultaneously allowing for the nested structure of
the data. The use of this type of statistical methodology is an
important direction for future BFLPE research (see Marsh et al.,
2008).
In summary, universally applicable theories have worldwide
implications. Many countries across the globe (e.g., Germany, the
Netherlands) espouse a system of academic selectivity whereby
high-ability students are taught in academically selective schools
and classes. Worldwide, educational policy documents typically
emphasize the importance of allowing every child to achieve his or
her full potential. For example, educational policy in Texas aims to
“ensure that all Texas children have access to a quality education
that enables them to achieve their potential and fully participate
now and in the future in the social, economic, and educational
opportunities of our state and nation” (Texas Education Code,
2007, Public Education Mission and Objectives, ¶ 1). Hence, these
results imply that the BFLPE is not solely a problem for econom-
ically developed and individualist societies. When academically
able students are segregated on the basis of their academic ability,
their self-worth suffers irrespective of their cultural heritage, their
ability level, or the economic development of their country. Con-
sequently, finding the means to overcome the negative effects of
the BFLPE is a truly global challenge.
References
Alwin, D. F., & Otto, L. B. (1977). High school context effects on
aspirations. Sociology of Education, 50, 259 –273.
Chanal, J. P., Marsh, H. W., Sarrazin, P. G., & Bois, J. E. (2005).
Big-fish-little-pond effects on gymnastics self-concept: Social compar-
ison processes in a physical setting. Journal of Sport & Exercise Psy-
chology, 27(1), 53–70.
Chen, X., Wasti, A., & Triandis, H. C. (2007). When does group norm or
group identity predict cooperation in a public goods dilemma? The
moderating effects of idiocentrism and allocentrism. International Jour-
nal of Intercultural Relations, 31, 259 –276.
Cohen, J. (1988). Statistical power analysis for the behavioral sciences
(2nd ed.). Hillsdale, NJ: Erlbaum.
Coleman, J. M., & Fults, B. A. (1985). Special class placement, level of
intelligence, and the self-concept of gifted children: A social comparison
perspective. Remedial and Special Education, 6, 7–11.
Craven, R. G., Marsh, H. W., & Print, M. (2000). Gifted, streamed and
mixed-ability programs for gifted students: Impact on self-concept,
motivation, and achievement. Australian Journal of Education, 44(1),
51–75.
Davis, J. (1966). The campus as a frogpond: An application of the theory
of relative deprivation to career decisions for college men. American
Journal of Sociology, 72, 17–31.
Festinger, L. (1954). A theory of social comparison processes. Human
Relations, 7, 117–140.
Fox, K. R., & Corbin, C. B. (1989). The Physical Self-Perception Profile:
Development and preliminary validation. Journal of Sport & Exercise
Psychology, 11(4), 408 – 430.
Guay, F., Larose, S., & Boivin, M. (2004). Academic self-concept and
educational attainment level: A ten-year longitudinal study. Self and
Identity, 3, 53– 68.
Guay, F., Marsh, H. W., & Boivin, M. (2003). Academic self-concept and
academic achievement: Developmental perspectives on their causal or-
dering. Journal of Educational Psychology, 95(1), 124 –136.
Ha, M., Marsh, H. W., & Halse, C. (2004, July). What you see is not
always. . . Paper presented at the SELF Research Biennial International
Conference, Berlin, Germany.
Hattie, J. (2002). Classroom composition and peer effects. International
Journal of Educational Research, 37, 449 – 481.
Helson, H. (1964). Adaptation-level theory. New York: Harper & Row.
Hofstede, G. (1980). Culture’s consequences: International differences in
work-related values. Beverly Hills, CA: Sage.
Hofstede, G., & Hofstede, G. J. (2005). Cultures and organizations:
Software of the mind. New York: McGraw-Hill.
Howell, D. C. (1997). Statistical methods for psychology. Belmont, CA:
Wadsworth.
Hox, J. (2002). Multilevel analysis: Techniques and applications. Mahwah,
NJ: Erlbaum.
Hyman, H. (1942). The psychology of subjective status. Psychological
Bulletin, 39, 473– 474.
Ireson, J., & Hallam, S. (2001). Ability grouping in education. London:
Paul Chapman.
Ireson, J., Hallam, S., & Plewis, I. (2001). Ability grouping in secondary
416 SEATON, MARSH, AND CRAVEN
schools: Effects on pupils’ self-concepts. British Journal of Educational
Psychology, 71, 315–326.
Keller, A., Ford, L. H., & Meacham, J. A. (1978). Dimensions in self-
concept in preschool children. Developmental Psychology, 14, 483– 489.
Marsh, H. W. (1974). Judgmental anchoring: Stimulus and response
variables. Unpublished doctoral dissertation, University of California,
Los Angeles.
Marsh, H. W. (1984). Self-concept, social comparison, and ability group-
ing: A reply to Kulik and Kulik. American Educational Research Jour-
nal, 21(4), 799 – 806.
Marsh, H. W. (1987). The big-fish-little-pond effect on academic self-
concept. Journal of Educational Psychology, 79(3), 280 –295.
Marsh, H. W. (1990). The structure of academic self-concept: The Marsh/
Shavelson model. Journal of Educational Psychology, 82(4), 623– 636.
Marsh, H. W. (1991). Failure of high-ability schools to deliver academic
benefits commensurate with their students’ ability levels. American
Educational Research Journal, 28(2), 445– 480.
Marsh, H. W. (1997). The measurement of physical self-concept: A con-
struct validation approach. In K. R. Fox (Ed.), The physical self: From
motivation to well-being (pp. 27–58).Champaign, IL: Human Kinetics.
Marsh, H. W. (2004). Negative effects of school-average achievement on
academic self-concept: A comparison of the big-fish-little-pond effect
across Australian states and territories. Australian Journal of Education,
48(1), 5–26.
Marsh, H. W. (2005). Big-fish-little-pond effect on academic self-concept.
German Journal of Educational Psychology, 19, 119 –127.
Marsh, H. W. (2007). Self-concept theory, measurement and research into
practice: The role of self-concept in educational psychology. Leicester,
England: British Psychological Society.
Marsh, H. W., Chessor, D., Craven, R., & Roche, L. (1995). The effects of
gifted and talented programs on academic self-concept: The big fish
strikes again. American Educational Research Journal, 32(2), 285–319.
Marsh, H. W., & Craven, R. G. (2002). The pivotal role of frames of
reference in academic self-concept formation: The big-fish-little-pond
effect. In F. Pajares & T. Urdan (Eds.), Adolescence and education (Vol.
2, pp. 83–123). Greenwich, CT: Information Age.
Marsh, H. W., & Craven, R. G. (2005). A reciprocal effects model of the
causal ordering of self-concept and achievement: New support for the
benefits of enhancing self-concept. In H. W. Marsh, R. G. Craven, & D.
McInerney (Eds.), International advances in self research (Vol. 2, pp.
15–52).Greenwich, CT: Information Age.
Marsh, H. W., & Craven, R. G. (2006). Reciprocal effects of self-concept
and performance from a multidimensional perspective: Beyond seduc-
tive pleasure and unidimensional perspectives. Perspectives on Psycho-
logical Science, 1, 133–163.
Marsh, H. W., & Hau, K. (2003). Big-fish-little-pond effect on academic
self-concept. A cross-cultural (26-country) test of the negative effects of
academically selective schools. American Psychologist, 58(5), 364–376.
Marsh, H. W., Hau, K., Artelt, C., Baumert, J., & Peschar, J. L. (2006).
OECD’s brief self-report measure of educational psychology’s most
useful affective constructs: Cross-cultural, psychometric comparisons
across 25 countries. International Journal of Testing, 6(4), 311–360.
Marsh, H. W., Ko¨ ller, O., & Baumert, J. (2001). Reunification of East and
West German school systems: Longitudinal multilevel modeling study
of the big-fish-little-pond effect on academic self-concept. American
Educational Research Journal, 38(2), 321–350.
Marsh, H. W., Kong, C. K., & Hau, K. (2000). Longitudinal multilevel
models of the big-fish-little-pond effect on academic self-concept:
Counterbalancing contrast and reflected glory effects in Hong Kong
schools. Journal of Personality and Social Psychology, 78, 337–349.
Marsh, H. W., & O’Mara, A. J. (2008a, July). Big-fish-little-pond-effect:
Total long-term negative effects of school-average ability on diverse edu-
cational outcomes over 8 adolescent/early adult years. Paper presented at the
29th International Congress of Psychology, Berlin, Germany.
Marsh, H. W., & O’Mara, A. J. (2008b). Self-concept is as multidisci-
plinary as it is multidimensional: A review of theory, measurement, and
practice in self-concept research. In H. W. Marsh, R. G. Craven, & D. M.
McInerney (Eds.). International advances in self research (Vol. 3, pp.
87–118). Charlotte, NC: Information Age.
Marsh, H. W., & Parker, J. W. (1984). Determinants of student self-
concept: Is it better to be a relatively large fish in a small pond even if
you don’t learn to swim as well? Journal of Personality and Social
Psychology, 47(1), 213–231.
Marsh, H. W., & Perry, C. (2005). Self-concept contributes to winning gold
medals: Causal ordering of self-concept and elite swimming perfor-
mance. Journal of Sport & Exercise Psychology, 27(1), 71–91.
Marsh, H. W., & Rowe, K. J. (1996). The negative effects of school-
average ability on academic self-concept: An application of multilevel
modelling. Australian Journal of Education, 40, 65– 87.
Marsh, H. W., Seaton, M., Trautwein, U., Lu¨ dtke, O., Hau, K. T., O’Mara,
A. J., & Craven, R. G. (2008). The big-fish-little-pond effect stands up
to critical scrutiny: Implications for theory, methodology, and future
research. Educational Psychology Review, 3, 319 –350.
Marsh, H. W., Trautwein, U., Lu¨ dtke, O., Baumert, J., & Ko¨ ller, O. (2007).
Big fish little pond effect: Persistent negative effects of selective high
schools on self-concept after graduation. American Educational Re-
search Journal, 44, 631– 669.
Marsh, H. W., Trautwein, U., Lu¨ dtke, O., Ko¨ ller, O., & Baumert, J. (2006).
Integration of multidimensional self-concept and core personality con-
structs: Construct validation and relations to well-being and achieve-
ment. Journal of Personality, 74(2), 403– 456.
Marsh, H. W., & Yeung, A. S. (1997a). Causal effects of academic
self-concept on academic achievement: Structural equation models of
longitudinal data. Journal of Educational Psychology, 89(1), 41–54.
Marsh, H. W., & Yeung, A. S. (1997b). Coursework selection: Relations to
academic self-concept and achievement. American Educational Re-
search Journal, 34, 691–720.
Matsumoto, D. (2002). Culture, psychology, and education. Online read-
ings in psychology and culture. Retrieved June, 23, 2005, from http://
www.wwu.edu/⬃culture
McFarland, C., & Buehler, R. (1995). Collective self-esteem as a moder-
ator of the frog-pond effect in reactions to performance feedback.
Journal of Personality and Social Psychology, 68(6), 1055–1070.
Meyer, J. W. (1970). High school effects on college intentions. American
Journal of Sociology, 76, 59 –70.
Ministerial Council on Education, Employment, Training and Youth Af-
fairs. (1999). The Adelaide declaration on national goals for schooling
in the twenty-first century. Retrieved September 12, 2007, from http://
www.dest.gov.au/sectors/school_education/policy_initiatives_reviews/
national_goals_for_schooling_in_the_twenty_first_century.htm
Morse, S., & Gergen, K. J. (1970). Social comparison, self-consistency,
and the concept of self. Journal of Personality and Social Psychology,
16, 148 –156.
Mulkey, L. M., Catsambis, S., Steelman, L. C., & Crain, R. L. (2005). The
long-term effects of ability grouping in mathematics: A national inves-
tigation. Social Psychology of Education, 8, 137–177.
Organisation for Economic Co-operation and Development. (2001). PISA
international database (Technical Report). Paris: Author.
Organisation for Economic Co-operation and Development. (2003). Stu-
dent engagement at school: A sense of belonging and participation.
Paris: Author.
Organisation for Economic Co-operation and Development. (2005a). PISA
2000 data analysis manual: SPSS users. Paris: Author.
Organisation for Economic Co-operation and Development. (2005b). PISA
2003 technical report. Paris: Author.
Parducci, A. (1995). Happiness, pleasure, and judgment: The contextual
theory and its applications. Mahwah, NJ: Erlbaum.
Rasbash, J., Browne, W., Goldstein, H., Yang, M., Plewis, I., Healy, M., et
417
A PAN-HUMAN THEORY
al. (2002). A user’s guide to MLwiN (Version 2.1d).London: Centre for
Multilevel Modelling, Institute of Education, University of London.
Raudenbush, S. W., & Bryk, A. S. (2002). Hierarchical linear models: Application
and data analysis methods (2nd ed.). Thousand Oaks, CA: Sage.
Raudenbush, S. W., Bryk, A. S., & Congdon, R. (2005). HLM6 for
Windows. Lincolnwood, IL: Scientific Software International.
Reuman, D. A. (1989). How social comparison mediates the relation between
ability-grouping practices and students’ achievement expectancies in math-
ematics. Journal of Educational Psychology, 81(2), 178 –189.
Rogers, C. R. (1951). Client-centered therapy. Boston: Houghton Mifflin.
Rowe, K. (2005). Practical multilevel analysis with MLwiN & LISREL.
Camberwell: Australian Council for Educational Research.
Schwartz, S. H., & Bilsky, W. (1990). Toward a theory of the universal
content and structure of values: Extensions and cross-cultural replica-
tions. Journal of Personality and Social Psychology, 58(5), 878 – 891.
Seaton, M. (2007). The big-fish-little-pond effect under the grill: Tests of its
universality, a search for moderators, and the role of social comparison.
Unpublished doctoral dissertation, University of Western Sydney, Australia.
Seaton, M., Marsh, H. W., Dumas, F., Huguet, P., Monteil, J.-M., Re´ gner,
I., et al. (2008). In search of the big fish: Investigating the coexistence
of the big-fish-little-pond effect with the positive effects of upward
comparisons. British Journal of Social Psychology, 47, 73–103.
Segall, M. H., Lonner, W. J., & Berry, J. W. (1998). Cross-cultural
psychology as a scholarly discipline: On the flowering of culture in
behavioral research. American Psychologist, 53, 1101–1110.
Shavelson, R. J., Hubner, J. J., & Stanton, G. C. (1976). Self-concept:
Validation of construct interpretations. Review of Educational Research,
46(3), 407– 441.
Stouffer, S. A., Suchman, E. A., DeVinney, L. C., Star, S. A., & Williams,
R. M. (1949). The American soldier: Adjustments during army life (Vol.
1). Princeton, NJ: Princeton University Press.
Texas Education Code. (2007). Retrieved September 12, 2007, from http://
tlo2.tlc.state.tx.us/statutes/docs/ED/content/htm/ed.002.00.000004
.00.htm#4.001.00
Trautwein, U., Gerlach, E., & Lu¨ dtke, O. (in press). Athletic classmates,
physical self-concept, and free-time physical activity: A longitudinal
study of frame of reference effects. Journal of Educational Psychology.
Triandis, H. C., Bontempo, R., Villareal, M. J., Asai, M., & Lucca, N.
(1988). Individualism and collectivism: Cross-cultural perspectives on
self-ingroup relationships. Journal of Personality and Social Psychol-
ogy, 54(2), 323–338.
Tymms, P. (2001). A test of the big fish in a little pond hypothesis: An
investigation into the feelings of seven-year-old pupils in school. School
Effectiveness and School Improvement, 12, 161–181.
Tymms, P. (2004). Effect sizes in multilevel models. In I. Schagen & K.
Elliot (Eds.), But what does it mean? The use of effect sizes in educa-
tional research (pp. 55– 66). London: National Foundation for Educa-
tional Research.
Upshaw, H. S. (1969). The personal reference scale: An approach to social
judgment. In L. Berkowitz (Ed.), Advances in experimental social psy-
chology, 4, 315–370.
Valentine, J. C., & DuBois, D. L. (2005). Effects of self-beliefs on
academic achievement and vice versa. In H. W. Marsh, R. G. Craven, &
D. McInerney (Eds.), The new frontiers of self research (pp. 53–77).
Greenwich, CT: Information Age.
Wedell, D. H., & Parducci, A. (2000). Social comparison: Lessons from
basic research on judgment. In J. Suls & L. Wheeler (Eds.), Handbook
of social comparison: Theory and research. The Plenum series in
social/clinical psychology (pp. 223–252). Dordrecht, the Netherlands:
Kluwer Academic.
World Bank. (2007). World development indicators. Retrieved January 12,
2007, from http://goworldbank.org/K2CKM78CCO
Wu, M. L., & Adams, R. J. (2002, April). Plausible values: Why are they
important? Paper presented at the International Objective Measurement
Workshop, New Orleans, LA.
Zeidner, M., & Schleyer, J. (1998). The big-fish-little-pond effect for
academic self-concept, test anxiety, and school grades in gifted children.
Contemporary Educational Psychology, 24, 305–329.
Appendix
Equations Used in Models
1. To test the generalizability of the BFLPE across the entire
cross-national sample, the following equation was used:
Math self-concept ⫽0ijkConstant ⫹1jklinear ability
⫹2jkquadratic ability ⫹3kschool-average ability
⫹4linear ability ⫻school-average ability interaction
⫹5quadratic ability ⫻school-average ability interaction
⫹v0k⫹v1k⫹v2k⫹v3k⫹u0jk ⫹u1jk ⫹u2jk ⫹e0ijk.
2. To test the generalizability of the BFLPE in 41 individual
countries, the following equation was used:
Math self-concept ⫽
0ijkConstant ⫹
1linear ability
⫹
2quadratic ability ⫹
3school-average ability
⫹v0k⫹u0jk ⫹e0ijk.
3. To test the moderating effect of cultural orientation and
economic development, the following equations were used:
Model 1: The base model with no predictors:
Math self-concept ⫽
0ijkConstant ⫹v0k⫹u0jk ⫹e0ijk.
Model 2: Linear ability, quadratic ability, and school-average
ability added as fixed predictors:
Math self-concept ⫽
0ijkConstant ⫹
1linear ability
⫹
2quadratic ability ⫹
3school-average ability
⫹v0k⫹u0jk ⫹e0ijk.
Model 3: The effect of the predictors was allowed to vary across
countries:
Math self-concept ⫽
0ijkConstant ⫹
1linear ability
⫹
2quadratic ability ⫹
3school-average ability
⫹v0k⫹v1k⫹v2k⫹v3k⫹u0jk ⫹e0ijk.
Model 4: Individualism– collectivism was added and left as a
fixed effect:
418 SEATON, MARSH, AND CRAVEN
Math self-concept ⫽
0ijkConstant ⫹
1linear ability
⫹
2quadratic ability ⫹
3school-average ability
⫹
4individualism– collectivism ⫹v0k⫹v1k
⫹v2k⫹v3k⫹u0jk ⫹e0ijk.
Model 5: The interaction of school-average ability and
individualism– collectivism was added:
Math self-concept ⫽
0ijkConstant ⫹
1linear ability
⫹
2quadratic ability ⫹
3school-average ability
⫹
4individualism– collectivism ⫹5individualism–
collectivism ⫻school-average ability interaction
⫹v0k⫹v1k⫹v2k⫹v3k⫹u0jk ⫹e0ijk.
Model 6: Economic development was added as a fixed effect:
Math self-concept ⫽
0ijkConstant ⫹
1linear ability
⫹
2quadratic ability ⫹
3school-average ability
⫹
4economic development ⫹v0k⫹v1k⫹v2k
⫹v3k⫹u0jk ⫹e0ijk.
Model 7: The interaction between school-average ability and
economic development was added:
Math self-concept ⫽
0ijkConstant ⫹
1linear ability
⫹
2quadratic ability ⫹
3school-average ability
⫹
4economic development ⫹5economic development
⫻school-average interaction ⫹v0k⫹v1k⫹v2k
⫹v3k⫹u0jk ⫹e0ijk.
Received November 27, 2007
Revision received June 10, 2008
Accepted June 11, 2008 䡲
Correction to Naumann et al. (2007)
In the article “Signaling In Expository Hypertexts Compensates for Deficits In Reading Skill,” by
Johannes Naumann, Tobias Richter, Ju¨rgen Flender, Ursula Christmann, and Norbert Groeben
(Journal of Education Psychology, 2007, Vol. 99, No. 4, pp. 791– 807), the URL published for the
supplemental material was incorrect. The correct URL is http://dx.doi.org/10.1037/
0022-0663.99.4.791.supp
DOI: 10.1037/a0015587
419
A PAN-HUMAN THEORY
A preview of this full-text is provided by American Psychological Association.
Content available from Journal of Educational Psychology
This content is subject to copyright. Terms and conditions apply.
