Content uploaded by Ulrich Orth
Author content
All content in this area was uploaded by Ulrich Orth
Content may be subject to copyright.
Does Low Self-Esteem Predict Depression and Anxiety? A Meta-Analysis
of Longitudinal Studies
Julia Friederike Sowislo and Ulrich Orth
University of Basel
Low self-esteem and depression are strongly related, but there is not yet consistent evidence on the nature
of the relation. Whereas the vulnerability model states that low self-esteem contributes to depression, the
scar model states that depression erodes self-esteem. Furthermore, it is unknown whether the models are
specific for depression or whether they are also valid for anxiety. We evaluated the vulnerability and scar
models of low self-esteem and depression, and low self-esteem and anxiety, by meta-analyzing the
available longitudinal data (covering 77 studies on depression and 18 studies on anxiety). The mean age
of the samples ranged from childhood to old age. In the analyses, we used a random-effects model and
examined prospective effects between the variables, controlling for prior levels of the predicted variables.
For depression, the findings supported the vulnerability model: The effect of self-esteem on depression
(⫽⫺.16) was significantly stronger than the effect of depression on self-esteem (⫽⫺.08). In
contrast, the effects between low self-esteem and anxiety were relatively balanced: Self-esteem predicted
anxiety with ⫽⫺.10, and anxiety predicted self-esteem with ⫽⫺.08. Moderator analyses were
conducted for the effect of low self-esteem on depression; these suggested that the effect is not
significantly influenced by gender, age, measures of self-esteem and depression, or time lag between
assessments. If future research supports the hypothesized causality of the vulnerability effect of low
self-esteem on depression, interventions aimed at increasing self-esteem might be useful in reducing the
risk of depression.
Keywords: self-esteem, depression, anxiety, longitudinal studies
There is an overwhelming amount of self-help literature that
explains how people can boost and sustain their self-esteem in
order to improve their psychological adjustment. But does self-
esteem indeed contribute to psychological health or, to put it
differently, does low self-esteem compromise a person’s psycho-
logical adjustment? Previous research suggests that self-esteem is
linked to indicators of psychological adjustment such as happiness
(H. Cheng & Furnham, 2004; Diener & Diener, 1995), high
positive affect and low negative affect (Orth, Robins, & Widaman,
2012), and to the absence, or a low number, of psychological
symptoms such as depression (Orth, Robins, Trzesniewski, Maes,
& Schmitt, 2009; J. E. Roberts & Monroe, 1992) and bulimia
(Vohs et al., 2001). However, with respect to many of these
variables, the precise nature of their relation with self-esteem has
not ultimately been established (Baumeister, Campbell, Krueger,
& Vohs, 2003).
In the present research, we focus on the relation of self-esteem
with two important indicators of low psychological adjustment,
specifically depression and anxiety.
1
The central goal of this study
was to evaluate the vulnerability and scar models of low self-
esteem and depression, by meta-analyzing the available longitudi-
nal data. Moreover, we tested whether the vulnerability and scar
models (if supported by the data) are specific for depression or
whether they are also valid models for anxiety. Finally, we exam-
ined moderators that might explain variability in the relation
between low self-esteem and depression.
Self-Esteem: Concept, Measurement, Function, and
Consequences
Concept of Self-Esteem
The concept of self-esteem has elicited a large body of theoret-
ical accounts and empirical research (see, e.g., Baumeister, 1998;
Kernis, 2006; Swann & Bosson, 2010). Historically, the first
influential definition of self-esteem dates back to James (1890),
who considered self-esteem to be the ratio of success and preten-
sions in important life domains. Whereas James focused to a
stronger degree on the individual processes that form self-esteem,
1
Throughout this article, we use the term depression to denote a con-
tinuous variable (i.e., individual differences in depressive affect) rather
than a clinical category such as major depressive disorder (American
Psychiatric Association, 2000). Taxometric analyses suggest that depres-
sion is best conceptualized as a continuous construct (Hankin, Fraley,
Lahey, & Waldman, 2005; Lewinsohn, Solomon, Seeley, & Zeiss, 2000;
Prisciandaro & Roberts, 2005; Ruscio & Ruscio, 2000).
This article was published Online First June 25, 2012.
Julia Friederike Sowislo and Ulrich Orth, Department of Psychology,
University of Basel, Basel, Switzerland.
This research was supported by Swiss National Science Foundation
Grant PP00P1-123370 to Ulrich Orth.
Correspondence concerning this article should be addressed to Julia
Friederike Sowislo, Department of Psychology, University of Basel, Mis-
sionsstrasse 62, 4055 Basel, Switzerland. E-mail: julia.sowislo@unibas.ch
Psychological Bulletin © 2012 American Psychological Association
2013, Vol. 139, No. 1, 213–240 0033-2909/12/$12.00 DOI: 10.1037/a0028931
213
later symbolic interactionism approaches stressed the social influ-
ences on self-esteem (Cooley, 1902; Goffman, 1959; Mead, 1934).
For instance, in his conception of the looking-glass self, Cooley
(1902) hypothesized that self-views are based upon information
gathered from explicit or implicit feedback from others. More
recent definitions of self-esteem emphasize the fact that self-
esteem should be distinguished from other components of the
self-concept (such as self-knowledge and self-efficacy), insofar as
self-esteem represents the affective, or evaluative, component of
the self-concept; it signifies how people feel about themselves
(Leary & Baumeister, 2000). This affective self-evaluation is sub-
jective at its core and is not based on specific behaviors (Robins,
Hendin, & Trzesniewski, 2001). According to Rosenberg (1989),
high self-esteem “expresses the feeling that one is ‘good enough.’
The individual simply feels that he is a person of worth. . . . He
does not necessarily consider himself superior to others” (p. 31).
Although Baumeister and his colleagues share the view of self-
esteem as self-appraisal with an affective component, they expand
the definition of self-esteem to include feelings of superiority,
arrogance, and pride (e.g., Baumeister, 1998; Baumeister, Smart,
& Boden, 1996).
In the literature, it is debated whether self-esteem is best con-
ceptualized as a global evaluation of the self (i.e., global self-
esteem) or as an evaluation in specific self-relevant domains such
as intellectual abilities, physical appearance, and social compe-
tence (i.e., domain-specific self-esteem; Swann & Bosson, 2010).
One finding that sheds more light on this debate is that both global
and domain-specific self-evaluations show predictive ability for
important outcomes, as long as these outcomes exhibit the same
degree of specificity as the self-evaluation that is used as a pre-
dictor (specificity-matching principle; Swann, Chang-Schneider,
& McClarty, 2007). More precisely, global self-esteem seems to
have predictive ability for outcomes measured at a global level
(such as several outcomes bundled together; for an example, see
Trzesniewski et al., 2006), whereas domain-specific self-esteem
seems to have predictive ability for outcomes measured at a
specific level (e.g., academic self-esteem predicts academic out-
comes; Marsh, Trautwein, Lu¨dtke, Koller, & Baumert, 2006).
With regard to the relation between self-esteem and psycholog-
ical adjustment, there are three reasons for focusing on global
self-esteem rather than domain-specific self-esteem. First, most of
the theories linking self-esteem to psychological adjustment ad-
dress global self-esteem but not domain-specific self-esteem (e.g.,
Abramson, Seligman, & Teasdale, 1978; Blatt, D’Afflitti, & Quin-
lan, 1976; G. W. Brown & Harris, 1978). Second and relatedly,
most studies in this field have used measures of global self-esteem
(for reviews, see Orth, Robins, & Roberts, 2008; Zeigler-Hill,
2010). Third, according to the specificity-matching principle, it
seems reasonable to examine global self-esteem in this context,
because indicators of psychological adjustment such as depression
and anxiety are relatively global constructs that combine a number
of cognitive, affective, and somatic symptoms (Swann et al.,
2007).
Measurement of Self-Esteem
Measures of self-esteem reflect the distinction between
global and domain-specific self-evaluations (for a review, see
Blascovich & Tomaka, 1991). Frequently used measures of
global self-esteem, all of which are multi-item scales, include
the Rosenberg Self-Esteem Scale (RSE; Rosenberg, 1965), the
Janis–Field Feelings of Inadequacy Scale (Fleming & Courtney,
1984), the Texas Social Behavior Inventory (Helmreich &
Stapp, 1974), and the Self-Liking/Self-Competence Scale (Ta-
farodi & Swann, 2001). Prominent measures of domain-specific
self-esteem are, for example, the Self-Description Question-
naire (Marsh, 1990), the Self-Perception Profile for Children
(Harter, 1985), and the Self-Perception Profile for Adolescents
(Harter, 1988). Research suggests that these measures generally
have good psychometric properties (Blascovich & Tomaka,
1991; Byrne, 1996; Fleming & Courtney, 1984; Gray-Little,
Williams, & Hancock, 1997; Marsh, Ellis, Parada, Richards, &
Heubeck, 2005; Marsh, Scalas, & Nagengast, 2010). For exam-
ple, the widely used RSE (Rosenberg, 1965) shows good inter-
nal consistency and test–retest reliability (Blascovich & To-
maka, 1991; Robins, Trzesniewski, Tracy, Gosling, & Potter,
2002). Moreover, research supports the construct validity of
this measure: First, factor analyses suggest that there is only
one substantive factor that explains responses to the RSE (Gray-
Little et al., 1997; Marsh et al., 2010; Schmitt & Allik, 2005).
Second, the RSE shows good discriminant validity, for in-
stance, regarding measures of life satisfaction, optimism, and
academic outcomes (Blascovich & Tomaka, 1991; Lucas, Die-
ner, & Suh, 1996; Robins et al., 2001). Third, the RSE shows
convergent validity with other measures of self-esteem
(Bosson, Swann, & Pennebaker, 2000). For example, in a
multisample study by Zeigler-Hill (2010), correlations between
the RSE and the above-mentioned measures of global self-
esteem ranged from .63 to .90.
All the measures discussed above are based on self-reports;
that is, respondents are explicitly asked to reflect on their global
or domain-specific self-worth. As self-esteem, by definition, is
a subjective construct, it cannot be validly assessed with ob-
jective criteria (Baumeister, 1998). However, in the past few
decades, researchers have also explored methods other than
self-report to assess self-esteem, namely, implicit measures
(Bosson et al., 2000; Krizan & Suls, 2009). According to a
recent review by Buhrmester, Blanton, and Swann (2011), the
most frequently used implicit measures of self-esteem are the
Implicit Association Test (Greenwald & Farnham, 2000) and
the Name–Letter Test (Greenwald & Farnham, 2000; Nuttin,
1985). However, research suggests that the currently available
implicit measures of self-esteem suffer from low reliability and
low convergent validity with each other (Bosson et al., 2000;
Krizan & Suls, 2009) and with explicit measures of self-esteem
(Bosson et al., 2000; Krizan & Suls, 2008). Moreover, implicit
measures of self-esteem show weak predictive validity for
theoretically relevant criteria such as personality (Krizan &
Suls, 2009) and well-being (Buhrmester et al., 2011; Schim-
mack & Diener, 2003). Buhrmester et al. concluded from their
review that the Implicit Association Test and the Name–Letter
Test measure generalized implicit affect and implicit egotism,
respectively, rather than self-esteem; thus, although implicit
measures are a promising avenue for self-esteem measurement,
there is not yet sufficiently strong support for their validity. For
these reasons, in the present research we restricted our analyses
to explicit measures of self-esteem.
214
SOWISLO AND ORTH
Function of Self-Esteem
People tend to have a pervasive motive to increase their self-
esteem and to maintain high self-esteem (Sedikides, 1993;
Sedikides, Gaertner, & Toguchi, 2003; but see Heine, Lehman,
Markus, & Kitayama, 1999). Correspondingly, many psychologi-
cal theories assume that people are motivated to enhance and
maintain their self-esteem without further delineating its functional
value (cf. Pyszczynski, Greenberg, Solomon, Arndt, & Schimel,
2004). However, there are a few approaches that seek to explain
why self-esteem is important for humans (for an outline of these
approaches, see Crocker & Park, 2004; Leary & Baumeister,
2000).
First, according to sociometer theory (Leary & Baumeister,
2000; Leary, Tambor, Terdal, & Downs, 1995), humans have a
fundamental need for belongingness, because social inclusion has
many adaptive benefits (e.g., the possibility of sharing knowledge
within social groups; see also Baumeister & Leary, 1995). The
theory states that self-esteem is a sociometer that serves as a
subjective monitor of the extent to which a person is valued as a
member of desirable groups and relationships. Thus, when people
perceive their relational value as low, their self-esteem should be
equally low, motivating behavior aimed at increasing or restoring
social inclusion.
Second, according to terror management theory (J. Greenberg,
Pyszczynski, & Solomon, 1986; Pyszczynski et al., 2004), people
have a central motive to identify with cultural values and groups,
because this identification promises either literal immortality (e.g.,
being part of a religious group that believes in reincarnation) or
symbolic immortality (e.g., being part of a cultural group whose
existence will endure after one’s own death) and consequently
reduces the deeply rooted fear of death. Thus, when people see
themselves as living up to these cultural values, their self-esteem
should be high, in turn serving as a buffer against the fear of death.
Interestingly, the fact that both theories stress the interpersonal
component of self-esteem is in line with early psychological ac-
counts of self-views as mentioned above (e.g., Cooley, 1902;
Goffman, 1959; Mead, 1934). Moreover, both theories imply an
association between self-esteem and psychological adjustment. For
terror management theory, this association is more evident, as
self-esteem is assumed to buffer against anxiety. From the per-
spective of sociometer theory, self-esteem is related to psycholog-
ical adjustment via beneficial aspects of social inclusion. For
example, socially excluded individuals may suffer from loneliness
and low social support, which increases the risk for depression
(e.g., Joiner, 1997; Nolan, Flynn, & Garber, 2003; Stice, Ragan, &
Randall, 2004).
Consequences of Self-Esteem
A much debated question in the literature is whether self-esteem
has an impact on real-life outcomes or whether self-esteem is
merely an epiphenomenon of success and well-being in the rela-
tionship, work, and health domain (Baumeister et al., 2003; Harter,
1999; Swann et al., 2007). Although research suggests that self-
esteem is correlated with many factors in important life domains
(e.g., relationship satisfaction, Shackelford, 2001; socioeconomic
status, Twenge & Campbell, 2002), this research does not dem-
onstrate that self-esteem actually influences these correlates. The
available longitudinal studies suggest that self-esteem might have
significant positive effects on important life outcomes (e.g., Orth et
al., 2012; Trzesniewski et al., 2006; but see Boden, Fergusson, &
Horwood, 2008), but further research is needed to test the causality
of the hypothesized effects of self-esteem. Moreover, research
suggests a causal link between self-esteem and task persistence
(Baumeister et al., 2003). More precisely, laboratory experiments
have repeatedly shown that high self-esteem facilitates more adap-
tive persistence behavior: Individuals with high self-esteem persist
longer in the face of failure (e.g., Perez, 1973; Shrauger & Sorman,
1977), but whenever persistence is maladaptive (e.g., when con-
fronted with unsolvable tasks), they persist less than individuals
with low self-esteem (e.g., Di Paula & Campbell, 2002; McFarlin,
1985). This adaptive self-regulatory behavior might contribute to
the link between self-esteem and psychological adjustment
(Baumeister et al., 2003). For instance, Shrauger and Sorman
(1977) argued that persistence is often needed for the accomplish-
ment of complex tasks and thereby helps to attain long-lasting
satisfaction and external rewards. Furthermore, they suggested that
task persistence may result in a sense of mastery and control
(Shrauger & Sorman, 1977), which is inversely related to phenom-
ena such as depression (Abramson et al., 1978).
Importantly, some researchers have proposed that self-esteem
may be associated not only with positive outcomes (i.e., the bright
side of high self-esteem) but also with negative attributes (i.e., the
dark side of high self-esteem). More precisely, Baumeister et al.
(1996) suggested that some forms of high self-esteem—
specifically, inflated and unstable high self-esteem—may cause
interpersonal aggression and violence, because people with overly
high self-esteem are more prone to experience ego threats and,
consequently, are more strongly motivated to defend their self-
esteem by devaluating and attacking people who question their
inflated self-views (see also Crocker & Park, 2004; Kernis, Gran-
nemann, & Barclay, 1989). However, other studies suggest that
low, but not high, self-esteem predicts antisocial behavior and
interpersonal violence, in particular when the confounding effect
of narcissism is statistically controlled for (e.g., Donnellan,
Trzesniewski, Robins, Moffitt, & Caspi, 2005; Paulhus, Robins,
Trzesniewski, & Tracy, 2004). Overall, the available research
suggests that high self-esteem may have positive consequences for
the well-being and success of the individual and that low self-
esteem may be a risk factor for negative outcomes.
Relation Between Low Self-Esteem and Depression
Depression is not only an important indicator of low psycho-
logical adjustment but also a universal major health concern
(Moussavi et al., 2007). According to the World Health Organi-
zation (2008), depressive disorders are among the leading contrib-
utors to the global burden of disease. For example, major depres-
sion affects a wide range of the population (e.g., a lifetime
prevalence of 16.6% was estimated in the study of Kessler, Ber-
glund, et al., 2005) and is highly recurrent (e.g., Kessler et al.,
2003; Solomon et al., 2000). It is associated with impaired func-
tioning in the relationship (e.g., Davila, Karney, Hall, & Bradbury,
2003; Wade & Pevalin, 2004), work (e.g., Adler et al., 2006;
Kessler et al., 2006), and health domain (e.g., Räikkönen, Mat-
thews, & Kuller, 2007; Wulsin & Singal, 2003) and with elevated
rates of suicidal behavior (e.g., Berman, 2009; Harris & Barra-
215
LOW SELF-ESTEEM, DEPRESSION, AND ANXIETY
clough, 1997). As yet, the etiology of depression is not fully
understood, but a biopsychosocial model is often assumed to best
explain the emergence of depression (Gotlib & Hammen, 2009).
Although it is generally undisputed that low self-esteem and
depression are related, researchers disagree about the nature of the
relation. Importantly, some researchers have argued that self-
esteem and depression are essentially one construct and should be
conceptualized as opposite poles of a single dimension (i.e., de-
pression being the same as low self-esteem; Watson, Suls, & Haig,
2002). Watson et al. (2002) found strong negative correlations
between self-esteem and depression and, on the basis of these
results, cautioned against treating self-esteem and depression as
distinct constructs (see also Judge, Erez, Bono, & Thoresen, 2002).
However, theoretical considerations suggest that it is useful to
distinguish between the two constructs. First, self-esteem plays an
important role in several classic theories of depression that do not
conceptualize low self-esteem as a synonym for depression but as
a distinct construct (Abramson et al., 1978; Blatt et al., 1976;
G. W. Brown & Harris, 1978); moreover, contemporary models of
depression and reviews of the literature also emphasize the role of
low self-esteem in the etiology of depressive disorders (Evraire &
Dozois, 2011; Hammen, 2005; Joiner, 2000; Morley & Moran,
2011; O’Brien, Bartoletti, & Leitzel, 2006; J. E. Roberts, 2006).
Second, although feelings of worthlessness are a symptom of
depressive disorders, they are neither a sufficient nor a necessary
criterion (American Psychiatric Association, 2000). Third, low
self-esteem is not only a symptom of depression but also an
associated feature of a wide range of other clinical conditions, such
as learning disorders, stuttering, social phobia, and attention-
deficit/hyperactivity disorder (American Psychiatric Association,
2000).
Likewise, empirical findings suggest that it is useful to distin-
guish between self-esteem and depression. First, the correlations
reported in previous research range from the ⫺.20s to the ⫺.70s
(for a review, see Orth et al., 2008). Thus, the correlation between
self-esteem and depression varies widely across studies, and al-
though some studies found strong correlations, the relation is not
as strong as would be expected if self-esteem and depression were
indicators of a common construct. Second, studies assessing the
frequency of individual depressive symptoms have found that
feelings of worthlessness are present only in a portion of individ-
uals diagnosed with depression and that feelings of worthlessness
do not belong to the most frequent depressive symptoms (Buch-
wald & Rudick-Davis, 1993; Minor, Champion, & Gotlib, 2005;
Spalletta, Troisi, Saracco, Ciani, & Pasini, 1996). In line with the
diagnostic criteria for depressive episodes, which do not require
that feelings of worthlessness are present (American Psychiatric
Association, 2000), these findings suggests that there can be de-
pression without low self-esteem. Third, in two independent sam-
ples, Orth et al. (2008) found that a common factor model did not
provide a good fit to the data, whereas a two-factor model did (but
see Hankin, Lakdawalla, Carter, Abela, & Adams, 2007). Fourth,
self-esteem and depression are differentially related to events that
happen in people’s lives. For example, whereas there is a robust
predictive effect of stressful life events on depression (Hammen,
2005; Kessler, 1997), the available evidence suggests that stressful
life events do not predict changes in self-esteem (Orth, Robins, &
Meier, 2009; Orth, Trzesniewski, & Robins, 2010). Moreover,
whereas there is consistent evidence that depression contributes to
the occurrence of future stressful life events (i.e., stress generation
effect; Cole, Nolen-Hoeksema, Girgus, & Paul, 2006; Hammen,
1991), the results of three independent studies suggest that self-
esteem does not predict whether stressful life events will occur
(Orth, Robins, & Meier, 2009). Finally, some studies have shown
that self-esteem and depression are cross-sectionally (McPherson
& Lakey, 1993) and prospectively (Orth, Robins, Trzesniewski, et
al., 2009) related to each other, even after controlling for prior
levels of each construct. It is unlikely that two indicators of a
common factor would have replicable cross-lagged effects because
their shared variance has been systematically removed in the
models. Given these conceptual arguments and empirical results,
we believe that it is useful to distinguish between self-esteem and
depression.
To further illustrate the difference between the constructs, it
might be useful to highlight characteristics of a prototypical person
with low self-esteem versus a prototypical person with depressive
symptoms. According to Rosenberg and Owens (2001; see also
Baumeister, 1993), individuals with low self-esteem can be de-
scribed as follows. For example, they tend to be sensitive to
criticism and to focus their attention on how others see them.
Moreover, they tend to avoid people by whom they feel their
self-esteem might be threatened and to conceal their inner thoughts
and feelings from others. Also, as members of a group, these
individuals have the tendency to stay at its fringes and not to
contribute much to the group discussion. More generally, individ-
uals with low self-esteem tend to avoid risk and try to protect their
self-esteem instead of putting their abilities to the test. Further-
more, they may be marked by an attitude of uncertainty, particu-
larly regarding the self and moral convictions. As a consequence,
individuals with low self-esteem may lack spontaneity, be shy, and
feel lonely and alienated from others. Current depression tends to
be clinically heterogeneous and can present with different patterns
of symptoms (Kendler, Gardner, & Prescott, 1999). For example,
as described in the Diagnostic and Statistical Manual of Mental
Disorders (American Psychiatric Association, 2000), individuals
with current depression might feel empty and sad, or have the
feeling of not being able to take it anymore. Moreover, they tend
to lose the ability to derive pleasure from things that used to
interest them, and they may feel a lack of drive and energy for
work, family, and recreational activities. They tend to have prob-
lems concentrating, and others may notice that their movement and
speech are slowed down. Individuals with depression might also
experience alterations in sleep and appetite. Again, individuals
with depression can, but do not have to, experience low self-
esteem.
Two dominant models on the relation between low self-esteem
and depression exist in the literature. Within a diathesis-stress
framework, the vulnerability model suggests that negative evalu-
ations of the self (which are conceptually close to low self-esteem;
A. T. Beck, Steer, Epstein, & Brown, 1990) constitute a causal risk
factor of depression (e.g., A. T. Beck, 1967; Butler, Hokanson, &
Flynn, 1994; Metalsky, Joiner, Hardin, & Abramson, 1993; J. E.
Roberts & Monroe, 1992; Whisman & Kwon, 1993). For example,
according to A. T. Beck’s (1967) cognitive theory of depression,
negative beliefs about the self are not just a symptom of depression
but a diathesis exerting causal influence in the onset and mainte-
nance of depression. Conversely, the scar model states that low
self-esteem is a consequence of depression, rather than a causal
216
SOWISLO AND ORTH
factor, because episodes of depression may leave permanent scars
in the self-concept of the individual (cf. Coyne, Gallo, Klinkman,
& Calarco, 1998; Coyne & Whiffen, 1995; Rohde, Lewinsohn, &
Seeley, 1990; Zeiss & Lewinsohn, 1988). It is important to note
that the vulnerability model and the scar model are not mutually
exclusive, because both processes (i.e., low self-esteem contribut-
ing to depression and depression eroding self-esteem) might op-
erate simultaneously.
The extant research has not yet provided unequivocal evidence
in favor of the vulnerability or scar model. Although a growing
body of longitudinal studies suggests that low self-esteem prospec-
tively predicts depression (e.g., Kernis et al., 1998; Orth, Robins,
& Meier, 2009; Orth et al., 2008; Orth, Robins, Trzesniewski, et
al., 2009; J. E. Roberts & Monroe, 1992), some studies have failed
to confirm this temporal pattern of results; moreover, the results of
some studies have found prospective effects in support of the scar
model (Burwell & Shirk, 2006; Shahar & Davidson, 2003; Shahar
& Henrich, 2010). It is possible not only that these inconsistencies
are due to within-study sampling error, but that systematic differ-
ences between studies (e.g., age of participants or measures used)
account for variability in the findings. In the present research, we
therefore test for moderating factors of vulnerability and scar
effects, or, in other words, whether the vulnerability and scar
effects replicate across sampling and method factors such as gen-
der, age, sample type, time lag between assessments, and measures
of self-esteem and depression.
Relation Between Low Self-Esteem and Anxiety
An important question is whether the vulnerability model and
scar models (if supported by the meta-analytic results) are specific
for depression or whether low self-esteem is related in similar
ways to affective symptoms other than depressive symptoms. To
address this question, we decided to focus on anxiety for several
reasons. First, anxiety is an important affective variable (Endler &
Kocovski, 2001), because it is the core symptom in the group of
anxiety disorders (American Psychiatric Association, 2000) that
cause a major burden of disease (P. E. Greenberg et al., 1999).
Second, anxiety is associated with depression: Self-report mea-
sures of depression and anxiety are strongly correlated in clinical
(Mendels, Weinstein, & Cochrane, 1972) and nonclinical samples
(Dobson, 1985; Gotlib, 1984; Tanaka-Matsumi & Kameoka,
1986), and depressive and anxiety disorders show a high diagnos-
tic comorbidity (T. A. Brown, Campbell, Lehman, Grisham, &
Mancill, 2001; Kessler, Chiu, Demler, & Walters, 2005). Third,
although depression and anxiety are related, the constructs are
conceptually distinct and can be empirically distinguished (e.g.,
B. J. Cox, Swinson, Kuch, & Reichman, 1993; Endler, Denisoff, &
Rutherford, 1998; McWilliams, Cox, & Enns, 2001; Watson &
Clark, 1992). Fourth, many previous studies have examined
whether risk factors and correlates of depressive disorders are
specific for depression or whether they are also related to anxiety
(A. T. Beck, Steer, & Epstein, 1992; R. Beck & Perkins, 2001; R.
Beck et al., 2001; Hankin, Abramson, Miller, & Haeffel, 2004;
Joiner, 1995; Mor & Winquist, 2002).
The relation between self-esteem and anxiety has only rarely
been studied (J. E. Roberts, 2006). Cross-sectional studies have
reported negative, medium-sized to strong correlations between
the constructs (Lee & Hankin, 2009; Riketta, 2004; Watson et al.,
2002). However, we are not aware of any longitudinal study that
has explicitly focused on the prospective relation between self-
esteem and anxiety.
2
Several theories postulate that self-esteem
serves as a buffer against anxiety (see Crocker & Park, 2004). For
example, terror management theory (J. Greenberg et al., 1986;
Pyszczynski et al., 2004) suggests that self-esteem may predict a
decrease in subsequent anxiety because high self-esteem buffers
against anxiety elicited by awareness of human mortality. How-
ever, the opposite causal direction is also plausible; that is, expe-
riences of intense anxiety might leave scars in the self-concept that
persistently threaten and reduce self-esteem.
Theoretical Perspectives on the Relation of Low
Self-Esteem With Depression and Anxiety
There are two established theories that allow for hypotheses about
how depression and anxiety might be differentially related to self-
esteem. First, according to the tripartite model (e.g., Clark, Watson, &
Mineka, 1994), depression should exhibit a stronger relation to self-
esteem than does anxiety. The tripartite model states that depression
and anxiety share the feature of high negative affectivity, that is, a
stable disposition to experience nonspecific distress and unpleasant
mood. However, the model also states that each construct includes a
unique component, with low positive affectivity being specific to
depression and with heightened autonomic arousal being specific to
anxiety. Thus, whereas depression is linked to both positive affect and
negative affect, anxiety is linked to negative affect only. Given that
self-esteem is correlated with both positive and negative affect at
about similar effect size (Aspinwall & Taylor, 1992; Joiner, 1995;
Watson et al., 2002), the tripartite model suggests that low self-esteem
is more relevant for depression than for anxiety.
Second, the cognitive content hypothesis of A. T. Beck et al.
(1992), which was derived from Beck’s cognitive theory of de-
pression (A. T. Beck, 1967), posits that depression and anxiety can
be distinguished by specific cognitive vulnerabilities. The cogni-
tive content hypothesis states that depressive cognitions reflect
negative evaluations of the self, the world, and the future, whereas
anxious cognitions reflect the anticipation of a physical or psycho-
logical threat. Accordingly, low self-esteem should be a stronger
diathesis for depression than for anxiety.
The Present Research
The first goal of our study was to evaluate the vulnerability and
scar models of low self-esteem and depression by means of meta-
analysis. To increase the validity of conclusions, we analyzed
effect size measures that were (a) based on longitudinal data and
(b) controlled for prior levels of the predicted variable (i.e., con-
trolled for autoregressive effects). Controlling for prior levels of
the variables is of crucial importance, because it rules out the
possibility that prospective effects are simply due to concurrent
relations between the variables and the stability of the predicted
variable (Finkel, 1995). Figure 1 provides a generic illustration of
2
Although no previous study has focused explicitly on the prospective
relations between low self-esteem and anxiety, some longitudinal studies
have included information on zero-order correlations between the con-
structs, which we used to compute the effect sizes examined in the present
meta-analysis (see below).
217
LOW SELF-ESTEEM, DEPRESSION, AND ANXIETY
the effect size measures used, exemplary for the relation of self-
esteem with depression. First, we examined the stability (i.e.,
autoregressive) coefficients for each construct (e.g., the effect of
depression at Time 1 on depression at Time 2). Second, we
examined the cross-lagged coefficients between the constructs,
which are controlled for autoregressive effects (e.g., the effect of
self-esteem at Time 1 on depression at Time 2, controlling for
depression at Time 1). Third, for reasons of completeness, we also
examined the concurrent correlation between self-esteem and de-
pression, using the data from Time 1. Given the findings from
primary studies discussed above, we hypothesized that self-esteem
has a significant negative effect on subsequent depression (corre-
sponding to the vulnerability model) and that the effect of depres-
sion on subsequent self-esteem is nonsignificant or, if significant,
smaller than the self-esteem effect on depression.
In this context, it is important to distinguish between two related
approaches, specifically cross-lagged correlation analysis and
cross-lagged regression analysis. Cross-lagged correlation analysis
has been critiqued because cross-lagged correlations not only
reflect the prospective influence of the predictor on the outcome
but also depend on the stability of the outcome (Locascio, 1982;
Rogosa, 1980). Thus, cross-lagged correlations are confounded by
the stability of the variables and may result in misleading inter-
pretations. It is possible that a large cross-lagged correlation sim-
ply reflects high stability of the outcome, when the constructs
simultaneously show a strong concurrent correlation at Time 1. In
contrast, cross-lagged regression analysis statistically controls for
the stability of the variables. In other words, whereas cross-lagged
correlations inform about whether the predictor at Time 1 is related
to the outcome at Time 2, cross-lagged regressions inform about
whether the predictor at Time 1 is related to change in the outcome
between Time 1 and Time 2 (because the level of the outcome at
Time 1 is controlled for; see Finkel, 1995). Therefore, in this
research we used the cross-lagged regression approach, which
avoids the possible confounding effect of the stability of the
variables.
The second goal of our study was to examine whether anxiety is
related to low self-esteem much as depression is, or whether the
vulnerability and scar models (if supported by the results) are
specific for depression. We therefore meta-analyzed the available
longitudinal data on self-esteem and anxiety, examining the same
effect size measures as for self-esteem and depression. On
the basis of the theoretical perspectives discussed above, we ex-
pected weaker concurrent and cross-lagged relations of self-esteem
with anxiety than with depression. However, we had no hypothe-
ses on the relative strength of the cross-lagged effects between
self-esteem and anxiety (i.e., whether the self-esteem effect on
anxiety would be stronger than the anxiety effect on self-esteem).
The third goal of our study was to test for moderators of the
effect sizes. Because the number of studies was low for the relation
between self-esteem and anxiety, we focused exclusively on the
relation of self-esteem with depression (see the Results section for
further information). Although previous studies tested whether the
prospective relation between self-esteem and depression holds
across gender (Orth et al., 2008; Orth, Robins, Trzesniewski, et al.,
2009) and across different age groups from adolescence to old age
(Orth et al., 2008; Orth, Robins, Trzesniewski, et al., 2009; Shahar
& Henrich, 2010), the meta-analytic approach provides for a more
powerful and valid test of the moderating effects of gender and
age. We also tested whether the results hold across different types
of samples, most importantly representative and clinical samples.
Another important moderator might be the temporal design of the
primary studies. Methodologists have advised that it is necessary
to study different time lags between assessments to gain a “com-
plete understanding of a variable’s effect” (Gollob & Reichardt,
1987, p. 82). More specifically, Collins and Graham (2002) high-
lighted the importance of studying the influence of time lags on the
effect size when longitudinal studies are meta-analyzed. We there-
fore tested for the moderating impact of time lag: for example,
whether a minimum time lag is required to observe any prospec-
tive effect between self-esteem and depression or whether effect
sizes become smaller when assessed across long time intervals.
Finally, we tested whether the effects replicate across different
measures of self-esteem and depression or whether the effects are
methodological artifacts of specific measures of the constructs. In
summary, we tested whether gender, age, sample type, time lag
between assessments, and measures of self-esteem and depression
moderate the strength of the prospective effect of self-esteem on
depression.
The effect size coefficients examined in the present research
were based on continuous measures of self-esteem, depression,
and anxiety. The measures typically employed in this field include
multiple indicators and have good psychometric properties, legit-
imating the statistical approach used in this meta-analysis. With
regard to self-esteem, the most frequently used measures and their
psychometric properties have already been discussed above. With
regard to depression, frequently used measures are the Beck De-
pression Inventory (BDI; A. T. Beck, Ward, Mendelson, Mock, &
Erbaugh, 1961) and the Center for Epidemiologic Studies Depres-
sion Scale (CES-D; Gotlib, Lewinsohn, & Seeley, 1995; Radloff,
1977), both of which can be used in nonclinical, subclinical, and
clinical populations. The BDI is a self-report instrument compris-
ing 21 items; research suggests that the BDI is a valid and reliable
measure of depressive symptoms (A. T. Beck, Steer, & Carbin,
1988; Nezu, Nezu, Friedman, & Lee, 2009; Osman et al., 2004).
Similarly, the CES-D—a 20-item self-report measure—is a well-
validated and reliable measure of depressive symptoms (Eaton,
Depression
Time 2
Self-Esteem
Time 2
Self-Esteem
Time 1
Depression
Time 1
Figure 1. The figure illustrates the coefficients meta-analyzed in the
present research, exemplary for the relation between self-esteem and de-
pression (the coefficients for the relation between self-esteem and anxiety
were specified accordingly). The relations between the variables at the two
measurement occasions are specified as cross-lagged effects and stability
effects. The cross-lagged effects indicate the prospective effect of one
variable on the other (e.g., effect of self-esteem at Time 1 on depression at
Time 2), after controlling for their stabilities across time (e.g., effect of
depression at Time 1 on depression at Time 2). In addition to cross-lagged
and stability effects, we examined the cross-sectional correlation between
the constructs, for Time 1 as an example.
218
SOWISLO AND ORTH
Smith, Ybarra, Muntaner, & Tien, 2004; Shaver & Brennan,
1991). With regard to anxiety, two frequently used measures are
the Beck Anxiety Inventory (BAI; A. T. Beck, Epstein, Brown, &
Steer, 1988) and the State–Trait Anxiety Inventory (STAI; Spiel-
berger & Sydeman, 1994). Both measures are multi-item self-
report scales; the available research supports the reliability and
validity of the BAI (Fydrich, Dowdall, & Chambless, 1992; Os-
man, Kopper, Barrios, Osman, & Wade, 1997) and STAI (Barnes,
Harp, & Jung, 2002; Spielberger & Sydeman, 1994).
This meta-analysis extends the primary studies on self-esteem
and depression in several ways. First, prospective relations were
estimated with greater power and based on a wide variety of study
characteristics. Methodological concerns unique to each primary
study were thus reduced, and more valid indications for the direc-
tion of the relation between self-esteem and depression were
procured. Second, we tested whether the vulnerability and scar
models are specific for the relation of self-esteem with depression,
or whether similar relations exist with anxiety. As yet, no previous
study has tested whether low self-esteem prospectively predicts
anxiety or, vice versa, whether anxiety prospectively predicts low
self-esteem. Third, the meta-analytic approach enabled us to test
for moderators that are difficult to examine in primary studies,
such as sample type and time lag between assessments. For ex-
ample, in this study we tested whether the prospective effects
between self-esteem and depression systematically differ when
assessed across a few days, weeks, months, or several years.
Method
Selection of Studies
To search for relevant studies, we used three strategies. First,
English-language journal articles, books, book chapters, and dis-
sertations were searched in the databases PsycINFO and Medline
for all years covered through July 2011.
3
We used the following
search terms: depress*, dysphori*, dysthym*, anxi*, fear, phobi*,
self-esteem, self-worth, self-liking, self-respect, longitudinal, pro-
spective, and antecedent. The asterisk (i.e., the truncation symbol)
allowed for the inclusion of alternate word endings of the search
term (e.g., depress* yielded articles containing depression, depres-
sive, etc.). Second, we examined the information provided by
relevant review articles (A. T. Beck, 1987; Dance & Kuiper, 1987;
J. Greenberg et al., 1992; O’Brien et al., 2006; Pyszczynski et al.,
2004; J. E. Roberts, 2006). Third, we examined the reference
sections of all articles included in the meta-analysis. The search
resulted in 251 potentially relevant journal articles and 44 disser-
tations. There were no relevant books or book chapters.
We decided to include dissertations in our meta-analysis be-
cause dissertations are a category of unpublished studies that has
important advantages for examining publication bias (Ferguson &
Brannick, 2012; McLeod & Weisz, 2004). Dissertations are in-
dexed in databases and, consequently, allow for an exhaustive
search and avoidance of selection bias in sampling the relevant
studies. In contrast, it is not possible to exhaustively search for
other types of unpublished studies such as presentations at confer-
ences and unpublished manuscripts. The empirical findings by
Ferguson and Brannick (2012) suggest that the unpublished liter-
ature included in meta-analyses is frequently plagued by selection
bias because several mechanisms prevent meta-analysts from ob-
taining a random sample of, for example, unpublished manu-
scripts. It is important to note that dissertations, although indexed
in databases, are generally not subject to publication bias because
dissertations are frequently accepted by dissertation committees
even if the research reported in the dissertation did not yield
significant effects. Moreover, Ferguson and Brannick found that
effect sizes that are based on nonindexed unpublished studies are,
on average, closer to effect sizes based on published studies than
effect sizes based on dissertations. Ferguson and Brannick there-
fore concluded that dissertations are better suited for examining
publication bias than other types of unpublished studies.
All studies were then assessed in full text by the first author of
this meta-analysis. In addition, a random sample of 79 studies was
rated by the second author to obtain estimates of interrater agree-
ment. The interrater agreement on inclusion or exclusion in the
meta-analysis was high (⫽.97), and all diverging assessments
were discussed until consensus was reached.
4
Studies were included in the meta-analysis if the following
criteria were fulfilled: (a) self-esteem was assessed with an explicit
measure of global self-esteem, (b) depression and/or anxiety was
assessed with continuous measures of the constructs, (c) the study
used a longitudinal study design, (d) at least one of the constructs
(i.e., self-esteem or depression/anxiety) was assessed on at least
two measurement occasions, and (e) enough information was
given to compute effect sizes. We included samples of all age
groups in the meta-analysis, covering the full life span from
childhood to old age. If a sample was analyzed by more than one
study, only one study was included in the meta-analysis to ensure
independence of effect sizes. In these cases, we included the study
that provided the most comprehensive coding information and
excluded the other studies. Finally, studies were excluded if in-
consistent information for the computation of effect sizes was
given.
This procedure left 53 journal articles and seven dissertations
for analysis. The articles of Chen (1995); Colarossi and Eccles
(2003); S. J. Cox et al. (2006); Le, Tv, and Taylor (2007); Orth et
al. (2008); Ostrowsky (2007); Rueger (2011); Schroevers, Ran-
chor, and Sanderman (2003); and Steinberg, Karpinski, and Alloy
(2007) provided two relevant samples each; the article of Orth,
Robins, Trzesniewski, et al. (2009) provided 12 relevant samples;
thus our data set comprised 80 samples. Of these, 77 samples
provided information on the prospective relations between self-
esteem and depression, and 18 samples provided information on
the prospective relations between self-esteem and anxiety.
Coding of Studies
We coded the following data: sample size, country of origin,
mean age of participants, proportion of female participants, sample
type (i.e., representative, clinical, college students, or convenience
sample other than college students), time lag between assessments,
3
Although our search covered the entire time span indexed in these
databases until July 2011, the earliest eligible study was published in 1984
(see Results section).
4
The qualifications of the coders were as follows: The first author had
a master’s degree in psychology, and the second author had a PhD in
psychology.
219
LOW SELF-ESTEEM, DEPRESSION, AND ANXIETY
measure used to assess self-esteem, measure used to assess de-
pression and anxiety, and effect sizes.
A few studies did not report the exact mean age or the exact time
lag between assessments. Yet, when valid indicators were given in
the studies, we used this information to estimate the variables. For
example, if a study that examined a sample of undergraduate
students did not report the mean age, we estimated it to be 20 years
(as done by, e.g., Starr & Davila, 2008). To take another example,
if a study reported that the first assessment was conducted in the
third trimester of pregnancy, we estimated Time 1 as 2 months
before delivery. In the meta-analytic data set, only few data were
missing on moderator variables (i.e., 2.6%). We therefore used the
complete case analysis method (i.e., listwise deletion) to deal with
missing data in the moderator analyses (Pigott, 2009).
5
In some cases, effect sizes were directly reported in the
article (i.e., the standardized regression coefficients as shown in
Figure 1). However, in most cases we computed effect sizes
using the zero-order correlations between the variables (e.g.,
correlations between self-esteem assessed at Time 1, self-
esteem assessed at Time 2, depression assessed at Time 1, and
depression assessed at Time 2). For the computation, we used
the following equation (Cohen, Cohen, West, & Aiken, 2003, p.
68), which is applicable when a criterion variable (Y) is influ-
enced by two predictors (X
1
, X
2
):

Y1.2
⫽
r
Y1
⫺ r
Y2
r
12
1 ⫺ r
12
2
. (1)
Here 
Y1.2
is the standardized regression coefficient of X
1
predict
-
ing Y, controlling for the effect of X
2
(e.g., the effect of self-esteem
at Time 1 on depression at Time 2, controlling for depression at
Time 1); r
Y1
and r
Y2
are the zero-order correlations between each
predictor (X
1
, X
2
; e.g., self-esteem at Time 1, depression at Time
1) and the criterion (Y; e.g., depression at Time 2); and r
12
is the
correlation between the two predictors (X
1
and X
2
; e.g., the cross-
sectional correlation of self-esteem at Time 1 and depression at
Time 1). For studies that provided more than one effect size for
one of the coefficients examined (e.g., because more than one
measure of self-esteem was used), we averaged the correlations
and standardized regression coefficients, respectively, using Fish-
er’s Z
r
transformations.
All articles were coded by the first author of this meta-analysis.
In addition, a random sample of 33 studies was coded by the
second author to obtain estimates of interrater agreement. The
interrater agreement was high ( ⱖ .95 for categorical variables
and r ⱖ .99 for continuous variables). All diverging assessments
were discussed until consensus was reached.
Meta-Analytic Procedure
We made all computations with effect sizes using Fisher’s Z
r
transformations and using study weights with ⫽n ⫺ 3 (see
Lipsey & Wilson, 2001). For the computations, we used SPSS and
the SPSS macros written by Daniel Wilson (see Lipsey & Wilson,
2001, Appendix D).
We conducted the following preliminary analyses. First, we
searched for statistical outliers on effect size variables. Second, we
determined whether there was evidence of publication bias, that is,
whether studies with nonsignificant results had a lower probability
of being published. We hypothesized that publication bias would
not be an issue in this research, because the majority of studies
included in the meta-analysis did not focus specifically on the
relations of low self-esteem with depression and anxiety but re-
ported their intercorrelations together with intercorrelations among
a larger set of constructs. Nevertheless, we tested for publication
bias, using two methods. First, if publication bias exists, studies
resulting in low effect sizes should have a low probability of being
published if the sample size is small (because of a low probability
of significant findings). In contrast, studies resulting in large effect
sizes have a high probability of being published even if the sample
size is small (because of a high probability of significant findings).
The relationship of sample size and effect size can be examined
visually with a funnel graph (cf. Sutton, 2009). If the funnel graph
does not show a symmetrical shape, and if studies with small
sample size show a bias toward larger effect sizes, there is evi-
dence for publication bias. Second, we tested whether effect sizes
based on dissertations differed significantly from effect sizes based
on published studies.
In the effect size analyses, we used a random-effects model,
following the recommendations by Field and Gillett (2010) and
Raudenbush (2009). We first computed weighted mean effect sizes
and tested for homogeneity of effect size distributions. Then we
examined moderators of the effect sizes using multiple regression
analysis and analysis of variance. In multiple regression analysis,
only continuous or dichotomous predictors can be used; therefore,
we dichotomized the categorical variable sample type into a vari-
able contrasting representative versus nonrepresentative samples.
We decided to focus on this contrast because representative samples
provide more valid results compared with nonrepresentative samples.
Finally, using analysis of variance, we investigated the influence of
the variable sample type in more detail using all the original catego-
ries and also examined the moderating effects of the self-esteem and
depression measures used.
Results
Description of Studies
The 80 studies included in the meta-analysis were published
between 1984 and 2010, with the median in 2004. Sample sizes
varied between 44 and 6,813 (M ⫽ 447.5, SD ⫽ 1,050.2, Mdn ⫽
214.5). The average proportion of female participants was 64%
(range: 0%–100%). The average mean age of the participants at the
time of the first assessment was 27.7 years (SD ⫽ 17.4; range:
8.2–79.3). The time lag between assessments varied between 1
week and 13 years (M ⫽ 1.23 years, SD ⫽ 1.81, Mdn ⫽ 0.75).
Forty-nine studies used convenience samples other than college
students, 19 used college student samples, nine used representative
samples, and three used clinical samples. Sixty-two studies were
conducted in the United States, six in Germany, three in the United
5
We tested whether the results of the multiple regression analysis used
in the moderator analyses were altered when we used a different method to
deal with missing data (i.e., the expectation–maximization algorithm;
Dempster, Laird, & Rubin, 1977). The results were very similar, and all the
significant effects remained significant and the nonsignificant effects re-
mained nonsignificant.
220
SOWISLO AND ORTH
Kingdom, two in Canada, two in Israel, and one each in Australia,
China, Korea, the Netherlands, and Sweden. Tables 1 and 2 show
the basic sample characteristics and effect sizes for each study,
separated for depression and anxiety.
A wide variety of measures were used in the studies. Self-
esteem (k ⫽ 80) was assessed by the RSE (Rosenberg, 1965) in 61
studies, by the global self-worth subscale of the Self-Perception
Profile for Children (Harter, 1985) or the Self-Perception Profile
for Adolescents (Harter, 1988) in 11 studies, and by a range of
other measures in eight studies. Depression (k ⫽ 77) was assessed
by the CES-D (Radloff, 1977) or its child version (Weissman,
Orvaschel, & Padian, 1980) in 30 studies, by the BDI (A. T. Beck
et al., 1961) in 20 studies, by the Children’s Depression Inventory
(Kovacs, 1985) in eight studies, and by other measures in 19
studies. Anxiety (k ⫽ 18) was assessed by the BAI (A. T. Beck,
Epstein, et al., 1988) in five studies, by the STAI or its child
version (Spielberger & Sydeman, 1994) in three studies, by the
anxiety subscale of the Mood and Anxiety Symptom Question-
naire (Watson et al., 1995) in two studies, by the anxiety subscale
of the Hospital Anxiety and Depression Scale (Zigmond & Snaith,
1983) in two studies, and by other measures in six studies.
Preliminary Analyses
First, the data revealed that there were no statistical outliers on
effect size variables. We therefore used the complete data set for
the subsequent analyses.
Second, the data showed evidence against any publication bias.
For each effect size, the funnel graphs indicated that studies with
small sample sizes were not biased toward larger effect sizes (see
Figures 2 and 3). The distributions of effect sizes exhibited a
symmetrical shape typical of nonbiased meta-analytic data sets.
Moreover, we tested whether effect sizes based on dissertations
differed significantly from effect sizes based on published studies.
These tests were possible only for effect sizes related to depres-
sion, but not anxiety, because only one dissertation related to
anxiety was included in the data set. The tests showed that there
were no significant differences between dissertations and pub-
lished studies (all ps ⬎ .05).
Effect Size Analyses
We computed weighted mean effect sizes for the relation be-
tween self-esteem and depression and for the relation between
self-esteem and anxiety. More specifically, we examined the cross-
sectional correlation between the constructs (for Time 1 as an
example), the stability coefficients of the constructs, and the cross-
lagged effects between the constructs (cf. Figure 1). Tables 3 and
4 show the results for depression and anxiety, respectively. As
reported in the tables, homogeneity statistics were significant for
most effect sizes, except for the cross-lagged effects between
self-esteem and anxiety. A significant homogeneity figure indi-
cates that the variance of the corresponding effect size must be
attributed not only to within-study sampling error but also to
between-study sampling error.
The results for depression supported the vulnerability model of
low self-esteem (see Table 3). The mean cross-lagged effect of
self-esteem on depression was ⫺.16 ( p ⬍ .05) and was larger than
the mean cross-lagged effect of depression on self-esteem, which
was ⫺.08 ( p ⬍ .05). No formal significance test for the difference
between cross-lagged effects is available; however, the confidence
intervals, which showed no overlap, clearly suggest that the two
cross-lagged effects differ from each other.
6
Moreover, the mean
stability coefficient of self-esteem was larger than the mean sta-
bility coefficient of depression. Because the cross-lagged effect of
self-esteem on depression was based on a much larger number of
studies (k ⫽ 77) than the cross-lagged effect of depression on
self-esteem (k ⫽ 42), we repeated the computation of the self-
esteem effect on depression using the same set of studies that was
used for the computation of the depression effect on self-esteem.
However, the self-esteem effect on depression was virtually unal-
tered (with a weighted mean effect size at ⫺.16). Similarly, the
stability of depression was virtually unaltered when computed with
the smaller set of studies (with a weighted mean effect size at .51).
The results for anxiety suggested a symmetric reciprocal rela-
tion between self-esteem and anxiety (see Table 4). The cross-
lagged effects between the constructs were significant and of
similar size (⫺.10 and ⫺.08; both ps ⬍ .05). As stated above, no
formal significance test is available for the difference between the
coefficients; however, the confidence intervals overlapped widely,
which suggests that the coefficients do not significantly differ.
Again, we tested whether the results for the self-esteem effect on
anxiety differed when computed with the smaller set of studies
(k ⫽ 10) used to compute the anxiety effect on self-esteem.
However, the self-esteem effect on anxiety was virtually unaltered
(with a weighted mean effect size at ⫺.10). Similarly, the stability
of anxiety was virtually unaltered when computed with the smaller
set of studies (with a weighted mean effect size at .47).
Moderator Analyses
The preceding analyses revealed heterogeneity of the distribu-
tions of most effect sizes; therefore, we investigated whether
moderator variables explain variation of effect sizes. In the mod-
erator analyses, we focused on the effect of self-esteem on depres-
sion for several reasons. First, the effect emerged as the strongest
cross-lagged effect in the effect size analyses and is of central
importance for the vulnerability model. Second, the number of
studies on which the other cross-lagged effects were based was
relatively low (i.e., 42, 18, and 10 studies), which limited the
statistical power of moderator analysis.
We first examined the simple correlations between the effect
size and the moderator variables (see Table 5). The results showed
6
No formal significance test for the difference between the cross-lagged
regression effects is available because the coefficients (a) do not involve
the same set of variables and (b) are based on a partially, but not fully,
overlapping set of studies. In this situation, none of the tests discussed in,
for example, Clogg, Petkova, and Haritou (1995); Cohen et al. (2003); and
Raghunathan, Rosenthal, and Rubin (1996) is applicable. We therefore
used the confidence intervals as an approximate means of comparing the
cross-lagged effects. For comparison purposes (although formally not
admissible), we also computed unpaired t tests, which bolstered our con-
clusions. For the cross-lagged effects between self-esteem and depression,
the test was significant, with t(117) ⫽ 3.78, p ⬍ .001, suggesting that the
coefficients differed significantly. For the cross-lagged effects between
self-esteem and anxiety, the test was nonsignificant, with t(26) ⫽ 1.18, p ⫽
.250, suggesting that the coefficients did not significantly differ.
221
LOW SELF-ESTEEM, DEPRESSION, AND ANXIETY
Table 1
Longitudinal Studies of the Relation Between Self-Esteem (SE) and Depression (D)
Study
Sample characteristics Effect sizes
N
Proportion
of female
participants
Mean age
(years)
Time lag
(years) Sample type r
SE,D
SE3D
a
D3SE
a
SE3SE
a
D3D
a
Abela & Payne (2003) 314 .45 11.2 0.11 Convenience ⫺.50 ⫺.20 .58
Bohon et al. (2008) 496 1.00 16.5 1.00 Convenience ⫺.34 ⫺.09 ⫺.06 .27 .61
Borelli & Prinstein (2006) 478 .51 12.7 0.92 Convenience ⫺.59 ⫺.06 .67
Burwell & Shirk (2006) 110 .58 13.6 0.58 Convenience ⫺.45 ⫺.30 ⫺.21 .48 .43
Butler et al. (1994), Part 2 73 .77 20.0 0.42 College students ⫺.36 ⫺.11 .36
Cambron et al. (2010), Study 3 230 .68 21.3 0.04 College students ⫺.61 ⫺.21 ⫺.16 .69 .64
Cast & Burke (2002) 574 .50 1.00 Convenience ⫺.31 ⫺.04 ⫺.20 .57 .46
Chen (1995), female subsample 374 1.00 1.00 Convenience ⫺.59 ⫺.14 ⫺.09 .72 .55
Chen (1995), male subsample 374 .00 1.00 Convenience ⫺.50 ⫺.10 ⫺.11 .63 .62
S. K. Cheng & Lam (1997) 286 .27 15.8 0.25 Convenience ⫺.64 ⫺.26 .49
Cikara & Girus (2010) 67 .63 20.9 0.08 College students ⫺.65 ⫺.26 ⫺.19 .74 .52
Colarossi & Eccles (2003), female
group 125 1.00 17.0 1.00 Convenience ⫺.52 ⫺.21 .19 .81 .46
Colarossi & Eccles (2003), male
group 92 .00 17.0 1.00 Convenience ⫺.52 ⫺.19 .00 .60 .44
Conley et al. (2001) 147 8.2 0.06 Convenience ⫺.45 ⫺.19 .59
Fernandez et al. (1998) 729 .52 2.00 Convenience ⫺.45 ⫺.14 .34
Flynn (2006) 160 .73 19.5 College students ⫺.69 ⫺.15 .57
Fontaine & Jones (1997) 45 1.00 31.0 0.21 Convenience ⫺.64 ⫺.39 .17
Hobfoll & Leiberman (1987) 99 1.00 28.0 0.25 Convenience ⫺.47 ⫺.15 ⫺.21 .48 .39
Hobfoll & Walfisch (1984) 55 1.00 38.2 0.25 Convenience ⫺.23 ⫺.10 .49
Hubbs-Tait et al. (1994), group of
mothers 44 1.00 17.7 3.42 Convenience ⫺.68 ⫺.59 ⫺.04 .70 ⫺.01
Jalajas (1994) 205 .38 23.4 0.25 College students ⫺.65 ⫺.23 ⫺.14 .72 .55
Joiner (1995), group of targets 100 .61 20.0 0.06 College students ⫺.62 ⫺.11 ⫺.02 .58 .40
Joiner et al. (1999) 177 .63 20.0 0.06 College students ⫺.58 ⫺.17 ⫺.02 .61 .39
Joiner et al. (2000) 143 .59 20.0 0.06 College students ⫺.56 ⫺.13 ⫺.03 .58 .45
Kakihara et al. (2010) 1,022 .47 15.3 1.00 Convenience ⫺.66 ⫺.15 ⫺.13 .54 .50
Katz et al. (1998), female group 134 1.00 19.0 0.11 Convenience ⫺.67 ⫺.24 .39
Kernis et al. (1998) 98 .88 20.0 0.08 College students ⫺.50 ⫺.04 .72
Kim et al. (2008) 60 1.00 31.8 0.25 Clinical ⫺.76 ⫺.36 .26
Klima & Repetti (2008) 226 .48 9.5 2.00 Convenience ⫺.67 ⫺.14 ⫺.33 .33 .51
Kling et al. (2003) 285 1.00 69.5 1.17 Convenience ⫺.33 ⫺.20 ⫺.10 .71 .42
Le et al. (2007), female group 6,813 1.00 16.0 1.00 Representative ⫺.46 ⫺.06 .60
Le et al. (2007), male group 6,504 .00 16.0 1.00 Representative ⫺.41 ⫺.08 .59
Lee & Hankin (2009) 350 .57 14.5 0.42 Convenience ⫺.60 ⫺.03 .68
Lewinsohn et al. (1988) 562 0.69 Convenience ⫺.51 ⫺.14 .35
McCarty et al. (2007) 331 .47 12.0 1.00 Clinical ⫺.60 ⫺.21 ⫺.07 .52 .46
Mindes et al. (2003) 67 1.00 33.0 0.75 Convenience ⫺.57 .05 ⫺.19 .61 .61
Ohannessian et al. (1994) 235 .56 12.2 1.00 Convenience ⫺.31 ⫺.13 .29
Orth et al. (2008), Study 1 2,403 .50 15.5 2.00 Representative ⫺.34 ⫺.09 ⫺.04 .51 .51
Orth et al. (2008), Study 2 359 .59 18.3 1.00 College students ⫺.60 ⫺.20 .00 .80 .35
Orth, Robins, Trzesniewski, et al.
(2009), Study 1, age 18–29 95 .58 22.0 3.00 Convenience ⫺.82 ⫺.44 .18 .89 .32
Orth, Robins, Trzesniewski, et al.
(2009), Study 1, age 30–39 673 .57 35.5 3.00 Convenience ⫺.70 ⫺.22 .01 .80 .45
Orth, Robins, Trzesniewski, et al.
(2009), Study 1, age 40–49 146 .45 41.9 3.00 Convenience ⫺.72 ⫺.23 ⫺.05 .84 .62
Orth, Robins, Trzesniewski, et al.
(2009), Study 1, age 50–59 270 .70 56.2 3.00 Convenience ⫺.59 ⫺.25 .02 .80 .37
Orth, Robins, Trzesniewski, et al.
(2009), Study 1, age 60–69 299 .49 63.4 3.00 Convenience ⫺.71 ⫺.23 .02 .90 .55
Orth, Robins, Trzesniewski, et al.
(2009), Study 1, age 70⫹ 202 .58 79.3 3.00 Convenience ⫺.44 ⫺.29 .11 .89 .29
Orth, Robins, Trzesniewski, et al.
(2009), Study 2, age 18–29 371 .51 24.5 2.00 Representative ⫺.77 ⫺.10 ⫺.09 .64 .64
222
SOWISLO AND ORTH
that proportion of female participants, mean age of participants,
and sample type did not significantly correlate with the effect size.
Only time lag showed a significant zero-order correlation with the
effect size (r ⫽⫺.25, p ⫽ .03). To control for multicollinearity of
the predictors, we computed a multiple regression analysis with
these variables as predictors of the effect size (see Table 5). The
variance explained was relatively small, and only one predictor
(i.e., sample type) yielded a significant regression weight, indicat-
ing that the self-esteem effect on depression was smaller in rep-
resentative than in nonrepresentative samples.
7
In a second mul
-
tiple regression analysis, we also tested whether time lag showed
a significant quadratic relation to effect size by additionally in-
cluding the square of the variable (the variable time lag was
centered for the analysis). However, neither the linear nor the
quadratic term was significant. Figure 4 further illustrates the
relation between time lag and effect size, showing that no linear or
7
In analyses of the moderating effect of time lag, we excluded one statistical
outlier. In the study by Schafer, Wickrama, and Keith (1998), the time lag between
assessments was 13 years, which was 6.5 standard deviations longer than the
average time lag. When this study was included in the multiple regression analysis,
time lag significantly predicted the effect size (⫽⫺.26, p ⫽ .02), indicating that
with increasing time lag the effect size became more negative (i.e., the absolute
value of the effect became larger). However, because this analysis was strongly
influenced by the statistical outlier (Cohen et al., 2003), we decided to report the
analyses without the outlier.
Table 1 (continued)
Study
Sample characteristics Effect sizes
N
Proportion
of female
participants
Mean age
(years)
Time lag
(years) Sample type r
SE,D
SE3D
a
D3SE
a
SE3SE
a
D3D
a
Orth, Robins, Trzesniewski, et al.
(2009), Study 2, age 30–39 437 .49 34.4 2.00 Representative ⫺.84 ⫺.23 ⫺.06 .81 .62
Orth, Robins, Trzesniewski, et al.
(2009), Study 2, age 40–49 476 .45 44.5 2.00 Representative ⫺.75 ⫺.17 .05 .92 .63
Orth, Robins, Trzesniewski, et al.
(2009), Study 2, age 50–59 545 .34 54.9 2.00 Representative ⫺.78 ⫺.16 ⫺.01 .80 .68
Orth, Robins, Trzesniewski, et al.
(2009), Study 2, age 60–69 434 .29 64.0 2.00 Representative ⫺.62 ⫺.22 ⫺.04 .82 .61
Orth, Robins, Trzesniewski, et al.
(2009), Study 2, age 70⫹ 216 .34 74.0 2.00 Representative ⫺.72 ⫺.02 .03 .96 .83
Ostrowsky (2007), female
subsample 253 1.00 14.0 1.00 Convenience ⫺.46 .01 ⫺.11 .44 .64
Ostrowsky (2007), male
subsample 675 .00 14.0 1.00 Convenience ⫺.43 ⫺.14 ⫺.13 .55 .49
Prinstein & La Greca (2002) 246 .60 16.8 6.00 Convenience ⫺.54 ⫺.15 ⫺.05 .29 .20
Procopio et al. (2006) 150 1.00 45.2 2.50 Convenience ⫺.63 ⫺.28 ⫺.36 .37 .48
Puckett (2010) 345 .58 14.0 0.50 Convenience ⫺.75 ⫺.07 .63
Ralph & Mineka (1998) 141 .54 20.0 0.02 College students ⫺.52 ⫺.09 .49
Ritter et al. (2000) 191 1.00 24.5 0.63 Convenience ⫺.49 ⫺.07 .46
J. E. Roberts & Kassel (1997) 213 .63 20.3 0.17 College students ⫺.58 ⫺.09 ⫺.13 .65 .52
J. E. Roberts & Monroe (1992) 192 .64 18.7 0.08 College students ⫺.51 ⫺.21 .65
Robinson et al. (1995) 381 .58 12.0 0.38 Convenience ⫺.71 ⫺.22 .50
Rosario et al. (2005) 156 .49 18.3 0.50 Convenience ⫺.62 ⫺.52 .02
Rueger (2011), female subsample 256 1.00 13.2 0.33 Convenience ⫺.67 ⫺.03 .57
Rueger (2011), male subsample 241 .00 13.2 0.33 Convenience ⫺.64 ⫺.19 .41
Schafer et al. (1998) 98 .50 56.0 13.00 Convenience ⫺.38 ⫺.38 .38
Schroevers et al. (2003), control
group 225 .70 57.0 1.00 Convenience ⫺.33 ⫺.13 .51
Schroevers et al. (2003), study
group 403 .73 58.0 1.00 Convenience ⫺.37 ⫺.10 .64
Settles et al. (2009) 128 1.00 24.2 2.00 Convenience ⫺.49 ⫺.14 .01 .50 .33
Shahar & Davidson (2003) 260 .43 42.2 0.33 Clinical ⫺.69 .01 ⫺.20 .62 .70
Southall & Roberts (2002) 115 .50 16.5 0.04 Convenience ⫺.69 ⫺.21 .47
Steinberg et al. (2007), high-risk
group 98 .61 20.0 0.34 College students ⫺.41 ⫺.01 .55
Steinberg et al. (2007), low-risk
group 83 .61 20.0 0.34 College students ⫺.52 .07 .66
Terry et al. (1996) 185 1.00 27.5 0.25 Convenience ⫺.37 ⫺.14 .45
Thoms (2006) 91 .81 21.2 0.13 College students ⫺.55 ⫺.19 ⫺.30 .52 .41
Vohs et al. (2001) 70 1.00 20.0 0.10 College students ⫺.45 .00 .58
Whisman & Kwon (1993) 80 .66 18.9 0.25 College students ⫺.77 .01 .69
Yang (2006) 1,149 .62 71.0 6.00 Convenience ⫺.36 ⫺.19 .38
Note. r
SE,D
⫽ correlation between the constructs at Time 1.
a
Standardized regression coefficient.
223
LOW SELF-ESTEEM, DEPRESSION, AND ANXIETY
curvilinear relation is discernible in the data. Thus, given that in
the multiple regression analysis only one significant predictor of
the effect size was identified, the important conclusion in this
context is that the vulnerability effect of low self-esteem on
depression replicated across samples with different gender and age
compositions and across different time lags between assessments.
Because sample type was used only as a dichotomous variable
in the preceding analysis, we computed an analysis of variance to
investigate the importance of sample type in more detail (see Table
6). Although the effect size differed for representative and non-
representative samples in the analysis reported above, the results of
the analysis of variance indicated that the self-esteem effect on
depression was present in all sample types (ranging from ⫺.12 to
⫺.19; all ps ⬍ .05).
Finally, we examined whether the effect size differed across
measures of self-esteem and depression, using analyses of vari-
ance. Table 7 shows that the effect size was very similar across
self-esteem measures (ranging from ⫺.15 to ⫺.18) and that there
was no significant heterogeneity between measures (Q
between
⫽
0.39, p ⫽ .821). Likewise, Table 8 shows that the effect size was
relatively similar across depression measures (ranging from ⫺.14
to ⫺.20), and again that the heterogeneity between measures was
nonsignificant (Q
between
⫽ 2.30, p ⫽ .531).
Together, the findings of the moderator analyses suggest that
low self-esteem serves as a general, stable risk factor for depres-
sion: the effect holds for samples with different gender and age
compositions, for different time lags, for different measures of
self-esteem and depression, and for representative, clinical, and
convenience samples.
Discussion
We investigated the prospective reciprocal relations of self-
esteem with depression and anxiety by meta-analyzing 77 longi-
tudinal studies providing information on the relation between
self-esteem and depression and 18 longitudinal studies providing
information on the relation between self-esteem and anxiety. The
studies included differed substantially with respect to sample char-
acteristics such as sample size, country of origin, sample type,
mean age of participants, and proportion of female participants.
Moreover, the studies differed significantly with respect to meth-
odological characteristics, such as the time lag between assess-
ments, and used a wide variety of measures to assess self-esteem,
depression, and anxiety. The heterogeneity of the studies strength-
ens the generalizability of the findings: First, the analyses yielded
consistent support for the vulnerability model of low self-esteem
and depression (i.e., low self-esteem contributes to depression) and
only weak support for the scar model (i.e., depression erodes
self-esteem). Second, the findings indicate that the relation be-
tween low self-esteem and anxiety is more symmetric, with small,
but significant, prospective effects in both directions. Third, mod-
erator analyses of the vulnerability effect of low self-esteem on
depression suggested that this effect is not significantly influenced
by gender and age composition of the sample, measures of self-
esteem and depression, or the time lag between assessments.
Moreover, although the vulnerability effect differed significantly
between representative, clinical, and convenience samples, the
effect was present in all types of samples examined in this re-
search.
Table 2
Longitudinal Studies of the Relation Between Self-Esteem (SE) and Anxiety (ANX)
Study
Sample characteristics Effect sizes
N
Proportion
of female
participants
Mean age
(years)
Time lag
(years) Sample type r
SE,ANX
SE3ANX
a
ANX3SE
a
SE3SE
a
ANX3ANX
a
Borelli & Prinstein (2006) 478 .51 12.7 0.92 Convenience ⫺.46 ⫺.02 .61
Cast & Burke (2002) 574 .50 1.00 Convenience ⫺.18 ⫺.11 ⫺.14 .60 .48
S. J. Cox et al. (2006),
IVF group 70 1.00 33.6 0.19 Convenience ⫺.67 ⫺.06 ⫺.01 .80 .70
S. J. Cox et al. (2006),
control group 111 1.00 29.3 0.19 Convenience ⫺.40 .02 ⫺.22 .63 .55
Ewen (2002) 115 .81 31.3 0.67 College students ⫺.66 ⫺.24 ⫺.15 .65 .42
Hobfoll & Walfisch
(1984) 55 1.00 38.2 0.25 Convenience ⫺.05 ⫺.22 .30
Jalajas (1994) 205 .38 23.4 0.25 College students ⫺.38 ⫺.15 ⫺.07 .78 .46
Joiner (1995) 100 .61 20.0 0.06 College students ⫺.31 ⫺.15 ⫺.02 .58 .32
Joiner et al. (1999) 177 .63 20.0 0.06 College students ⫺.42 ⫺.07 .11 .67 .41
Kim et al. (2008) 60 1.00 31.8 0.25 Clinical ⫺.38 ⫺.31 .61
Lee & Hankin (2009) 350 .57 14.5 0.42 Convenience ⫺.53 ⫺.21 .45
McCarty et al. (2007) 331 .47 12.0 1.00 Clinical ⫺.37 ⫺.04 ⫺.08 .53 .53
Ohannessian et al. (1994) 235 .56 12.2 1.00 Convenience ⫺.41 ⫺.07 .25
Prinstein & La Greca
(2002) 246 .60 10.8 6.00 Convenience ⫺.31 ⫺.05 .00 .31 .31
Procopio et al. (2006) 150 1.00 45.2 2.50 Convenience ⫺.33 ⫺.17 ⫺.14 .50 .48
Ralph & Mineka (1998) 141 .54 20.0 0.02 College students ⫺.30 .01 .49
Rosario et al. (2005) 156 .49 18.3 0.50 Convenience ⫺.35 ⫺.10 .50
Vohs et al. (2001) 70 1.00 20.0 0.10 College students ⫺.45 ⫺.13 .51
Note. r
SE,ANX
⫽ correlation between the constructs at Time 1; IVF ⫽ in vitro fertilization.
a
Standardized regression coefficient.
224
SOWISLO AND ORTH
Implications of the Findings
The present results suggest that the prospective relation be-
tween low self-esteem and depression is best described by the
vulnerability model, whereas the prospective relation between
low self-esteem and anxiety is best described as a symmetric
reciprocal relation. Consequently, it would be interesting to
gain further insight into (a) the mechanisms that account for the
vulnerability effect of low self-esteem on depression, (b) the
mechanisms that account for the small but significant scar
effect of depression on self-esteem, and (c) the mechanisms that
.00-.20-.40-.60-.80-1.00
Sample size
7000
6000
5000
4000
3000
2000
1000
0
A
Stability effect of depression
1.00.80.60.40.20.00-.20
Sample size
7000
6000
5000
4000
3000
2000
1000
0
Effect of self-esteem on depression
.40.20.00-.20-.40-.60
Sample size
7000
6000
5000
4000
3000
2000
1000
0
B
Correlation between self-esteem and depression at Time 1
D
1.00
1
.
00
Stability effect of self-esteem
.80.60.40.20.00-.20
7000
6000
5000
4000
3000
2000
1000
0
Effect of depression on self-esteem
.40.20.00-.20-.40-.60
Sample sizeSample size
7000
6000
5000
4000
3000
2000
1000
0
C
E
Figure 2. Funnel graphs for the effect sizes of the relation between self-esteem and depression. The graphs
display the relation between the effect size and sample size of the studies. The dashed lines show the weighted
mean effect sizes.
225
LOW SELF-ESTEEM, DEPRESSION, AND ANXIETY
account for the affective specificity of the results. Knowledge
about mediating processes is of crucial importance because it
provides for possible starting points for interventions, for in-
stance, interventions aimed at preventing or reducing depres-
sion.
The vulnerability effect of low self-esteem on depression might
operate through both interpersonal and intrapersonal psychological
pathways. One interpersonal pathway is that some individuals with
low self-esteem might excessively seek reassurance from friends
and relationship partners, which might lead to social disruptions
A
Correlation between self-esteem and anxiety at Time 1
.00-.20-.40-.60-.80-1.00
Sample size
600
500
400
300
200
100
0
.40
C
Effect of anxiety on self-esteem
.20.00-.20-.40
Sample size
600
500
400
300
200
100
0
1.00
Sample size
E
Stability effect of self-esteem
.80.60.40.20.00
600
500
400
300
200
100
0
1.00
B
D
Effect of self-esteem on anxiety
.40.20.00-.20-.40
Sample size
600
500
400
300
200
100
0
Stability effect of anxiety
.80.60.40.20.00
Sample size
600
500
400
300
200
100
0
Figure 3. Funnel graphs for the effect sizes of the relation between self-esteem and anxiety. The graphs display
the relation between the effect size and sample size of the studies. The dashed lines show the weighted mean
effect size.
226
SOWISLO AND ORTH
that in turn foster depressive symptoms (Joiner, Alfano, & Metal-
sky, 1992; Potthoff, Holahan, & Joiner, 1995). A second interper-
sonal pathway is that individuals with low self-esteem seek neg-
ative feedback from their relationship partners to verify their
negative self-concept. Negative feedback seeking might lead to
rejection by close others and might undermine social support,
which in turn increases the risk of depression (Giesler, Josephs, &
Swann, 1996; Joiner, Katz, & Lew, 1997; Swann, Wenzlaff, &
Tafarodi, 1992). An intrapersonal pathway explaining how low
self-esteem contributes to depression might operate through self-
focused attention (Mor & Winquist, 2002). Individuals with low
self-esteem are prone to ruminating about negative aspects of the
self, which in turn increases depression (e.g., Nolen-Hoeksema,
2000; Spasojevic´ & Alloy, 2001). Overall, reassurance seeking,
negative feedback seeking, and rumination are theoretically linked
to low self-esteem and depression, and there is some evidence that
self-esteem contributes to these processes (Evraire & Dozois,
2011; Joiner, Katz, & Lew, 1999; Kuster, Orth, & Meier, 2012),
making it less likely that reassurance seeking, negative feedback
seeking, and rumination are third variables that cause the relation
between low self-esteem and depression. However, the hypothe-
sized mediational pathways should be tested directly. As yet, only
one study has identified a mediator of the vulnerability effect of
low self-esteem on depression. Using longitudinal mediation anal-
ysis (Cole & Maxwell, 2003), Kuster et al. (2012) found that
rumination partially mediated the prospective effect of low self-
esteem on depression across several waves of data. Future research
should continue to test for possible mediators of the vulnerability
effect, such as reassurance seeking or negative feedback seeking.
Similarly, the small but significant scar effect of depression on
self-esteem might unfold through interpersonal and intrapersonal
psychological pathways. One interpersonal pathway is that depres-
sive episodes may cause damage to important sources of self-
esteem such as close relationships or social networks. A second
interpersonal pathway is that depression might change how the
individual is perceived by others; these representations may be
relatively persistent and may cause the individual to be treated by
others with low regard or in ways that minimize the individual’s
self-esteem, even if the depression has already remitted (Joiner,
2000). A possible intrapersonal pathway is that the experience of
depression might influence self-esteem by persistently altering the
way in which individuals process self-relevant information; for
example, the chronic negative mood associated with depression
may lead the individual to selectively attend to, encode, and
retrieve negative information about the self, resulting in the for-
mation of more negative self-evaluations.
In addition to the psychological pathways through which the vul-
nerability effect (and the small scar effect) might operate, biological
factors might play a role. As yet, there is little knowledge about the
possible biological mechanisms underlying self-esteem and underly-
ing its association with psychological adjustment (cf. Pruessner et al.,
2005; Putnam & McSweeney, 2008). Generally, self-esteem shows a
genetic component, with heritability estimates ranging widely, from
29% to 73% (cf. Saphire-Bernstein, Way, Kim, Sherman, & Taylor,
Table 3
Summary of Effect Sizes for Relation Between Self-Esteem (SE) and Depression (D)
Variable kN
Weighted mean
effect size 95% CI Homogeneity (Q)
r
SE,D
77 35,501 ⫺.57
ⴱ
[⫺.60, ⫺.54] 1338.79
ⴱ
SE3D
a
77 35,501 ⫺.16
ⴱ
[⫺.18, ⫺.14] 226.63
ⴱ
D3SE
a
42 14,049 ⫺.08
ⴱ
[⫺.11, ⫺.05] 115.05
ⴱ
SE3SE
a
42 14,049 .69
ⴱ
[.63, .74] 1692.57
ⴱ
D3D
a
77 35,501 .51
ⴱ
[.48, .54] 758.05
ⴱ
Note. Computations were made with a random-effects model. k ⫽ number of studies; N ⫽ total number of
participants in the k samples; r
SE,D
⫽ correlation between the constructs at Time 1; CI ⫽ confidence interval.
a
Standardized regression coefficient.
ⴱ
p ⬍ .05.
Table 4
Summary of Effect Sizes for Relation Between Self-Esteem (SE) and Anxiety (ANX)
Variable kN
Weighted mean
effect size 95% CI Homogeneity (Q)
r
SE,ANX
18 3,597 ⫺.40
ⴱ
[⫺.46, ⫺.33] 83.01
ⴱ
SE3ANX
a
18 3,597 ⫺.10
ⴱ
[⫺.14, ⫺.06] 20.84
ANX3SE
a
10 2,052 ⫺.08
ⴱ
[⫺.13, ⫺.02] 13.83
SE3SE
a
10 2,052 .62
ⴱ
[.53, .69] 80.00
ⴱ
ANX3ANX
a
18 3,597 .47
ⴱ
[.42, .52] 63.85
ⴱ
Note. Computations were made with a random-effects model. k ⫽ number of studies; N ⫽ total number of
participants in the k samples; r
SE,ANX
⫽ correlation between the constructs at Time 1; CI ⫽ confidence interval.
a
Standardized regression coefficient.
ⴱ
p ⬍ .05.
227
LOW SELF-ESTEEM, DEPRESSION, AND ANXIETY
2011). More specifically, biological variables that have been associ-
ated with low self-esteem and depression include reduced hippocam-
pal volume (Pruessner et al., 2005), higher cortisol stress response
(Pruessner, Hellhammer, & Kirschbaum, 1999), specific patterns of
prefrontal electroencephalography alpha activity (De Raedt, Franck,
Fannes, & Verstraeten, 2008; Putnam & McSweeney, 2008), varia-
tions in the oxytocin receptor gene (Saphire-Bernstein et al., 2011),
and reduced cardiac vagal tone (Martens, Greenberg, & Allen, 2008).
Future research should examine whether these factors contribute (e.g.,
as third variables, moderators, or mediators) to the explanation of the
effect of low self-esteem on depression (for an example, see Scarpa &
Luscher, 2002).
We can only speculate as to why depression and anxiety are
differentially linked to low self-esteem. Divergent mediating
mechanisms might provide an explanation. For example, self-
focused attention is differentially related to depression and anxiety.
First, self-focused attention is more strongly related to depression
than to anxiety (Mor & Winquist, 2002), and if self-focused
attention is a mediator of the vulnerability effect, it might account
for the stronger effect of low self-esteem on depression than on
anxiety. Second, depression is more strongly related to a focus on
private aspects of the self, whereas anxiety is more strongly related
to public aspects of the self (Mor & Winquist, 2002). If the
vulnerability effect of low self-esteem is mediated more strongly
by private self-focus than by public self-focus, this might provide
a further explanation for the diverging effects on depression and
anxiety. Third, given that the evidence suggests that self-focused
attention has a reciprocal relation with depression and anxiety
(Mor & Winquist, 2002), self-focused attention might also account
for the small, but significant, reverse effects (i.e., the scar effects
of depression and anxiety on self-esteem).
Another mechanism that might account for the divergent rela-
tions of self-esteem with depression and anxiety is that excessive
reassurance seeking might lead to increases in depressive, but not
in anxious, symptoms (Joiner & Schmidt, 1998). Consequently, if
excessive reassurance seeking is a mediator of the vulnerability
effect, it might at least partially explain why low self-esteem has
stronger predictive effects on depression than on anxiety. Future
research should therefore explore the mediating mechanisms of the
relation between self-esteem and depression and anxiety from a
perspective of specificity: Which common pathways mediate the
vulnerability effect of low self-esteem on both depression and
anxiety? And which additional unique pathways explain that low
self-esteem has a stronger effect on depression than on anxiety?
The results suggest that the strength of the vulnerability effect of
low self-esteem on depression is not moderated by gender and age.
Thus, although the mean levels of self-esteem and depression vary as
a function of gender (Hyde, Mezulis, & Abramson, 2008; Kling,
Hyde, Showers, & Buswell, 1999) and age (Kessler, Foster, Webster,
& House, 1992; Lewinsohn, Rohde, Seeley, & Fischer, 1991; Orth et
al., 2010, 2012; Robins et al., 2002), the structural relations between
self-esteem and depression are unaffected by gender and age. The
present meta-analytic findings, which are based on study-level data,
are consistent with the findings from primary studies that suggested
that the vulnerability effect of low self-esteem holds across gender
(Orth et al., 2008; Orth, Robins, Trzesniewski, et al., 2009) and
replicates across age groups from young adulthood to old age (Orth,
Robins, Trzesniewski, et al., 2009; but see the findings on adolescent
age groups by Shahar & Henrich, 2010). From a theoretical perspec-
tive, the evidence that the effect of low self-esteem on subsequent
depression operates independently from gender and age is in line with
the vulnerability model, which states that low self-esteem is a global
risk factor for depression. In particular, we note that the vulnerability
effect was present not only in samples of adolescents and adults
(which represent the majority of the samples examined in this re-
search) but also in samples of children.
8
Major depression in child
-
hood is a concern (although the prevalence in childhood is lower than
in adolescence; Costello, Erkanli, & Angold, 2006; Kessler, Ave-
nevoli, & Merikangas, 2001), but as yet, few studies have explicitly
focused on the longitudinal relations between low self-esteem and
depression in children (Abela & Payne, 2003; Abela & Taylor, 2003;
8
The vulnerability effect of low self-esteem on depression, based on
seven samples (N ⫽ 2,112) with a mean age below 13 years, was ⫺.16
(p ⬍ .05).
Time lag (years)
14121086420
Self-esteem effect on depression
.00
-.20
-.40
-.60
Figure 4. Scatterplot displaying the relation between the cross-lagged
effect of self-esteem on depression (standardized regression weight) and
the time lag between the two assessments. The dashed line shows the
weighted mean effect size.
Table 5
Correlations and Standardized Regression Coefficients for
Sample Characteristics Predicting the Self-Esteem Effect on
Depression (k ⫽ 69)
Predictor r 
Proportion of female participants .00 .13
Mean age ⫺.11 ⫺.10
Time lag ⫺.25
ⴱ
⫺.19
Sample type
a
.09 .25
ⴱ
Note. Computations for the multiple regression analysis were made with
a random-effects model. Homogeneity Q
model
⫽ 10.25 (df ⫽ 4, p ⫽ .036);
homogeneity Q
residual
⫽ 84.55 (df ⫽ 64, p ⫽ .044); R
2
⫽ .11. k ⫽ number
of studies.
a
1 ⫽ representative, 0 ⫽ nonrepresentative.
ⴱ
p ⬍ .05.
228
SOWISLO AND ORTH
Borelli & Prinstein, 2006; Conley, Haines, Hilt, & Metalsky, 2001;
McCarty, Vander Stoep, & McCauley, 2007; Robinson, Garber, &
Hilsman, 1995). The finding that the vulnerability model holds for
children is important because children’s self-esteem is subject to
relatively strong developmental changes (Robins et al., 2002; Trzesni-
ewski, Donnellan, & Robins, 2003). Moreover, typical depressive
symptoms of children may differ from typical symptoms among
adolescents and adults (specifically, childhood depression can be
characterized more strongly by irritable than depressed mood; Amer-
ican Psychiatric Association, 2000). The finding that low self-esteem
shows a similar relation to depression in children as in adults is in line
with findings on other vulnerability factors for depression (Abela &
Hankin, 2008).
The moderator analyses also indicated that the vulnerability
effect of low self-esteem holds in different sample types. One
important finding is that the effect replicates in representative
samples (our data set included nine representative samples with
altogether more than 18,000 individuals), which significantly
strengthens the generalizability of the findings. Another important
finding is that the effect also replicates in clinical samples, sup-
porting the hypothesis that low self-esteem is a risk factor not only
for moderate but also for clinically relevant levels of depression
and, possibly, for depressive disorders. Although this conclusion
must be treated with caution because of the small number of data
points (i.e., three clinical samples including about 650 individu-
als), additional aspects support the conclusion. First, longitudinal
studies have demonstrated a relation between low self-esteem and
clinically diagnosed depression (Ormel, Oldehinkel, & Volle-
bergh, 2004; Trzesniewski et al., 2006). Second, in the general
population the prevalence of clinical depression is high (Kessler,
Berglund, et al., 2005), which consequently should be reflected in
the representative samples included in this meta-analysis. Third, as
mentioned in Footnote 1, the available evidence suggests that
depression is best conceptualized as a continuous rather than a
categorical construct: Representative samples, which cover the full
range of depression levels from absence of any depressive symp-
tom to severe levels of depression, should therefore provide for
valid insights into the structural relations between self-esteem and
depression.
We also tested for the moderating influence of the time lag
between assessments, in response to a previous call for studying its
influence on cross-lagged effects in a meta-analytic framework
(Collins & Graham, 2002). We found that the vulnerability effect
of low self-esteem on depression did not significantly vary as a
function of the time interval between assessments, which is some-
what unexpected in view of the findings of Cole and Maxwell
(2003). Cole and Maxwell’s analysis suggests that the effect size
should be zero when the time interval is zero (because any causal
effect needs a minimum amount of time to unfold); that the effect
size should increase when the time interval becomes larger, reach-
ing a maximum at a specific time interval; and that subsequently
the effect size should decrease again and approach zero (because
after long time intervals the causal effect will have disappeared).
Although our meta-analysis covered a large range of time lags
(from several days to several years), no linear or curvilinear trend
was detectable after controlling for other study characteristics. One
reason might be that the number of studies was too small for this
type of moderator analysis, restricting the statistical power. Nev-
Table 6
Analysis of Variance of Self-Esteem Effect on Depression by Sample Type (k ⫽ 77)
Sample type kN
Weighted mean
effect size
a
95% CI Homogeneity (Q)
Representative 9 18,199 ⫺.12
ⴱ
[⫺.17, ⫺.07] 6.47
Convenience 47 14,078 ⫺.17
ⴱ
[⫺.20, ⫺.14] 66.93
ⴱ
College students 18 2,573 ⫺.13
ⴱ
[⫺.18, ⫺.08] 10.92
Clinical 3 651 ⫺.15
ⴱ
[⫺.26, ⫺.03] 5.70
Note. Computations were made with a random-effects model. Homogeneity Q
between
⫽ 3.63 (df ⫽ 3, p ⫽
.304); homogeneity Q
within
⫽ 89.99 (df ⫽ 73, p ⫽ .086). k ⫽ number of studies; N ⫽ total number of participants
in the k samples; CI ⫽ confidence interval.
a
Standardized regression coefficient.
ⴱ
p ⬍ .05.
Table 7
Analysis of Variance of Self-Esteem Effect on Depression by Self-Esteem Measure (k ⫽ 77)
Self-esteem measure kN
Weighted mean
effect size
a
95% CI Homogeneity (Q)
Rosenberg Self-Esteem Scale 60 30,954 ⫺.15
ⴱ
[⫺.18, ⫺.13] 71.08
Harter Self-Perception Profile 11 2,721 ⫺.16
ⴱ
[⫺.21, ⫺.10] 6.60
Other 6 1,826 ⫺.18
ⴱ
[⫺.25, ⫺.10] 12.59
ⴱ
Note. Computations were made with a random-effects model. Homogeneity Q
between
⫽ 0.34 (df ⫽ 2, p ⫽
.843); homogeneity Q
within
⫽ 90.27 (df ⫽ 74, p ⫽ .096). k ⫽ number of studies; N ⫽ total number of participants
in the k samples; CI ⫽ confidence interval.
a
Standardized regression coefficient.
ⴱ
p ⬍ .05.
229
LOW SELF-ESTEEM, DEPRESSION, AND ANXIETY
ertheless, our analysis indicates that the vulnerability effect is
stable and detectable across a wide range of time intervals. This
finding has two implications. First, it indicates that self-esteem has
predictive power over a long period. Consequently, an important
avenue for future research is to further investigate which mediating
mechanisms account for the large temporal stability of this effect.
Second, the finding indicates that the vulnerability effect of low
self-esteem is already detectable after short time intervals. Conse-
quently, future research should examine which mediating mecha-
nisms account for the self-esteem effect across a few weeks or
even a few days (and whether the mediating processes across short
vs. long time intervals are identical). For example, low self-esteem
might elicit rumination on one day, which in turn exacerbates
depressive symptoms over the following days (Nolen-Hoeksema,
2000). Thus, in future research it would be intriguing to study
these phenomena in a higher temporal resolution, for example,
using diary data.
An important task of future research is to further examine
moderators of the vulnerability effect of low self-esteem and to
explain why some people with low self-esteem develop depression
while others do not. For example, previous research suggests that
the vulnerability effect might be stronger if a person’s self-esteem
is not only low but also temporally stable (Kernis, Grannemann, &
Mathis, 1991). Another example is research by Michalak, Teis-
mann, Heidenreich, Strohle, and Vocks (2011), suggesting that
mindful acceptance buffers the detrimental effect of low self-
esteem on depression. Moreover, situational factors could moder-
ate the vulnerability effect. One hypothesis is that low self-esteem
might have stronger effects on depression when the individual
simultaneously suffers from stressful life circumstances (e.g., J. E.
Roberts, 2006). However, in three independent studies, Orth, Rob-
ins, and Meier (2009) did not find evidence that the occurrence of
stressful life events or daily hassles influenced the prospective
effect of low self-esteem on depression. Nevertheless, it is possible
that other characteristics of the situation moderate the strength of
the vulnerability effect. For example, social support by relation-
ship partners, family, and friends might protect individuals with
low self-esteem from spiraling downward into depression.
The results on the stability coefficients provide an additional
argument in favor of the distinction between the concepts of
self-esteem and depression. More precisely, the present results
suggest that self-esteem is a more stable, trait-like construct than
depression, corresponding to findings reported in the literature
(Lovibond, 1998; Trzesniewski et al., 2003).
9
Given that the
average time lag between assessments was more than 1 year, the
stability coefficients for self-esteem are not much smaller than
the stability coefficients of broad personality constructs such as the
Big Five personality factors (Ferguson, 2010; B. W. Roberts &
DelVecchio, 2000). This result is in line with the findings by
Trzesniewski et al. (2003), who reported that the stability of
self-esteem is moderately high across the life span (disattenuated
correlations averaging in the .50s–.70s). Moreover, the present
results are consistent with the notion that, typically, the more
dispositional factor (i.e., self-esteem) influences the more fluctu-
ating, state-like factor (i.e., depression) rather than vice versa. If
low self-esteem and depression were two interchangeable indica-
tors of the same construct, then they should have comparable
stabilities, because their individual stabilities should each reflect
the stability of the common factor. However, we note that the
stability coefficient for depression was still relatively large, indi-
cating a moderate degree of stability.
Limitations
An important limitation of this research is that it does not allow
for strong conclusions regarding the causality of the relations
between self-esteem, depression, and anxiety, because all the stud-
ies included in the meta-analysis used correlational designs. There-
fore, the effects under investigation were not experimentally in-
duced but may be caused by third variables that were not
controlled for (Finkel, 1995; Little, Preacher, Selig, & Card, 2007).
For example, neuroticism is related to low self-esteem (Judge et
al., 2002; Robins et al., 2001) and depression (Kendler, Neale,
Kessler, Heath, & Eaves, 1993; Ormel, Oldehinkel, & Brilman,
2001), and therefore might be a third variable influencing both
constructs. Another example might be common genetic factors of
low self-esteem and depression (Neiss, Stevenson, Legrand,
Iacono, & Sedikides, 2009; S. B. Roberts & Kendler, 1999). Future
9
No formal significance test for the difference between the stability
coefficients of self-esteem and depression is available, for the reasons
given in Footnote 6. We therefore used the confidence intervals as an
approximate means of comparing the coefficients. Moreover, for compar-
ison purposes, we also computed an unpaired t test, which was significant,
with t(117) ⫽ 5.10, p ⬍ .001, suggesting that the coefficients differed
significantly.
Table 8
Analysis of Variance of Self-Esteem Effect on Depression by Depression Measure (k ⫽ 77)
Depression measure KN
Weighted mean
effect size
a
95% CI Homogeneity (Q)
CES-D 30 24,872 ⫺.14
ⴱ
[⫺.18, ⫺.11] 39.86
BDI 20 4,390 ⫺.16
ⴱ
[⫺.21, ⫺.12] 18.28
CDI 8 2,487 ⫺.13
ⴱ
[⫺.19, ⫺.07] 5.03
Other 19 3,752 ⫺.18
ⴱ
[⫺.23, ⫺.14] 28.87
Note. Computations were made with a random-effects model. Homogeneity Q
between
⫽ 2.66 (df ⫽ 3, p ⫽
.447); homogeneity Q
within
⫽ 92.04 (df ⫽ 73, p ⫽ .065). k ⫽ number of studies; N ⫽ total number of participants
in the k samples; CI ⫽ confidence interval; CES-D ⫽ Center for Epidemiologic Studies Depression Scale;
BDI ⫽ Beck Depression Inventory; CDI ⫽ Children’s Depression Inventory.
a
Standardized regression coefficient.
ⴱ
p ⬍ .05.
230
SOWISLO AND ORTH
research should test relevant third-variable models that might
account for the relations between low self-esteem and depression,
and low self-esteem and anxiety. Nevertheless, when experimental
designs are not feasible for ethical or practical reasons, longitudi-
nal analyses are useful because they can indicate whether the data
are consistent with a causal model of the relation between the
variables, by establishing the direction of the effects and ruling out
some (but not all) alternative causal hypotheses.
Another limitation is that nearly all studies included in the
meta-analysis employed self-report measures of the constructs.
Although the vast majority of the measures used are reliable and
well validated, a problem of the exclusive reliance on self-report
methodology is that correlations between measures may be artifi-
cially inflated by shared method variance. Note, however, that
shared method variance cannot account for the prospective cross-
lagged effects because shared method variance has already been
statistically removed by controlling for prior levels of the predicted
construct. Nevertheless, future research would benefit from includ-
ing measures based on informant reports (e.g., ratings by relation-
ship partners) and diagnostic interviews to further control for
possible self-report biases.
Furthermore, the studies included in the meta-analysis were
predominantly conducted in Western cultural contexts (i.e., only
two studies were conducted in Asia). Therefore, future research
should test whether the results hold in other cultural contexts, such
as in Asian or African cultures (Arnett, 2008; Henrich, Heine, &
Norenzayan, 2010). The function of self-esteem and the frequency
or intensity of depressive and anxious symptoms may vary cross-
culturally. For example, individuals from Asian and Western cul-
tures show different self-construal styles and different tendencies
toward self-enhancement (Heine et al., 1999; Markus & Kitayama,
1991). As another example, research suggests that there are cul-
tural differences in the reporting of depressive symptoms (Parker,
Gladstone, & Chee, 2001; Ryder et al., 2008). These cross-cultural
differences might have consequences for the relation of low self-
esteem with depression and anxiety. Therefore, whether studies
with samples from other cultural contexts would yield the same
results as the present meta-analysis is currently unknown.
An additional limitation is that our data did not allow us to
investigate several, more nuanced characteristics of the relations
between self-esteem, depression, and anxiety. First, it would be
interesting to test for other models that could explain the relations
between the constructs. For example, in addition to being a vul-
nerability factor, self-esteem might influence the course or treat-
ment of depressive disorders (e.g., G. W. Brown, Bifulco, &
Andrews, 1990; Ezquiaga et al., 2004), corresponding to the
pathoplasty model (Clark, 2005; Klein, Kotov, & Bufferd, 2011;
Santor, Bagby, & Joffe, 1997). The present meta-analysis did not
allow examining the pathoplasty model because very few studies
reported information on diagnoses and treatment of depressive
disorders (information that would be needed to assess whether
self-esteem predicts the course of depressive disorders). Another
example is the common cause model (Klein et al., 2011), which
states that low self-esteem and depression have a shared etiology
accounting for the observed association and that corresponds to the
third-variable models discussed above. The present meta-analysis
did not allow testing the common cause model because very few of
the primary studies examined third variables that could serve as a
common cause (as mentioned above, it would, for example, be
interesting to test whether neuroticism is a common cause of low
self-esteem and depression). Second, because clinical anxiety is a
heterogeneous category (Heimberg et al., 1989; Mor & Winquist,
2002), research should clarify how low self-esteem relates to
different forms of anxiety (e.g., social anxiety, worry, panic, and
phobias). Recent studies with child, adolescent, and adult samples
have found that some forms of anxiety (e.g., generalized anxiety)
load together with depression and dysthymia on one factor, rather
than on another factor together with the remaining forms of anx-
iety (Krueger, 1999; Lahey et al., 2004). Accordingly, low self-
esteem might be a stronger vulnerability factor for certain forms of
anxiety such as generalized anxiety (i.e., having a prospective
effect of similar size as for depression). Similarly, it might be
interesting to further investigate how low self-esteem relates do
different forms of depression (e.g., depressive episodes with atyp-
ical or melancholic features; American Psychiatric Association,
2000).
Another limitation of the meta-analytic approach is that we
could not control for potential content overlap between the con-
structs. Although self-esteem and anxiety measures typically do
not overlap in their item content, depression measures frequently
include one or two items that are conceptually related to self-
esteem. However, the fact that the vulnerability effect of low
self-esteem on depression replicated across different combinations
of self-esteem and depression measures (which may differ in their
degree of content overlap) suggests that the effect is not biased by
potential content overlap. Moreover, the findings from four longi-
tudinal studies (Orth et al., 2008; Orth, Robins, Trzesniewski, et
al., 2009) that controlled for content overlap between self-esteem
and depression scales suggest that the vulnerability effect of low
self-esteem is not influenced by depression items that tap into the
self-esteem construct. A related issue is that depression and anx-
iety measures frequently overlap in their item content, which in
turn might affect the relative strength of their individual relations
to self-esteem. To further address this issue, future research should
employ designs in which self-esteem, depression, and anxiety are
simultaneously examined and potential content overlap is con-
trolled for. Moreover, this approach would afford the additional
benefit of enabling tests of prospective effects between all three
variables.
Furthermore, it is possible that the relations between self-
esteem, depression, and anxiety are influenced by narcissism,
which is conceptually related to high self-esteem (Morf & Rho-
dewalt, 2001; Tracy, Cheng, Robins, & Trzesniewski, 2009).
Although measures of self-esteem and narcissism are only mod-
erately correlated (Ackerman et al., 2011; R. P. Brown & Zeigler-
Hill, 2004), it is possible that the prospective effects of low
self-esteem on depression and anxiety are even stronger when
narcissism is statistically controlled for.
Finally, it is possible that self-esteem, depression, and anxiety
have been subject to generational changes in the past decades, and
consequently an important question is whether these possible
secular trends in the mean levels of the constructs can be recon-
ciled with the findings of our meta-analysis. First, we note that the
evidence regarding generational changes in self-esteem, depres-
sion, and anxiety is inconsistent. For example, whereas some
studies suggest that there are generational increases in self-esteem
(Gentile, Twenge, & Campbell, 2010; Twenge & Campbell, 2001),
the results of other studies—two of which used longitudinal data
231
LOW SELF-ESTEEM, DEPRESSION, AND ANXIETY
from national probability samples—suggest that the average level
of self-esteem has not changed across the generations born in the
20th century (Erol & Orth, 2011; Orth et al., 2010, 2012). Simi-
larly, the evidence regarding generational increases in constructs
related to self-esteem, such as self-enhancement and narcissism, is
inconsistent. For example, whereas some studies find supporting
evidence (Twenge & Foster, 2008; Twenge, Konrath, Foster,
Campbell, & Bushman, 2008), the results of other studies suggest
that there are no generational changes (Trzesniewski & Donnellan,
2010; Trzesniewski, Donnellan, & Robins, 2008). Also, with re-
gard to depression and anxiety, some studies report significant
generational increases (e.g., Twenge, 2000), whereas other studies
did not find supporting evidence (Booth, Sharma, & Leader, 2011;
Orth et al., 2012; Simon & VonKorff, 1992). In sum, the available
evidence on generational changes in the constructs examined in
this research is inconsistent and a topic of current debate in the
literature.
Second, even if generational changes in the constructs were
present, they do not necessarily contradict the findings of the
present study. The reason is that mean levels of variables and the
structural relations between these variables can vary independently
from one another. For example, it is possible that the mean levels
of self-esteem, depression, and anxiety change over time, whereas
the structural relations between the constructs remain unaltered. In
accordance with this reasoning, although previous research has
documented significant age differences in the level of self-esteem
(Meier, Orth, Denissen, & Ku¨hnel, 2011; Orth et al., 2010, 2012)
and depression (Kessler et al., 1992; Mirowsky & Kim, 2007)
across the life course, this meta-analysis and a previous study
(Orth, Robins, Trzesniewski, et al., 2009) found that age did not
significantly moderate the prospective relations between self-
esteem and depression. We therefore believe that the validity of
our meta-analytic findings is not called into question by possible
generational changes in the constructs.
Conclusions
The present research suggests that self-esteem shows diverging
structural relations with depression and anxiety. As yet, drawing
clinical recommendations from this affective specificity would be
premature. Nevertheless, continuing this line of research might
ultimately lead to the identification of mechanisms specific to
depression and anxiety, which in turn might provide important
information for the further development of disorder-specific treat-
ment approaches.
Moreover, the present research shows that the effect of low
self-esteem on depression is robust and holds across different
sample and design characteristics of studies. The robustness of the
effect has important implications for research, suggesting that the
conclusions of extant studies in this field are probably generaliz-
able and that future studies can build on this effect and investigate
it in more detail. Furthermore, when studying vulnerability factors
for depression, researchers should control for low self-esteem in
order not to overestimate the effects of other vulnerability factors.
The robustness of the effect also strengthens the potential impor-
tance of self-esteem interventions. If future research supports the
hypothesized causality of the relations between the constructs,
interventions aimed at increasing self-esteem might be useful in
reducing the risk of depression, regardless of the gender and age of
the individuals, and might not only reduce the short-term risk of
depression but have a long-lasting, positive influence.
References
References marked with an asterisk indicate studies included in the
meta-analysis.
Abela, J. R. Z., & Hankin, B. L. (2008). Cognitive vulnerability to
depression in children and adolescents: A developmental psychopathol-
ogy perspective. In J. R. Z. Abela & B. L. Hankin (Eds.), Handbook of
depression in children and adolescents (pp. 35–78). New York, NY:
Guilford Press.
*Abela, J. R. Z., & Payne, A. V. (2003). A test of the integration of the
hopelessness and self-esteem theories of depression in schoolchildren.
Cognitive Therapy and Research, 27, 519–535. doi:10.1023/A:
1026303020478
Abela, J. R. Z., & Taylor, G. (2003). Specific vulnerability to depressive
mood reactions in schoolchildren: The moderating role of self-esteem.
Journal of Clinical Child and Adolescent Psychology, 32, 408 –418.
doi:10.1207/S15374424JCCP3203_09
Abramson, L. Y., Seligman, M. E. P., & Teasdale, J. D. (1978). Learned
helplessness in humans: Critique and reformulation. Journal of Abnor-
mal Psychology, 87, 49–74. doi:10.1037/0021-843X.87.1.49
Ackerman, R. A., Witt, E. A., Donnellan, M. B., Trzesniewski, K. H.,
Robins, R. W., & Kashy, D. A. (2011). What does the Narcissistic
Personality Inventory really measure? Assessment, 18, 67– 87. doi:
10.1177/1073191110382845
Adler, D. A., McLaughlin, T. J., Rogers, W. H., Chang, H., Lapitsky, L.,
& Lerner, D. (2006). Job performance deficits due to depression. Amer-
ican Journal of Psychiatry, 163, 1569 –1576. doi:10.1176/appi.ajp
.163.9.1569
American Psychiatric Association. (2000). Diagnostic and statistical man-
ual of mental disorders (4th ed., text rev.). Washington, DC: Author.
Arnett, J. J. (2008). The neglected 95%: Why American psychology needs
to become less American. American Psychologist, 63, 602– 614. doi:
10.1037/0003-066X.63.7.602
Aspinwall, L. G., & Taylor, S. E. (1992). Modeling cognitive adaptation:
A longitudinal investigation of the impact of individual differences and
coping on college adjustment and performance. Journal of Personality
and Social Psychology, 63, 989 –1003. doi:10.1037/0022-3514.63.6.989
Barnes, L. L., Harp, D., & Jung, W. S. (2002). Reliability generalization of
scores on the Spielberger State–Trait Anxiety Inventory. Educational
and Psychological Measurement, 62, 603–618. doi:10.1177/
0013164402062004005
Baumeister, R. F. (1998). The self. In D. T. Gilbert, S. T. Fiske, & G.
Lindzey (Eds.), The handbook of social psychology (Vol. 1, pp. 680 –
740). Boston, MA: McGraw-Hill.
Baumeister, R. F. (Ed.). (1993). Self-esteem: The puzzle of low self-regard.
New York, NY: Plenum Press.
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. doi:10.1111/1529-1006.01431
Baumeister, R. F., & Leary, M. R. (1995). The need to belong: Desire for
interpersonal attachment as a fundamental human motivation. Psycho-
logical Bulletin, 117, 497–529. doi:10.1037/0033-2909.117.3.497
Baumeister, R. F., Smart, L., & Boden, J. M. (1996). Relation of threatened
egotism to violence and aggression: The dark side of high self-esteem.
Psychological Review, 103, 5–33. doi:10.1037/0033-295X.103.1.5
Beck, A. T. (1967). Depression: Clinical, experimental, and theoretical
aspects. New York, NY: Harper & Row.
Beck, A. T. (1987). Cognitive models of depression. Journal of Cognitive
Psychotherapy, 1, 5–37.
Beck, A. T., Epstein, N., Brown, G., & Steer, R. A. (1988). An inventory
232
SOWISLO AND ORTH
for measuring clinical anxiety: Psychometric properties. Journal of
Consulting and Clinical Psychology, 56, 893– 897. doi:10.1037/0022-
006X.56.6.893
Beck, A. T., Steer, R. A., & Carbin, M. G. (1988). Psychometric properties
of the Beck Depression Inventory: Twenty-five years of evaluation.
Clinical Psychology Review, 8, 77–100. doi:10.1016/0272-
7358(88)90050-5
Beck, A. T., Steer, R. A., & Epstein, N. (1992). Self-concept dimensions
of clinically depressed and anxious outpatients. Journal of Clinical
Psychology, 48, 423– 432. doi:10.1002/1097-4679(199207)48:4⬍423::
AID-JCLP2270480402⬎3.0.CO;2-G
Beck, A. T., Steer, R. A., Epstein, N., & Brown, G. (1990). Beck Self-
Concept Test. Psychological Assessment, 2, 191–197. doi:10.1037/
1040-3590.2.2.191
Beck, A. T., Ward, C. H., Mendelson, M., Mock, J., & Erbaugh, J. (1961).
An inventory for measuring depression. Archives of General Psychiatry,
4, 561–571. doi:10.1001/archpsyc.1961.01710120031004
Beck, R., & Perkins, T. S. (2001). Cognitive content-specificity for anxiety
and depression: A meta-analysis. Cognitive Therapy and Research, 25,
651– 663. doi:10.1023/A:1012911104891
Beck, R., Perkins, T. S., Holder, R., Robbins, M., Gray, M., & Allison,
S. H. (2001). The cognitive and emotional phenomenology of depression
and anxiety: Are worry and hopelessness the cognitive correlates of NA
and PA? Cognitive Therapy and Research, 25, 829 –838. doi:10.1023/
A:1012983726272
Berman, A. L. (2009). Depression and suicide. In I. H. Gotlib & C. L.
Hammen (Eds.), Handbook of depression (2nd ed., pp. 510 –530). New
York, NY: Guilford Press.
Blascovich, J., & Tomaka, J. (1991). Measures of self-esteem. In J. P.
Robinson, P. R. Shaver, & L. S. Wrightsman (Eds.), Measures of
personality and social psychological attitudes (pp. 115–160). San Di-
ego, CA: Academic Press.
Blatt, S. J., D’Afflitti, J. P., & Quinlan, D. M. (1976). Experiences of
depression in normal young adults. Journal of Abnormal Psychology, 85,
383–389. doi:10.1037/0021-843X.85.4.383
Boden, J. M., Fergusson, D. M., & Horwood, L. J. (2008). Does adolescent
self-esteem predict later life outcomes? A test of the causal role of
self-esteem. Development and Psychopathology, 20, 319 –339. doi:
10.1017/S0954579408000151
*Bohon, C., Stice, E., Burton, E., Fudell, M., & Nolen-Hoeksema, S.
(2008). A prospective test of cognitive vulnerability models of depres-
sion with adolescent girls. Behavior Therapy, 39, 79 –90. doi:10.1016/
j.beth.2007.05.003
Booth, R., Sharma, D., & Leader, T. (2011, July). Is trait anxiety increas-
ing? STAI scores across 40 years and 40 nations. Paper presented at the
12th European Congress of Psychology, Istanbul, Turkey.
*Borelli, J. L., & Prinstein, M. J. (2006). Reciprocal, longitudinal associ-
ations among adolescents’ negative feedback-seeking, depressive symp-
toms, and peer relations. Journal of Abnormal Child Psychology, 34,
154 –164. doi:10.1007/s10802-005-9010-y
Bosson, J. K., Swann, W. B., & Pennebaker, J. W. (2000). Stalking the
perfect measure of implicit self-esteem: The blind men and the elephant
revisited? Journal of Personality and Social Psychology, 79, 631– 643.
doi:10.1037/0022-3514.79.4.631
Brown, G. W., Bifulco, A., & Andrews, B. (1990). Self-esteem and
depression. IV. Effect on course and recovery. Social Psychiatry and
Psychiatric Epidemiology, 25, 244–249. doi:10.1007/BF00788645
Brown, G. W., & Harris, T. (1978). Social origins of depression: A study
of psychiatric disorder. New York, NY: Free Press.
Brown, R. P., & Zeigler-Hill, V. (2004). Narcissism and the nonequiva-
lence of self-esteem measures: A matter of dominance? Journal of
Research in Personality, 38, 585–592. doi:10.1016/j.jrp.2003.11.002
Brown, T. A., Campbell, L. A., Lehman, C. L., Grisham, J. R., & Mancill,
R. B. (2001). Current and lifetime comorbidity of the DSM–IV anxiety
and mood disorders in a large clinical sample. Journal of Abnormal
Psychology, 110, 585–599. doi:10.1037/0021-843X.110.4.585
Buchwald, A. M., & Rudick-Davis, D. (1993). The symptoms of major
depression. Journal of Abnormal Psychology, 102, 197–205. doi:
10.1037/0021-843X.102.2.197
Buhrmester, M. D., Blanton, H., & Swann, W. B. (2011). Implicit self-
esteem: Nature, measurement, and a new way forward. Journal of
Personality and Social Psychology, 100, 365–385. doi:10.1037/
a0021341
*Burwell, R. A., & Shirk, S. R. (2006). Self processes in adolescent
depression: The role of self-worth contingencies. Journal of Research on
Adolescence, 16, 479– 490. doi:10.1111/j.1532-7795.2006.00503.x
*Butler, A. C., Hokanson, J. E., & Flynn, H. A. (1994). A comparison of
self-esteem lability and low trait self-esteem as vulnerability factors for
depression. Journal of Personality and Social Psychology, 66, 166 –177.
doi:10.1037/0022-3514.66.1.166
Byrne, B. M. (1996). Measuring self-concept across the life span: Issues
and instrumentation. Washington, DC: American Psychological Asso-
ciation. doi:10.1037/10197-000
*Cambron, M. J., Acitelli, L. K., & Steinberg, L. (2010). When friends
make you blue: The role of friendship contingent self-esteem in predict-
ing self-esteem and depressive symptoms. Personality and Social Psy-
chology Bulletin, 36, 384–397. doi:10.1177/0146167209351593
*Cast, A. D., & Burke, P. J. (2002). A theory of self-esteem. Social Forces,
80, 1041–1068. doi:10.1353/sof.2002.0003
*Chen, W.-L. (1995). A longitudinal investigation of the relationships
among parental life stress, parental self-esteem, marital satisfaction,
and parental depression (Unpublished doctoral dissertation). Iowa State
University, Ames.
Cheng, H., & Furnham, A. (2004). Perceived parental rearing style, self-
esteem and self-criticism as predictors of happiness. Journal of Happi-
ness Studies, 5, 1–21. doi:10.1023/B:JOHS.0000021704.35267.05
*Cheng, S. K., & Lam, D. J. (1997). Relationships among life stress,
problem solving, self-esteem, and dysphoria in Hong Kong adolescents:
Test of a model. Journal of Social and Clinical Psychology, 16, 343–
355. doi:10.1521/jscp.1997.16.3.343
*Cikara, M., & Girgus, J. S. (2010). Unpacking social hypersensitivity:
Vulnerability to the absence of positive feedback. Personality and Social
Psychology Bulletin, 36, 1409–1423. doi:10.1177/0146167210383288
Clark, L. A. (2005). Temperament as a unifying basis for personality and
psychopathology. Journal of Abnormal Psychology, 114, 505–521. doi:
10.1037/0021-843X.114.4.505
Clark, L. A., Watson, D., & Mineka, S. (1994). Temperament, personality,
and the mood and anxiety disorders. Journal of Abnormal Psychology,
103, 103–116. doi:10.1037/0021-843X.103.1.103
Clogg, C. C., Petkova, E., & Haritou, A. (1995). Statistical methods for
comparing regression coefficients between models. American Journal of
Sociology, 100, 1261–1293. doi:10.1086/230638
Cohen, J., Cohen, P., West, S. G., & Aiken, L. S. (2003). Applied multiple
regression/correlation analysis for the behavioral sciences (3rd ed.).
Mahwah, NJ: Erlbaum.
*Colarossi, L. G., & Eccles, J. S. (2003). Differential effects of support
providers on adolescents’ mental health. Social Work Research, 27,
19 –30. doi:10.1093/swr/27.1.19
Cole, D. A., & Maxwell, S. E. (2003). Testing mediational models with
longitudinal data: Questions and tips in the use of structural equation
modeling. Journal of Abnormal Psychology, 112, 558 –577. doi:
10.1037/0021-843X.112.4.558
Cole, D. A., Nolen-Hoeksema, S., Girgus, J., & Paul, G. (2006). Stress
exposure and stress generation in child and adolescent depression: A
latent trait-state-error approach to longitudinal analyses. Journal of
Abnormal Psychology, 115, 40–51. doi:10.1037/0021-843X.115.1.40
Collins, L. M., & Graham, J. W. (2002). The effect of the timing and
spacing of observations in longitudinal studies of tobacco and other drug
233
LOW SELF-ESTEEM, DEPRESSION, AND ANXIETY