Reciprocal Effects Between Academic
Self-Concept, Self-Esteem, Achievement,
and Attainment Over Seven Adolescent
Years: Unidimensional and Multidimensional
Perspectives of Self-Concept
Herbert W. Marsh
Alison O’Mara
University of Oxford
2005; Hunter & Csikszentmihalyi, 2003; Marsh &
Craven, 2006). In a potentially serious threat to this
positive psychology movement, Baumeister, Campbell,
Krueger, and Vohs (2003, 2005) challenged the prevailing optimistic perspective of the value of positive selfbeliefs in a highly influential review commissioned for
Psychological Science in the Public Interest. They posed
the question, “Does high self-esteem cause better performance, interpersonal success, happiness, or healthier
lifestyles?” Arguing for a negative response to their
question, Baumeister et al. (2003) concluded that “selfesteem per se is not the social panacea that many people
hoped it was” (p. 38), a point reiterated by Baumeister
et al. (2005) in their article in Scientific American when
they concluded “that efforts to boost people’s selfesteem are of little value in fostering academic achievement or preventing undesirable behaviour” (p. 84).
Because of the strength of these conclusions and the
prestige of the journals in which they appeared, this
might seem to be the definitive word for mainstream
psychology on this construct that has been so central in
the development of psychology from the time of
William James. However, as noted by Baumeister et al.
(2003, see p. 7), their conclusions apply only to global
Authors’ Note: Requests for further information about this investigation should be sent to Professor Herbert W. Marsh, Department of
Education, University of Oxford, 15 Norham Gardens, Oxford, OX2
6PY, UK; e-mail: This email address is being protected from spambots. You need JavaScript enabled to view it..
PSPB, Vol. 34 No. 4, April 2008 542-552
DOI: 10.1177/0146167207312313
© 2008 by the Society for Personality and Social Psychology, Inc.
In their influential review, Baumeister, Campbell,
Krueger, and Vohs (2003) concluded that self-esteem—
the global component of self-concept—has no effect on
subsequent academic performance. In contrast, Marsh
and Craven’s (2006) review of reciprocal effects models
from an explicitly multidimensional perspective demonstrated that academic self-concept and achievement are
both a cause and an effect of each other. Ironically, both
reviews cited classic Youth in Transition studies in support
of their respective claims. In definitive tests of these counter
claims, the authors reanalyze these data—including self-esteem
(emphasized by Baumeister et al.), academic self-concept
(emphasized by Marsh & Craven), and postsecondary educational attainment—using stronger statistical methods
based on five waves of data (grade 10 through 5 years after
graduation; N = 2,213). Integrating apparently discrepant
findings under a common theoretical framework based
on a multidimensional perspective, academic self-concept
had consistent reciprocal effects with both achievement
and educational attainment, whereas self-esteem had
almost none.
Keywords: self-concept; self-esteem; reciprocal effects model;
structural equation modeling
There is a revolution sweeping psychology, one that
emphasizes a positive psychology focusing on how
healthy, normal, and exceptional individuals can get the
most from life (e.g., Fredrickson, 2006; Lopez et al.,
2006; Seligman & Csikszentmihalyi, 2000). Positive
self-beliefs are at the heart of this revolution (Furr,
Marsh, O’Mara / SELF-ESTEEM, SELF-CONCEPT, AND PERFORMANCE 543
self-esteem and not to specific components of self-concept.
Emphasizing the importance of this distinction, we
demonstrate that Baumeister et al.’s conclusions need
not sound the death knell for the relevance of self-beliefs
to achievement if self-concept is appropriately considered from a multidimensional perspective. Indeed, there
is convincing evidence for the consistent positive effects
of academic self-concept on subsequent achievement
after controlling the effects of prior achievement (e.g.,
Byrne, 1996; Marsh & Craven, 2006; Valentine &
DuBois, 2005; Valentine, DuBois, & Cooper, 2004).
Marsh and Craven argued that conclusions drawn by
Baumeister and colleagues were based largely on research
studies, statistical methodology, and theoretical conceptualizations of self-concept that are no longer current.
Here, we provide an empirical test of a theoretical
model that integrates both of these apparently contradictory conclusions.
There were important areas of agreement between
Baumeister et al. (2003; see also Baumeister et al., 2005)
and Marsh and Craven (2006) on appropriate methodology. In particular, all parties agreed that correlations
based on a single wave of data cannot be used to infer causation and the need for longitudinal panel designs (as in
the reciprocal effects model outlined by Marsh & Craven,
1997, 2006), in which achievement and self-beliefs are
each measured on at least two different occasions. Noting
the strength and appropriateness of this design,
Baumeister et al. (2003) added the caveat,
Insisting that self-esteem [at Time 1] must predict achievement at Time 2 after controlling for achievement at Time
1 could obscure some actual causal relationships, so it
should be regarded as a highly conservative way of testing
the hypothesis . . . one may be throwing a very large baby
out with the statistical bathwater. (p. 9)
Despite such areas of agreement, there were key areas
of disagreement between the two sets of reviews in terms
of the following:
a. Use of current research: Baumeister et al. (2003) only
considered publications from before 1990, whereas
Marsh and Craven mostly considered studies from the
past 10 years;
b. Research methodology: Research reviewed by Baumeister
et al. (2003) was largely based on multiple regression
that was typical of research of that earlier era, whereas
Marsh and Craven (2006) focused on studies that used
structural equation models (SEM) based on multiple
indicators;
c. Unidimensional versus multidimensional perspective:
Baumeister et al. (2003) focused on an implicit unidimensional perspective of self-concept through their sole
reliance on self-esteem—the global component of
multidimensional, hierarchical models of self-concept
(see Marsh, 1993; Shavelson, Hubner, & Stanton,
1976). Marsh and Craven (2006) took an explicitly multidimensional perspective based on multiple, relatively
distinct components of self-concept.

CONSTRUCT DEFINITION OF
SELF-CONCEPT: A MULTIDIMENSIONAL,
HIERARCHICAL CONSTRUCT
In their classic review of self-concept research, theory,
and measurement, Shavelson et al. (1976) developed a
multidimensional, hierarchical model of self-concept that
fundamentally impacted self-concept research (Marsh &
Hattie, 1996). Self-concept, broadly defined by Shavelson
et al., is a person’s self-perceptions formed through experience with and interpretations of one’s environment.
These self-perceptions are influenced especially by evaluations by significant others, reinforcements, and attributions for one’s own behavior. Self-concept is not an entity
within the person but a hypothetical construct that is
potentially useful in explaining and predicting how a
person acts. Shavelson et al. noted that self-concept is
important both as an outcome and as a mediating variable
that helps to explain other outcomes. Self-perceptions
influence the way one acts, and behaviors in turn influence
one’s self-perceptions.
Although some researchers reserve the term selfesteem for the evaluative component of self-perception
and use the term self-concept for descriptive components of self-perception, Shavelson et al. (1976) argue
that self-concept is both descriptive and evaluative (also
see Marsh, 1993; Marsh & Craven, 2006; Swann,
Chang-Schneider, & Larsen McClarty, 2007). Thus, for
example, statements such as “I am good at mathematics,” “I can run a long way without stopping,” and “I
am good looking” all have both evaluative and descriptive components and clearly reflect specific components
of self-concept (math, physical and appearance self-
|concepts, respectively). More recently, Swann et al. also
concluded that there is little basis for such evaluativedescriptive and cognitive-affective distinctions, as both
self-esteem and self-concepts involve cognitive and
affective components, noting that researchers in related
fields (attitude researchers, interpersonal expectancy
researchers, trait theorists) have not found this to be a
useful distinction.
In support of the conceptualization of self-esteem as
a global component of self-concept, Rosenberg (1979)
maintained that the self-esteem construct based on
instruments such as the Rosenberg Scale emphasizes an
overarching, general, or global construct that at least
implicitly incorporates many (or all) specific components according to the subjective weight put on each by
the respondent (also see Marsh, 1986). In higher order
factor analysis studies (see Marsh & Hattie, 1996),
global measures of self-esteem consistently correlated
about .95 with the highest-order factor (representing
the apex of Shavelson et al.’s hierarchical model) based
on specific components of self-concept. It is important
to note that multidimensional self-concept theorists
(e.g., Marsh & Craven, 2006) do not deny the existence
of an overarching, global self-concept or self-esteem.
Indeed, multidimensional hierarchical models of selfconcept integrate specific and global self-esteem dimensions of self-concept, such that global self-esteem is a
component of the multidimensional self-concept structure (Marsh, 1993; Marsh, Craven, & Martin, 2006).
Hence, for the purposes of this study, we use the term
self-esteem to refer to the global component of selfconcept and to distinguish between this and specific
components of self-concept (e.g., physical, social, and
academic).
Factor analytic research (e.g., Marsh, Byrne, &
Shavelson, 1988; Marsh & Hattie, 1996) showed that
the hierarchical aspect of the multidimensional, hierarchical model proposed by Shavelson et al. (1976)
was much weaker than originally hypothesized. In
particular, specific components of self-concept were
more differentiated and less highly correlated with
each other than anticipated, so that much of the variance in domain specific factors of self-concept could
not be explained in terms of higher order self-concept
factors or self-esteem. Thus, for example, factor analysis of adolescent responses to a recent adaptation of
the multidimensional Self-Definition Questionnaire III
clearly supported the 17 self-concept factors that the
instrument was designed to measure (Marsh, Trautwein,
Lüdtke, Köller, & Baumert, 2006). Furthermore,
the average correlation among the 17 self-concept
factors—even after controlling for unreliability—was
only .14, thus verifying the distinctiveness of these
dimensions.
Consistent with a large amount of subsequent
research in support of the model by Shavelson et al.
(1976), Marsh and Craven (1997, 2006) argued for the
importance of a multidimensional perspective of selfconcept. They suggested that researchers should focus
on specific components of self-concept most logically
related to the goals of their particular research instead
of, or in addition to, a global measure of self-esteem.
The distinctive nature of the various domains of selfconcept means that global self-esteem scales may conceal or distort the multiplicity of self-beliefs (Harter,
1999; Hattie, 1992). For example, if a child has a low
reading self-concept and a high math self-concept, then
a measure of global self-esteem is not useful as a diagnostic tool. This has ramifications for self-concept intervention research, as the underuse of multidimensional
instruments may lead researchers to underestimate the
effectiveness of their intervention (Bracken, 1996;
Craven, Marsh, & Burnett, 2003; O’Mara, Marsh,
Craven, & Debus, 2006).
Swann et al. (2007) made a related point about the
predictive power of content-specific components of selfconcept compared to the more global component of selfesteem. Drawing on historical parallels from attitude
research, personality research, and trait-theory research,
they emphasized the importance of the specificity
matching principle, that specific predictors should be
used to predict specific behaviors and general predictors
should be used to predict general behaviors. More
specifically, they concluded, “From the perspective of
the specificity matching principle, then, recent reviews
of the self-esteem literature (e.g., Baumeister et al.,
2003) have violated the specificity matching principle
by focusing on the capacity of global measures of selfesteem to predict specific outcomes (e.g., Does selfesteem predict grades in a math class?)” (p. 87). They
reviewed evidence from different areas of research to
support the contention that predictive validity is maximized when predictors and outcomes referred to the
same conceptual variable.
Particularly in educational psychological research,
Marsh and Craven (2006; also see Byrne, 1996; Marsh,
1993) reviewed a large body of research showing that
diverse academic outcomes were systematically related to
academic self-concept but nearly unrelated (or even negatively related) to global self-esteem and other nonacademic
components of self-concept. This extreme multidimensionality was highlighted by Marsh et al. (2006), who showed
that nine academic outcomes (standardized test scores,
school grades, and course work selection in different school
subjects) were systematically related to corresponding
academic self-concepts. For example, math self-concept
for their late-adolescent sample was substantially related
to math school grades (.71), math standardized achievement test scores (.59), and advanced math courses (.51).
In contrast, the academic outcomes were nearly unrelated
to global self-esteem (rs ranging from –.03 to .05) as well
as nine other nonacademic specific domains of self-concept. While support for this multidimensional perspective
of self-concept was particularly strong in educational
research, Marsh and Craven (1997, 2006) reviewed other
research demonstrating its importance in social, personality, developmental, and sport psychology, as well as elite
sport, mental health, and gender studies. Hence, Marsh and
Craven (1997) argued, “If the role of self-concept research
is to better understand the complexity of self in different
contexts, to predict a wide variety of behaviors, to provide outcome measures for diverse interventions, and to
relate self-concept to other constructs, then the specific
domains of self-concept are more useful than a general
domain” (p. 191).
THE PRESENT INVESTIGATION
For present purposes, the most important finding of the
Marsh and Craven (2006) review was the consistent support
for the reciprocal effects model predictions based on
544 PERSONALITY AND SOCIAL PSYCHOLOGY BULLETIN
Marsh, O’Mara / SELF-ESTEEM, SELF-CONCEPT, AND PERFORMANCE 545
longitudinal panel studies showing that academic self-concept
and achievement were mutually reinforcing constructs, each
having an impact on the other. Indeed, they concluded that
clear support for the reciprocal effects models based on academic self-concept, coupled with the clear lack of support
for the reciprocal effects models based on self-esteem, provided good evidence for the discriminant validity of academic self-concept in relation to self-esteem and for their
multidimensional perspective. Whereas the Marsh and
Craven review focused primarily on their own research
program, Valentine’s (Valentine & DuBois, 2005; Valentine
et al., 2004) comprehensive meta-analysis found consistent
support for reciprocal effects between academic self-beliefs
and achievement but little or no reciprocal effects based on
self-esteem (also see Trautwein, Lüdtke, Köller, & Baumert,
2006). Because the substantial body of support for the
reciprocal effects model focused primarily on academic selfconcept, it was ignored in the Baumeister et al. (2003, 2005)
reviews that focused on global self-esteem.
Indeed, a critical limitation in reviews by both
Baumeister et al. (2003, 2005) and by Marsh and
Craven, 1997, 2006)—as well as research reviewed by
Valentine et al. (2004)—was the dearth of studies available to compare and contrast the effects of academic
self-concept and self-esteem based on appropriate SEM
models that included both constructs. Furthermore,
even though the Baumeister et al. (2003, 2005) and
Marsh and Craven (1997, 2006) reviews addressed
basically the same question about the effects of selfbeliefs on subsequent academic performance, there was
almost no overlap in the studies that they drew upon to
support their apparently contradictory conclusions.
Ironically, one exception was that both of these reviews
emphasized publications based on the classic Youth in
Transition study as critical in support of their apparently conflicting conclusions. Marsh and Craven
pointed to Marsh’s (1990) Youth in Transition study of
academic self-concept as the classic reciprocal effects
model study—the first to propose the reciprocal effects
model, superseding earlier self-enhancement and skill
development models, and supporting reciprocal effects
model predictions using an appropriate design and statistical procedures (also see Byrne, 1996).
Baumeister et al. (2003) commented that the Bachman
and O’Malley (1977) study was an “early and still well
respected study” (p. 11) based on the large, nationally
representative Youth in Transition database. Baumeister
et al. interpreted their results to mean that “Although
Bachman and O’Malley found that self-esteem correlated
with school performance, their more sophisticated statistical tests (i.e., path analyses) did not point to any causal
role for self-esteem” (p. 11). However, a careful reading
of the original Bachman and O’Malley (1977) study
shows that, whereas prior school performance did have a
moderate effect on subsequent self-esteem, the authors
did not actually test effects of prior self-esteem to subsequent high school performance. Indeed, their main focus
was on postsecondary educational attainment rather than
school performance per se, showing that self-esteem during high school had little positive, direct effect on educational attainment 5 years after graduation from high
school. Hence, although the Youth in Transition data are
clearly appropriate for testing a reciprocal effects model
based on self-esteem and school performance, this model
was not actually tested by Bachman and O’Malley.
Furthermore, although subsequent educational attainment is a critical outcome variable for testing the longterm effects of academic self-concept, Marsh (1990) did
not include attainment in his Youth in Transition study.
While the two different studies based on the same Youth
in Transition data yielded apparently contradictory conclusions about the effects of self-beliefs, Marsh and Craven
(2006) argued that the findings were not inconsistent:
The juxtaposition of the two seemingly contradictory sets
of conclusions based on analyses of the same data [the
Bachman & O’Malley, 1977, and the Marsh, 1990, studies] reinforces the importance of considering academic selfconcept measures in causal-ordering studies of school
performance and the need to account for the multidimensionality of the self-concept construct. (p. 150)
Given the critical role of the Youth in Transition data in
this debate, we extended previous Youth in Transition studies to include academic self-concept (emphasized by Marsh,
1990; Marsh & Craven, 2006), self-esteem (emphasized
by Baumeister et al., 2003, 2005), and educational attainment (emphasized by Bachman & O’Malley, 1977), using
current statistical procedures that provide a definitive test to
the apparently contradictory conclusions about self-beliefs
and their integration into the multidimensional perspective
proposed by Marsh and Craven (2006). Consistent with
the reciprocal effects model and the multidimensional perspective, we predict that (a) there will be consistently positive links relating prior academic self-concept to subsequent
school achievement (school grades) and educational attainment, and relating prior achievement and attainment to
subsequent academic self-concept; and (b) links relating
self-esteem with school grades and attainment will be consistently smaller (or nonsignificant) than corresponding
links involving academic self-concept.
METHOD
Database, Variables, and Model
The Youth in Transition database is a longitudinal,
large, nationally representative database of 10th grade
boys in U.S. public high schools in 1966; for more
detailed description of the database and the variables
546 PERSONALITY AND SOCIAL PSYCHOLOGY BULLETIN
used here, see Bachman, 2002 (also see Bachman &
O’Malley, 1977; Marsh, 1990; appendix). Five waves
of data were collected between 1966 and 1974: Time 1
(T1, early 10th grade; N = 2,213), Time 2 (T2, late 11th
grade; N = 1,886), Time 3 (T3, late 12th grade; N =
1,799), Time 4 (T4, 1 year after graduation; N = 1,620),
and Time 5 (T5, 5 years after graduation; N = 1,608).
Following earlier Youth in Transition studies (Bachman
& O’Malley, 1977; Marsh, 1990), we considered 15
latent constructs based on 73 indicators: socioeconomic
status (SES; T1, 6 indicators); academic ability (T1, 4
standardized test scores); academic self-concept (T1,
T2, and T3 based on 3, 3, and 2 items, respectively);
global self-esteem (T1, T2, T3, T4, T5, 10 items each);
school grades (GPA; T1, T2, T3, 1 item each); and prior
educational attainment (T4, T5, highest level of education completed; Bachman & O’Malley, 1977).
Statistical Analyses
The initial a priori model (see Figure 1) was primarily based on the temporal ordering of the data collection
(i.e., T1 variables precede T2 variables). Control variables (ability and SES) came first in the posited causal
ordering, followed by school grades and then academic
self-concept and global self-esteem (school grades came
before academic self-concept and self-esteem, because
students reported their grades from the previous year).
Academic self-concept and self-esteem were posited to
be correlated with no causal ordering within a single
wave of data. Educational attainment at T4 followed
T3 variables, from before the end of high school, but
proceeded T4 academic self-concept and T4 self-esteem.
Statistical analyses were SEMs using the complex
modeling procedure in MPLUS (Version 4.0; Muthén &
Muthén, 2006) that takes into account the nonindependence of the scores for students from the same school—
the clustering effect of students nested within schools.
We used the robust maximum likelihood estimator with
maximum likelihood parameter estimates, standard
errors, and a chi-square test statistic that are robust to
nonnormality and nonindependence of observations
and full information estimation for missing data
(Muthén & Muthén, 2006). We considered a very
restrictive measurement model in which each indicator
was allowed to load on only the latent variable it was
designed to measure.
Figure 1 A structural equation model of the reciprocal relations between academic self-concept, global self-esteem, academic achievement,
and educational attainment.
NOTE: A structural equation model of the reciprocal relations between academic self-concept (ASC), global self-esteem (Estm), academic achievement (GPA), and educational attainment (EdAt), controlling for initial academic ability and socioeconomic status (SES) across five data collection waves (T1-T3, last 3 years of high school; T4-T5, 1 and 5 years postsecondary). Path coefficients not relevant to the reciprocal effects model
or not significant are excluded for purposes of clarity (but see the Appendix for the presentation of all parameter estimates). Critical significant
paths representing reciprocal effects of ASC with GPA and attainment are represented with bold black lines (and path coefficients in black boxes
with gray shading). Critical significant paths representing reciprocal effects of self-esteem with GPA and attainment are represented with bold
gray lines (and white path coefficients in solid black boxes). Despite a significant chi square (χ2
(2318) = 4893.9, p < .001) due in part to the large
sample size, goodness of fit indices provided good support for the ability of the a priori model to fit the data: root mean square error of approximation = .022; Tucker-Lewis Index = .93; confirmatory fit index = .94; and standardized root mean square residual = .046.
Marsh, O’Mara / SELF-ESTEEM, SELF-CONCEPT, AND PERFORMANCE 547
Error Structure
School grades and educational attainment measures
were each based on one indicator, so that it was not
possible to estimate their reliabilities (and correct for
measurement error) as part of the analysis. Consistent
with Marsh (1990) and more general recommendations
(e.g., Jöreskog, 1979), we constrained the standardized
uniqueness for each of these measures to be .10 (reflecting a conservative estimate of reliability of .90). Also,
consistent with analyses by Marsh (1990) and recommendations for longitudinal panel data more generally
(Jöreskog, 1979; Marsh & Hau, 1996), correlated
residuals were posited a priori between matching indicators of the same construct administered on different
occasions (e.g., responses to the first self-esteem item
administered at T1-T5). Failure to control this error
structure would result in positively biased estimates of
stability over time and distort parameter estimates
based on the reciprocal effects model (see Marsh &
Hau, 1996). Responses to the many variations of the
Rosenberg self-esteem instrument do not have a simple
unidimensional structure. Consistent with recommendations and empirical results by Marsh (1996), the original design of the scale, and the way it was used by
Bachman and O’Malley (1977), we posited one substantive global self-esteem factor based on all (positively and
negatively worded items) and a negative item–method
effect represented by correlated uniquenesses among
responses to just the negatively worded items administered
within the same data collection wave. Although the a
priori error structure posited in this study is complex, it
is important to emphasize that all correlated uniquenesses were posited a priori, following from previous
empirical research and theory.
RESULTS AND DISCUSSION
Relations Between Academic Self-Concept and
Self-Esteem to School Grades
In terms of testing the reciprocal effects model, the
most important predictions are the reciprocal effects
relating academic self-concept and school grades.
Critical paths (bold black arrows with path coefficients
in black boxes with gray shading, Figure 1; also see
appendix) are those leading from prior school grades to
subsequent academic self-concept and those leading
from prior academic self-concept to subsequent school
grades (controlling for SES, ability, and prior school
grades). All of these paths are statistically significant, at
least moderate in size (.14 to .36), and highly consistent
across the different waves. These results replicate the
findings of the Marsh (1990) study within the context of
this larger model and stronger statistical methodology.
Results based on the corresponding paths relating
self-esteem and school grades (GPA; bold gray arrows
with white coefficients in black boxes, Figure 1; also see
appendix) are all small; only the path from T2 selfesteem to T3 school grades (.07) is marginally significant, and no paths from school grades to self-esteem are
significant. Whereas the results provide some minimal
support for the effect of prior self-esteem on subsequent
school grades, the reciprocal effects involving academic
self-concept are clearly stronger and more consistent.
Even though academic self-concept and self-esteem are
moderately correlated (rs = .44 to .46 for different waves;
see appendix), academic self-concept shows a strong
pattern of reciprocal effects with subsequent achievement, whereas self-esteem shows almost none.
Relations Between Academic Self-Concept
and Self-Esteem to Attainment
The most important new feature of the present investigation is the inclusion of educational attainment collected 1 (T4) and 5 years (T5) after the normal
graduation from high school. Both T1 academic selfconcept and T2 academic self-concept (recalling that
there is no T3 academic self-concept) are significant predictors of T4 attainment, T4 attainment is a significant
predictor of T4 academic self-concept, and T4 academic
self-concept is a significant predictor of T5 attainment.
Hence, there is a clear pattern of reciprocal effects relating academic self-concept and attainment. Particularly
given that T4 attainment is, not surprisingly, the best
predictor of T5 attainment and that there is a 4-year
gap between T4 and T5, the contribution of T4 academic self-concept (.22) is strong—clearly larger than any
other predictors of T5 attainment (school grades, ability, SES, self-esteem). Hence, T4 academic self-concept
is a significant predictor of subsequent growth in educational attainment between 1 and 5 years after the normal graduation of high school.
In contrast to the consistent pattern of relations
between academic self-concept and attainment, relations involving self-esteem are small and mostly nonsignificant. Whereas T1 self-esteem has a small (.08)
statistically significant effect on T4 attainment, T2 and
T3 self-esteem do not. T4 self-esteem has no significant
effect on T5 attainment. It is interesting that T3 selfesteem had a small (.08) statistically significant effect on
T5 attainment but not on T4 attainment. T5 attainment, however, was a statistically significant predictor
of T5 self-esteem.
In summary, the results clearly support reciprocal
effects between academic self-concept and educational
attainment. However, the results also suggest that prior
self-esteem has a small positive effect on subsequent
educational attainment beyond what can be explained
in terms of prior measures of SES, test scores, school
grades, and academic self-concept. Although somewhat
at odds with Baumeister et al.’s (2003, 2005) depiction
of results based on these data in their interpretation of
the Bachman and O’Malley’s (1977) study, these findings are consistent findings actually reported by
Bachman and O’Malley’s results that were based on a
very different analytical strategy and different variables.
CONCLUSIONS AND IMPLICATIONS
Marsh and Craven (2006) sought a rapprochement
between Baumeister et al.’s (2003) pessimistic conclusions about the importance of self-beliefs and their own
more optimistic perspective based on their research
showing a pattern of reciprocal effects between academic self-concept and academic achievement. From their
multidimensional perspective (also see related support
for the specificity matching hypothesis by Swann et al.,
2007), Marsh and Craven argued that it is entirely logical that there are few, if any, significant linkages
between self-esteem and school grades even though the
reciprocal linkages between academic self-concept and
school grades are consistently positive. They lamented
that Baumeister et al. had limited their consideration to
older studies of self-esteem and had not considered the
growing body of newer academic self-concept studies
based on stronger statistical methodology and theory,
but they conceded that their reciprocal effects model
studies had focused on academic self-concept to the virtual exclusion of self-esteem. What was needed, they
argued, was more research that considered the complicated pattern of reciprocal effects between academic selfconcept, self-esteem, academic achievement, and other
academic criterion measures. Noting that the reviews by
Baumeister et al. and by Marsh and Craven both
emphasized apparently contradictory interpretations of
Youth in Transition studies, we reanalyzed this classic
database, including the self-esteem measures emphasized by Baumeister et al., the academic self-concept
measures emphasized by Marsh and Craven, and the
educational attainment measures originally emphasized
by Bachman and O’Malley (1977). Based on a wider
selection of variables and stronger statistical methodology,
we found (a) consistent support for positive reciprocal
effects between academic self-concept and school grades,
(b) consistent support for reciprocal effects relating academic self-concept and educational attainment, and (c) only
weak and inconsistent support for linkages between
self-esteem and either school grades or attainment.
It is also important to emphasize that we are not
claiming that self-esteem is never a useful construct.
Rather, consistent with the specificity matching principle (Swann et al., 2007), our conclusion is that when the
focus of a study is on educational outcomes, it is important to focus on academic components of self-concept.
While our results provide clear support for the specificitymatching principle, they also demonstrate implications
of this principle when taken to its logical extreme.
Indeed, a growing body of educational research
reviewed by Marsh and Craven (2006) shows that
global self-esteem is nearly unrelated to a wide variety
of academic outcomes, even though these outcomes are
substantially and logically related to specific components of academic self-concept. Nevertheless, the results
of the present investigation provide some weak support
for the positive effects of prior global self-esteem on subsequent academic achievement and educational attainment. Swann et al. (2007) also reviewed other research
consistent with the specificity-matching principle, showing that self-esteem significantly but weakly predicted
specific outcomes and more strongly predicted global outcomes (e.g., Donnellan, Trzesniewski, Robins, Moffitt, &
Caspi, 2005; Trzesniewski et al., 2006). Although beyond the
scope of the present investigation, there is need for more rigorous tests of the specificity-matching principle, comparing
the predictive power of self-esteem and a comprehensive
set of specific components of self-concept across a range
of specific and global outcome measures (e.g., Marsh
et al., 2006) to evaluate when (or if) global self-esteem
is better able to predict substantively important outcomes than appropriately constructed specific components of self-esteem.
Following from this, we suggest that the behavioral
implications of enhancing academic self-concept are
clear. Some examples of the behavioral implications of
having higher academic self-concept include a reduction
in test anxiety (e.g., Zeidner & Schleyer, 1999), taking
advanced course work (e.g., Marsh, 1993; Marsh &
Yeung, 1998), not dropping out of school (e.g., House,
1993), and as shown in the present investigation, higher levels of long-term educational attainment. Thus, unlike global
self-esteem, which has no clear behavioral implications for
academic achievement or future educational attainment, improved academic self-concept has obvious
direct and indirect implications.
In summary, our new results extend support for the
reciprocal effects model debate in three important directions. First, they provide definitive support for Marsh
and Craven’s (2006) proposed rapprochement in their
debate with Baumeister et al. (2003, 2005), integrating
apparently contradictory results into a single theoretical
548 PERSONALITY AND SOCIAL PSYCHOLOGY BULLETIN
Marsh, O’Mara / SELF-ESTEEM, SELF-CONCEPT, AND PERFORMANCE 549
framework based on a multidimensional perspective of
self-concept. Second, they extend previous reciprocal
effects model research, demonstrating the effects of
prior academic self-concept on subsequent educational
attainment, beyond the effects of prior attainment, SES,
academic ability, school grades, and self-esteem. Third,
the integration of Baumeister’s et al.’s implicit unidimensional approach with Marsh and Craven’s explicit
multidimensional approach provides a substantively
important, new application of the specificity matching
principle (Swann et al., 2007).
Despite the high quality of the Youth in Transition
data, the application of new, more powerful statistical
analyses, and new theoretical developments underpinning our research, there are limitations that dictate
caution in the interpretation of our results. Because
the data are based on responses by boys, the generalizability to girls cannot be tested (although Marsh and
Craven, 2006, reported that the pattern of results is
similar for boys and girls based on other research; also
see Marsh, 1989; Marsh & Yeung, 1998). In addition,
the use of historical data suggests the need for further
research to test the generalizability of the results
(although the results are consistent with Valentine and
DuBois’s 2005 meta-analysis). Although self-concept
is necessarily based on self-perceptions, school grades
and educational attainment were also based on selfreport measures (even though these measures were verified in interviews and are clearly less subjective than
self-concept responses). While the reciprocal effects
model clearly posits causal relations and the results
support these predictions, it is important to be cautious about drawing causal interpretations from correlational data. However, we agree with Baumeister et
al. (2003) that longitudinal panel studies provide the
strongest tests of the reciprocal effects model—and
indeed may even be overly conservative—and that the
Youth in Transition is one of the best longitudinal
panel studies for this purpose. Finally, although the
statistical methods used here are clearly stronger than
those used by Marsh (1990) and particularly those
used by Bachman and O’Malley (1977) and those in
studies reviewed by Baumeister et al., this is a rapidly
developing area in which further methodological
developments are likely.
Our results also bear on another concern expressed by
Baumeister et al. (2003) who argued that it was counterproductive to enhance self-esteem in ways that were not
contingent upon meritorious performances; they were
particularly critical of the so-called self-esteem movement.
Although we take a somewhat different perspective, the
Marsh and Craven (2006) review of reciprocal effects
model research and the results of the present investigation
largely support Baumeister et al.’s concerns and
have important practical implications for practitioners,
counselors, and policy makers. Support for our reciprocal
effects model implies that academic self-concept and performance are reciprocally related and mutually reinforcing. Improved academic self-concepts lead to better
performance, and improved performance leads to better
academic self-concepts. However, if practitioners enhance
academic self-concepts without improving performance,
then the gains in academic self-concept are likely to be
short lived. Even less useful would be interventions such
as those criticized by Baumeister et al. that aim to improve
self-concept in areas unrelated to intended areas of performance gain. Drawing from the reciprocal effects model
research, if practitioners improve performances without
also fostering participants’ self-beliefs in their capabilities,
then the performance gains are also unlikely to be long
lasting. Worse yet, interventions may unintentionally
undermine academic self-concept in ways that will eventually undermine the short-term gains in performance
(e.g., the social comparison/competitive intervention in
Marsh & Peart, 1988). Importantly, if practitioners focus
on either one of these constructs to the exclusion of the
other, then both are likely to suffer. Hence, based on the
reciprocal effects model results, practitioners should strive
to improve both academic self-concept and performance.
Thus, for example, Hilyer and Mitchell (1979) found the
need to supplement performance training with a counseling intervention to enhance self-concepts of participants
with initially low levels of self-concept (also see Marsh &
Peart, 1988).
The juxtaposition of research in support of the multidimensional perspective and the reciprocal effects
model suggests that practitioners need to target specific
components of self-concept logically related to their performance goals and intended outcomes. Whereas targeting global components such as self-esteem may result in
increased happiness, as suggested by Baumeister et al.
(2003), these results may not generalize to the desired
outcomes. This has particular implications for intervention research, as the success in enhancing academic selfconcept or its relation with the desired outcome (e.g.,
achievement) may be inaccurately assessed and substantially underestimate the true effects of the intervention
in relation to appropriate, specific components of selfconcept, as has been shown to be the case in recent
meta-analytic research (O’Mara et al., 2006).
Finally, although our focus has been on substantive
issues, it is significant to emphasize that the study
demonstrates the importance of methodological–
substantive synergies that combine strong theory, the
most appropriate methodologies, and the best statistical
analyses to evaluate substantively important issues
(Marsh & Hau, 2007). Marsh and Hau argued that
important methodological advances in latent variable
modelling allow researchers to evaluate new theoretical
models and revisit unresolved issues in ways that were
TABLE A1: Factor Loadings on Latent Factors
Control T1 T2 T3 T4 T5
Indicator SES ABIL GPA ASC Estm GPA ASC Estm GPA Estm EdAt ASC Estm EdAt Estm
1 .67 .76 .95 .68 .61 .95 .72 .63 .95 .62 .95 .74 .65 .95 .70
2 .76 .88 .79 .66 .80 .63 .66 .53 .69 .75
3 .64 .72 .52 .62 .58 .60 .63 .65 .67
4 .56 .81 .26 .31 .38 .32 .28
5 .54 .53 .62 .61 .64 .62
6 .41 .28 .39 .43 .44 .44
7 .47 .5 .49 .51 .59
8 .28 .33 .40 .38 .38
9 .46 .43 .45 .49 .44
10 .35 .35 .38 .39 .33
NOTE: Selected estimated parameters are presented in a completely standardized format (Muthén & Muthén, 2006). SES = socioeconomic status; ABIL = academic ability; GPA = grade point
average; ASC = academic self-concept; Estm = self-esteem; EdAt = educational attainment. T1-T5 = Time 1, Time 2, . . . Time 5. Factor loadings are presented in condensed format in which each
of the 15 latent factors has between 1 and 10 indicators. For example, each of the self-esteem factors has 10 indicators so that there are 10 factor loadings, whereas GPA has a single indicator
so that there is only 1 factor loading. Importantly, each indicator is allowed to load on one and only one factor. Seventy-three indicators were used to infer 15 latent constructs.
550
APPENDIX
SELECTED PARAMETER ESTIMATES FROM THE STRUCTURAL EQUATION MODEL
Marsh, O’Mara / SELF-ESTEEM, SELF-CONCEPT, AND PERFORMANCE 551
TABLE A2: Path Coefficients
Independent
Latent Factors Dependent Latent Factors
SES —
ABIL .39 —
T1GPA –.15 .11 —
T1ASC .22 .28 .35 —
T1Estm .00 –.08 –.01 (.12) —
T2GPA –.11 –.04 .4 .14 .03 —
T2ASC .04 .39 –.08 .62 .11 .30 —
T2Estm –.04 –.05 –.03 .01 .42 .02 (.09) —
T3GPA .01 .07 .21 –.05 –.02 .50 .15 .07 —
T3Estm –.07 –.18 –.00 .03 .17 .01 .13 .57 .07 —
T4EdAt .21 .19 .04 .09 .09 .06 .14 .04 .35 .08 —
T4ASC –.01 –.15 –.05 .14 –.09 –.00 .64 .02 .36 .66 –.03 —
T4Estm .10 .20 .03 –.01 .08 –.06 –.12 .17 –.11 .04 .25 (.27) —
T5EdAt .10 .06 .06 –.00 –.00 .07 .01 –.05 .12 .08 .49 .22 .05 —
T5Estm –.04 .03 –.02 .01 .00 .01 .06 .06 .01 .02 .02 .26 .56 .16 —
NOTE: Selected estimated parameters are presented in a completely standardized format (Muthén & Muthén, 2006). SES = socioeconomic status;
ABIL = academic ability; GPA = grade point average; ASC = academic self-concept; Estm = self-esteem; EdAt = educational attainment. T1-T5 =
Time 1, Time 2, . . . Time 5. Path coefficients (also see Figure 1) relate each predictor factor (independent variables, the factors listed along the
top) to each predicted factor (dependent factors, the variables listed in the left-hand column). Thus, for example, consistent with the a priori
model, SES—the first factor in the causal ordering—is a predictor of all subsequent variables. Those paths that are statistically significant and
that are pertinent to the reciprocal effects model are shown in Figure 1. Numbers in parentheses indicate a correlation between the variables
rather than a path coefficient, as no causal ordering is posited between ASC and Estm at Times 1, 2, and 4.
TABLE A3: Latent Factor Correlations
Latent Factor Correlations
SES 1.00
ABIL .56 1.00
T1GPA .28 .54 1.00
T1ASC .46 .64 .64 1.00
T1Estm .17 .21 .27 .46 1.00
T2GPA .28 .51 .73 .62 .26 1.00
T2ASC .43 .64 .58 .85 .37 .64 1.00
T2Estm .14 .22 .25 .42 .64 .28 .46 1.00
T3GPA .31 .50 .66 .54 .23 .73 .58 .26 1.00
T3Estm .12 .18 .24 .38 .58 .25 .40 .76 .24 1.00
T4EdAt .46 .53 .51 .49 .24 .51 .48 .21 .55 .20 1.00
T4ASC .34 .48 .48 .72 .30 .52 .79 .39 .51 .42 .37 1.00
T4Estm .17 .23 .19 .33 .50 .18 .34 .66 .18 .75 .16 .44 1.00
T5EdAt .34 .39 .40 .33 .16 .40 .34 .14 .41 .17 .55 .26 .14 1.00
T5Estm .12 .21 .16 .28 .34 .18 .29 .46 .17 .49 .16 .29 .56 .16 1.00
NOTE: Selected estimated parameters are presented in completely standardized format (Muthén & Muthén, 2006). SES = socioeconomic status;
ABIL = academic ability; GPA = grade point average; ASC = academic self-concept; Estm = self-esteem; EdAt = educational attainment. T1-T5 =
Time 1, Time 2, . . . Time 5. Estimated latent factor correlations reflect that correlation between each factor within the context of the a priori
model. Because some latent factors are substantially correlated, it is useful to compare path coefficients with corresponding correlations. Thus,
for example, even though SES has a small negative effect on T1GPA (path coefficient of –.15) the corresponding correlation is positive (.28), indicating that the indirect effect of SES on T1GPA mediated through ability is substantial and positive.
not previously possible. More specifically, methodological and statistical approaches used here were not available for either the earlier Youth in Transition studies
(Bachman & O’Malley, 1977; Marsh, 1990) or studies
reviewed by Baumeister et al. (2003), so that issues
emphasized here could not be pursued appropriately in
that earlier research. Combining these methodological
advances with strong theory allowed us to demonstrate
that apparently contradictory conclusions—those by
Marsh and Craven (1997, 2006) and those by Baumeister
et al. (2003, 2005)—were actually consistent with
each other and consistent with a growing body of
research emphasizing a multidimensional perspective to
self-concept.
552 PERSONALITY AND SOCIAL PSYCHOLOGY BULLETIN
REFERENCES
Bachman, J. G. (2002). Volume I of the documentation manual. Ann
Arbor, MI: Inter-university Consortium for Political and Social
Research.
Bachman, J. G., & O’Malley, P. M. (1977). Self-esteem in young men:
A longitudinal analysis of the impact of educational and occupational attainment. Journal of Personality and Social Psychology,
35, 365-380.
Baumeister, R. F., Campbell, J. D., Krueger, J. I., & Vohs, K. D.
(2003). Does high self-esteem cause better performance, interpersonal success, happiness, or healthier lifestyles? Psychological
Science in the Public Interest, 4, 1-44.
Baumeister, R. F., Campbell, J. D., Krueger, J. I., & Vohs, K. D. (2005).
Exploding the self-esteem myth. Scientific American, 292, 84-92.
Bracken, B. A. (1996). Clinical applications of a context-dependent,
multidimensional model of self-concept. In B. A. Bracken (Ed.),
Handbook of self-concept (pp. 463-503). New York: John Wiley.
Byrne, B. M. (1996). Academic self-concept: Its structure, measurement,
and relation to academic achievement. In B. A. Bracken (Ed.),
Handbook of self-concept (pp. 287-316). New York: John Wiley.
Craven, R. G., Marsh, H. W., & Burnett, P. C. (2003). Cracking the
self-concept enhancement conundrum: A call and blueprint for the
next generation of self-concept enhancement research. In H. W.
Marsh, R. G. Craven, & D. M. McInerney (Eds.), International
advances in self research (Vol. 1, pp. 67-90). Greenwich, CT:
Information Age.
Donnellan, B., Trzesniewski, K., Robins, R., Moffitt, T., & Caspi, A.
(2005). Low self-esteem is related to aggression, antisocial behavior, and delinquency. Psychological Science, 16, 328-335.
Fredrickson, B. L. (2006). Unpacking positive emotions: Investigating the
seeds of human flourishing. Journal of Positive Psychology, 1, 57-59.
Furr, R. M. (2005). Differentiating happiness and self-esteem.
Individual Differences Research, 3, 105-127.
Harter, S. (1999). The construction of the self: A developmental perspective. New York: Guilford.
Hattie, J. (1992). Self-concept. Hillsdale, NJ: Lawrence Erlbaum.
Hilyer, J. C., & Mitchell, W. (1979). Effect of systematic physicalfitness training combined with counseling on the self-concept of
college-students. Journal of Counseling Psychology, 26, 427-436.
House, J. D. (1993). The relationship between academic self-concept and
school withdrawal. Journal of Social Psychology, 133, 125-127.
Hunter, J. P., & Csikszentmihalyi, M. (2003). The positive psychology of interested adolescents. Journal of Youth and Adolescence,
32, 27-35.
Jöreskog, K. G. (1979). Statistical estimation of structural models in
longitudinal investigations. In J. R. Nesselroade & B. Baltes (Eds.),
Longitudinal research in the study of behavior and development
(pp. 303-351). New York: Academic Press.
Lopez, S. J., Magyar-Moe, J. L., Petersen, S. E., Ryder, J. A.,
Krieshok, T. S., O’Byrne, K. K., et al. (2006). Counseling psychology’s focus on positive aspects of human functioning. The
Counseling Psychologist, 34, 205-227.
Marsh, H. W. (1986). Global self esteem: Its relation to specific facets
of self-concept and their importance. Journal of Personality and
Social Psychology, 51, 1224-1236.
Marsh, H. W. (1989). Sex differences in the development of verbal
and math constructs: The High School and Beyond study.
American Educational Research Journal, 26, 191-225.
Marsh, H. W. (1990). The causal ordering of academic self-concept
and academic achievement: A multiwave, longitudinal panel
analysis. Journal of Educational Psychology, 82, 646-656.
Marsh, H. W. (1993). Academic self-concept: Theory measurement
and research. In J. Suls (Ed.), Psychological perspectives on the self
(Vol. 4, pp. 59-98). Hillsdale, NJ: Lawrence Erlbaum.
Marsh, H. W. (1996). Positive and negative global self-esteem: A substantively meaningful distinction or artifactors? Journal of
Personality & Social Psychology, 70, 810-819.
Marsh, H. W., Byrne, B. M., & Shavelson, R. (1988). A multifaceted academic self-concept: Its hierarchical structure and its relation to academic achievement. Journal of Educational Psychology, 80, 366-380.
Marsh, H. W., & Craven, R. G. (1997). Academic self-concept:
Beyond the dustbowl. In G. Phye (Ed.), Handbook of classroom
assessment: Learning, achievement, and adjustment (pp. 131-198).
Orlando, FL: Academic Press.
Marsh, H. W., & Craven, R. G. (2006). Reciprocal effects of selfconcept and performance from a multidimensional perspective:
Beyond seductive pleasure and unidimensional perspectives.
Perspectives on Psychological Science, 1, 133-163.
Marsh, H. W., Craven, R. G., & Martin, A. (2006). What is
the nature of self-esteem? Unidimensional and multidimensional
perspectives. In M. Kernis (Ed.), Self-esteem: Issues and answers
(pp. 16-25). New York: Psychology Press.
Marsh, H. W., & Hattie, J. (1996). Theoretical perspectives on the
structure on self-concept. In B. A. Bracken (Ed.), Handbook of
self-concept (pp. 38-90). New York: John Wiley.
Marsh, H. W., & Hau, K.-T. (1996). Assessing goodness of fit: Is parsimony
always desirable? Journal of Experimental Education, 64, 364-390.
Marsh, H. W., & Hau, K.-T. (2007). Applications of latent-variable
models in educational psychology: The need for methodologicalsubstantive synergies. Contemporary Educational Psychology, 32,
151-171.
Marsh, H. W., & Peart, N. (1988). Competitive and cooperative
physical fitness training programs for girls: Effects on physical fitness and on multidimensional self-concepts. Journal of Sport and
Exercise Psychology, 10, 390-407.
Marsh, H. W., Trautwein, U., Lüdtke, O., Köller, O., & Baumert, J.
(2006). Integration of multidimensional self-concept and core personality constructs: Construct validation and relations to wellbeing and achievement. Journal of Personality, 74, 403-455.
Marsh, H. W., & Yeung, A. S. (1998). Longitudinal structural equation models of academic self-concept and achievement: Gender
differences in the development of math and English constructs.
American Educational Research Journal, 35, 705-738.
Muthén, L. K., & Muthén, B. (2006). Mplus user’s guide (Version 4).
Los Angeles: Muthén & Muthén.
O’Mara, A. J., Marsh, H. W., Craven, R. G., & Debus, R. (2006). Do
self-concept interventions make a difference? A synergistic blend of
construct validation and meta-analysis. Educational Psychologist,
41, 181-206.
Rosenberg, M. (1979). Conceiving the self. New York: Basic Books.
Seligman, M. E. P., & Csikszentmihalyi, M. (2000). Positive psychology: An introduction. American Psychologist, 55, 5-14.
Shavelson, R. J., Hubner, J. J., & Stanton, G. C. (1976). Validation of
construct interpretations. Review of Educational Research, 46,
407-441.
Swann, W. B. Jr., Chang-Schneider, C., & Larsen McClarty, K.
(2007). Do people’s self-views matter? Self-concept and selfesteem in everyday life. American Psychologist, 62, 84-94.
Trautwein, U., Lüdtke, O., Köller, O., & Baumert, J. (2006). Selfesteem, academic self-concept, and achievement: How the learning
environment moderates the dynamics of self-concept. Journal of
Personality and Social Psychology, 90, 334-349.
Trzesniewski, K., Donnellan, B., Moffitt, T., Robins, R., Poulton, R.,
& Caspi, A. (2006). Low self-esteem during adolescence predicts
poor health, criminal behavior, and limited economic prospects
during adulthood. Developmental Psychology, 42, 381-390.
Valentine, J. C., & DuBois, D. L. (2005). Effects of self-beliefs on academic achievement and vice-versa: Separating the chicken from the
egg. In H. W. Marsh, R. G. Craven, & D. M. McInerney (Eds.),
International advances in self research (Vol. 2, pp. 53-78).
Greenwich, CT: Information Age.
Valentine, J. C., DuBois, D. L., & Cooper, H. (2004). The relations
between self-beliefs and academic achievement: A systematic
review. Educational Psychologist, 39, 111-133.
Zeidner, M., & Schleyer, E. J. (1999). The big-fish-little-pond effect
for academic self-concept, test anxiety, and school grades in gifted
children. Contemporary Educational Psychology, 24, 305-329.