Discussion 14 - Glacier

Research Article

Low Self-Esteem Is Related to Aggression, Antisocial Behavior, and Delinquency M. Brent Donnellan,1 Kali H. Trzesniewski,2,3 Richard W. Robins,4 Terrie E. Moffitt,2,3 and

Avshalom Caspi2,3

1Michigan State University; 2Institute of Psychiatry, King’s College London, London, United Kingdom; 3University

of Wisconsin-Madison; and 4University of California, Davis

ABSTRACT—The present research explored the controver-

sial link between global self-esteem and externalizing

problems such as aggression, antisocial behavior, and

delinquency. In three studies, we found a robust relation

between low self-esteem and externalizing problems. This

relation held for measures of self-esteem and externalizing

problems based on self-report, teachers’ ratings, and

parents’ ratings, and for participants from different na-

tionalities (United States and New Zealand) and age

groups (adolescents and college students). Moreover, this

relation held both cross-sectionally and longitudinally and

after controlling for potential confounding variables such

as supportive parenting, parent-child and peer relation-

ships, achievement-test scores, socioeconomic status, and

IQ. In addition, the effect of self-esteem on aggression was

independent of narcissism, an important finding given

recent claims that individuals who are narcissistic, not low

in self-esteem, are aggressive. Discussion focuses on clar-

ifying the relations among self-esteem, narcissism, and

externalizing problems.

The link between global self-esteem and aggression is currently

being debated by researchers (Baumeister, Campbell, Krueger,

& Vohs, 2003; DuBois & Tevendale, 1999) and in the popular

media (e.g., Slater, 2002). Researchers on one side of the

debate have argued that individuals with low self-esteem are

prone to real-world externalizing problems such as delinquency

and antisocial behavior (e.g., Fergusson & Horwood, 2002;

Rosenberg, Schooler, & Schoenbach, 1989; Sprott & Doob,

2000). However, others have questioned this claim, noting that

several studies have failed to find a relation between low self-

esteem and externalizing problems (e.g., Bynner, O’Malley, &

Bachman, 1981; Jang & Thornberry, 1998; McCarthy & Hoge,

1984) or between low global self-esteem and laboratory mea-

sures of aggression (Bushman & Baumeister, 1998; Kirkpatrick,

Waugh, Valencia, &Webster, 2002; Twenge & Campbell, 2003).

On the basis of this research, Baumeister, Bushman, and

Campbell (2000) suggested that ‘‘future research can benefit

from discarding the obsolete view that low self-esteem causes

violence’’ (p. 29). Instead, Baumeister and his colleagues have

posited that any link between self-esteem and aggression

probably occurs at the high end of the self-esteem continuum;

that is, unrealistically high self-esteem (best captured by

measures of narcissism), not low self-esteem, contributes to

aggression and crime (e.g., Baumeister, Smart, & Boden, 1996).

At least three distinct traditions in the social sciences posit a

link between low self-esteem and externalizing problems.

Rosenberg (1965) suggested that low self-esteem weakens ties

to society; according to social-bonding theory, weaker ties to

society decrease conformity to social norms and increase de-

linquency (Hirschi, 1969). Humanistic psychologists such as

Rogers (e.g., 1961) have argued that a lack of unconditional

positive self-regard is linked to psychological problems, in-

cluding aggression. Finally, neo-Freudians also posit that low

self-regard motivates aggression. For example, Horney (1950)

and Adler (1956) theorized that aggression and antisocial be-

havior are motivated by feelings of inferiority rooted in early

childhood experiences of rejection and humiliation. More spe-

cifically, Tracy and Robins (2003) suggested that individuals

protect themselves against feelings of inferiority and shame by

externalizing blame for their failures, which leads to feelings of

hostility and anger toward other people. Thus, three separate

Address correspondence to M. Brent Donnellan, Department of Psychology, Michigan State University, East Lansing, MI 48823; e-mail: [email protected].

PSYCHOLOGICAL SCIENCE

328 Volume 16—Number 4Copyright r 2005 American Psychological Society

theoretical perspectives posit that externalizing behaviors are

motivated, in part, by low self-esteem.

Despite these theoretical arguments, research on the link

between low self-esteem and externalizing problems has failed

to produce consistent results. An understanding of the precise

nature of this relation has important theoretical implications, as

well as practical implications given the media attention sur-

rounding the issue. To bring new data to bear on this contro-

versy, we report results from three studies that extend previous

research in several ways. First, we used a multimethod ap-

proach to assessing self-esteem and externalizing problems.

Previous research has relied almost exclusively on self-report

measures, so it is possible that the relations that have been

observed are due to shared method variance. Second, we ex-

amined several theoretically relevant variables that might ac-

count for the effects of self-esteem on externalizing problems,

including IQ, academic achievement, socioeconomic status

(SES), and the quality of parent-child relationships. Third, we

used longitudinal data to test the hypothesis that low self-es-

teem is related to future externalizing problems (Study 2). Fi-

nally, we assessed narcissism to examine the possibility that

unrealistically high self-esteem is related to aggression and to

determine whether self-esteem and narcissism have inde-

pendent effects (Study 3).

STUDY 1

Study 1 investigated the relation between self-reports and

teacher ratings of self-esteem and self-reports of delinquency in

a sample of 11- and 14-year-olds. We also controlled for two

theoretically relevant variables—supportive parenting and ac-

ademic achievement—that might account for the effects of self-

esteem on delinquency.

Method

Participants

The sample included 292 (78% response rate) 11- and 14-year-

old participants (mean age 5 12.66 years, SD 5 1.57; 55%

female; 56.5% European American, 4.8% Asian American or

Pacific Islander, 20.5% Hispanic American, 9.2% African

American, and 9.0% ‘‘other’’ or not reported) from two schools

in northern California.

Measures

Self-esteem was measured with the 10-item Rosenberg (1965)

Self-Esteem Scale (RSE; a 5 .81) and the 6-item Global sub-

scale of the Harter (1985) Self-Perception Profile for Children

(SPPC; a 5 .75). Teachers completed a modified teacher ver-

sion of the SPPC (a 5 .88).

Delinquency was measured using a 12-item delinquent-be-

haviors scale adapted from Elliott, Huizinga, and Ageton (1985;

a 5 .85).

Supportive parenting (warmth, monitoring, use of inductive

reasoning, and consistent discipline) was measured using a

modified scale from the Iowa Youth and Families Project (e.g.,

Conger et al., 1992; a 5 .89).

Academic achievement was measured by a composite of the

Math and Reading percentile scores from the Stanford Achieve-

ment Test Battery.

Results and Discussion

Self-esteem was consistently negatively correlated with delin-

quency, regardless of whether self-esteem was assessed by the

RSE (r 5�.35), the self-report version of the SPPC (r 5�.39),

or the teacher version of the SPPC (r 5�.29; all ps < .05).1 To

explore these effects further, we compared the self-esteem

scores of individuals who reported at least one delinquent act

(76% of the sample) and those who reported no delinquent acts.

The delinquent group had lower self-esteem than the nonde-

linquent group on all three self-esteem measures (Cohen’s d 5

0.48, 0.63, and 0.35 for the RSE, self-report SPPC, and teacher

SPPC, respectively; all ps < .05).

Baumeister et al. (1996) focused their critique of the low-self-

esteem hypothesis on aggression, and it was possible that our

results were due to delinquent behaviors not involving aggres-

sion. To address this issue, we divided the delinquency scale

into a 2-item aggression scale (‘‘got into a fight,’’ ‘‘beat someone

up’’) and a 10-item nonaggression scale (e.g., ‘‘lied to parents or

teachers,’’ ‘‘used drugs or alcohol’’). All the effects of self-es-

teem remained significant for both the aggression scale (rs

ranged from �.17 to �.26, ps < .05) and the nonaggression

scale (rs ranged from �.28 to �.39, ps < .05).

To test whether supportive parenting and academic

achievement could account for the relation between low self-

esteem and delinquency, we used structural equation modeling

with latent variables defined by item parcels rather than indi-

vidual items (Kishton &Widaman, 1994). Supportive parenting

was defined by three parcels of eight items; self-esteem was

defined by the RSE and the self- and teacher-based SPPC

scales; delinquency was defined by three parcels of four items;

and academic achievement was modeled as a manifest variable.

An initial base model that included only self-esteem and de-

linquency had good fit, w2(8)5 9.09, n.s. (comparative fit index,

CFI 5 1.00; root mean square error of approximation, RMSEA

5 .02, p close fit 5 .76), and the path linking self-esteem to

delinquency was negative (b5�.52, p < .05). Figure 1 shows a

model controlling for both supportive parenting and academic

achievement. This model also had acceptable fit, w2(30) 5

60.07, p < .05 (CFI 5 .996; RMSEA 5 .06, p close fit 5 .24),

and the relation between self-esteem and delinquency remained

1Age and gender did not moderate the relation between self-esteem and de- linquency.

Volume 16—Number 4 329

M.B. Donnellan et al.

significant (b5�.28, p < .05). Thus, supportive parenting and

academic achievement could not explain the relation between

self-esteem and delinquency.

STUDY 2

The results of Study 1 provided support for the low-self-esteem

hypothesis. In Study 2, we extended Study 1 in several ways.

First, we used a longitudinal design to examine the prospective

relation between self-esteem and externalizing problems. Sec-

ond, Study 2 included non-self-report measures of externalizing

problems, specifically, teacher- and parent-rated antisocial

behavior. Third, Study 2 examined additional control variables,

including the quality of parent-child and peer relationships,

SES, and IQ. Finally, Study 2 was based on data from a repre-

sentative birth cohort of New Zealanders, so the range of ex-

ternalizing problems in the sample reflects the variation found

in the general population.

Method

Sample

Participants were members of the Dunedin Multidisciplinary

Health and Development Study (for details, see Moffitt, Caspi,

Rutter, & Silva, 2001), a longitudinal investigation of a com-

plete cohort of consecutive births between April 1, 1972, and

March 31, 1973, in Dunedin, New Zealand. The present study

included participants who completed a measure of self-esteem

at age 11 (n 5 812; 48% female; 78% of the initial cohort) or

age 13 (n 5 736; 48% female; 71% of the initial cohort).2

Measures

Self-esteem was measured at age 11 and age 13 with the RSE.

The use of a yes/no response format resulted in reliabilities that

were somewhat lower than usual for the RSE (a5 .64 at age 11

and .60 at age 13).

Externalizing problems were assessed using the Rutter Child

Scale (RCS; Rutter, Tizard, &Whitmore, 1970) and the Revised

Behavior Problem Checklist (RBPC; Quay & Peterson, 1987).

Teachers completed the RCS when the participants were ages

11 and 13; parents completed the RCS when the participants

were age 11 and the RBPC when they were age 13. Information

about the reliability and validity of these measures is provided

by Moffitt et al. (2001).

Relationship with parents and peers was assessed at age 13

using the Inventory of Parent Attachment (a 5 .77) and the

Inventory of Peer Attachment (Armsden & Greenberg, 1987; a 5 .80). These scales measure the degree to which adolescents

Fig. 1. Model linking global self-esteem and delinquency in Study 1 (N 5 292). Standardized coefficients are reported, and asterisks indicate structural coefficients significant at the .05 level. Self-esteem was measured with the Rosenberg (1965) Self-Esteem Scale, the Global subscale from Harter’s (1985) Self-Perception Profile for Children (SPPC), and a modified teacher report version of the Global subscale from the SPPC. The four-item parcels of delinquency items are labeled D1 through D3. Supportive parenting is a latent construct indexed by three eight-item parcels.

2Missing data do not appear to be a major cause for concern given the few differences between participants with and without self-esteem scores: Partici- pants without self-esteem scores at age 11 were rated as more antisocial by their parents than participants with self-esteem scores (d 5 0.23, p < .05), and participants without self-esteem scores at age 13 were rated as more antisocial by their parents and teachers than participants with self-esteem scores (ds 5 0.17 and 0.20, respectively, ps < .05).

330 Volume 16—Number 4

Self-Esteem and Externalizing Problems

feel they can trust, communicate with, and are not alienated

from their parents or peers.

IQ was assessed using the mean of each participant’s scores

on the Wechsler Intelligence Scale for Children–Revised

(Wechsler, 1974) at ages 7, 9, 11, and 13.

SES was calculated as the average social class of each par-

ticipant’s family from birth to age 15. Scores at each assessment

ranged from 1 (parents are unskilled laborers) to 6 (parents are

professionals).

Results and Discussion

Relation Between Self-Esteem and Externalizing Problems

Results were consistent with those of Study 1: Self-esteem was

negatively correlated with parent reports of externalizing prob-

lems (r 5�.18 at age 11 and r 5�.27 at age 13, ps < .05) and

with teacher reports of externalizing problems (r5�.16 at age 11

and r5�.18 at age 13, ps< .05). Moreover, self-esteem at age 11

was prospectively related to both parent and teacher reports of

externalizing problems at age 13 (both rs5�.20, ps< .05). As in

Study 1, the cross-method effects were significant; individuals

with low self-esteem were more likely to engage in antisocial

behaviors as reported by their parents and teachers. We divided

the items on the antisocial-behavior scales according to whether

they involved aggressive (e.g., fighting, bullying) or nonaggres-

sive (e.g., lying, disobedient) behaviors, and the effects of self-

esteem remained significant for both the aggression items (rs

ranged from�.13 to�.26, ps< .05) and the nonaggression items

(rs ranged from �.18 to �.21, ps < .05).

We next tested whether theoretically relevant third variables

could account for the relation between low self-esteem and

delinquency. The base model of self-esteem and externalizing

problems had good fit, w2(1)5 3.61, n.s. (CFI5 .99; RMSEA5

.06, p close fit 5 .32), and the path linking self-esteem to ex-

ternalizing problems was negative (b5�.49, p < .05). Figure 2

shows a model linking self-esteem and externalizing problems,

controlling for parent and peer relationships, IQ, and SES. This

model had good fit, w2(9)5 19.81, p < .05 (CFI5 .99; RMSEA

5 .04, p close fit 5 .78), and the relation between self-esteem

and externalizing problems remained significant (b 5�.32,

p < .05). Thus, parent and peer relationships, IQ, and SES

could not explain the relation between self-esteem and

delinquency.3

Cross-Lagged Relations Between Self-Esteem and Externalizing

Problems

To examine the effect of self-esteem at age 11 on externalizing

problems at age 13, we conducted cross-lagged analyses con-

trolling for prior levels of externalizing problems. Although

these analyses do not establish causal direction, they help rule

out alternative causal interpretations related to temporal se-

quence. We created latent measures of self-esteem at ages 11

and 13 using five two-item parcels of RSE items, and latent

measures of externalizing problems at ages 11 and 13 using an

indicator of aggressive behaviors (a standardized composite of

Fig. 2. Model linking self-esteem and externalizing problems in Study 2 (N 5 830). Standardized coefficients are reported, and asterisks indicate structural coefficients significant at the .05 level.

3There was no evidence that gender, IQ, or SES moderated the relation be- tween self-esteem and externalizing problems.

Volume 16—Number 4 331

M.B. Donnellan et al.

parent and teacher reports) and an indicator of nonaggressive

behaviors (a standardized composite of parent and teacher re-

ports). To improve model fit, we allowed the errors for the in-

dicators at age 11 to correlate with the same errors at age 13.

Figure 3 shows the cross-lagged model. This model had good

fit, w2(64)5 116.06, p < .05 (CFI5 .99; RMSEA5 .03, p close

fit 5 1.00), and self-esteem was concurrently related to exter-

nalizing problems at age 11 (r 5�.31, p < .05) and age 13

(r 5�.25, p < .05). The path linking self-esteem at age 11 to

externalizing problems at age 13 was negative and significant

(b 5�.15, p < .05), whereas the path linking externalizing

problems at age 11 to self-esteem at age 13 was negative but not

statistically significant (b 5�.08, p 5 .11). These results are

consistent with the claim that low self-esteem leads to increases

in externalizing problems. However, given the magnitude of the

effect and the nonexperimental design, we are hesitant to con-

clude that self-esteem causes future externalizing problems.

STUDY 3

The results of Studies 1 and 2 support the low-self-esteem hy-

pothesis. In Study 3, we tested the hypothesis (Baumeister et al.,

1996, 2003) that unrealistically high, not low, self-esteem

predicts aggression by assessing both self-esteem and narcis-

sism and examining their relations with reports of real-world

aggression. Previous research on narcissism has used laboratory

measures of aggression, and it is not clear whether the findings

generalize to real-world aggression.

Method

Sample

The sample consisted of 3,143 undergraduate students (68.3%

female; mean age5 19.6 years, SD 5 1.6) from a large research

university in northern California. They participated in exchange

for course credit.

Measures

Self-esteem was measured with the RSE (a 5 .90). Narcissism

was measured by the 40-item Narcissistic Personality Inventory

(Raskin & Terry, 1988; a5 .84). Aggression was assessed using

the 29-item Buss-Perry Aggression Questionnaire (AQ; Buss &

Perry, 1992). The AQ includes a Total Aggression scale (a 5

.90) and four subscales: Physical Aggression (a 5 .83), Verbal

Aggression (a5 .76), Anger (a5 .81), and Hostility (a5 .84).

Results and Discussion

Table 1 shows correlations of self-esteem and narcissism with

aggression.4 Results were consistent with the findings of Studies

1 and 2: Self-esteem was negatively correlated with the Total

Aggression scale of the AQ (r 5�.30, p < .05) and with all of

the subscales except Verbal Aggression. Note that self-esteem

was related to the Physical Aggression subscale, which has

been linked to real-world displays of violence (Bushman &

Wells, 1998). In contrast, narcissism was positively correlated

Fig. 3.Cross-lagged model linking self-esteem and externalizing problems in Study 2 (N 5 830). Standardized coefficients are reported, and asterisks indicate structural coefficients significant at the .05 level. The five two-item parcels of self-esteem items are labeled SE1 through SE5.

4Gender did not moderate any of these relations.

332 Volume 16—Number 4

Self-Esteem and Externalizing Problems

with the Total Aggression scale (r 5 .18, p < .05) and with all of

the subscales except Hostility. Thus, we found support for the

claim that narcissistic individuals are prone to aggression.

Self-esteem and narcissism were moderately related (r 5 .32,

p < .05), so we conducted multiple regression analyses to test

whether they had independent effects on aggression (Table 1).

In general, the effect sizes increased in the multiple regression

analyses (e.g., the zero-order relation between self-esteem and

Total Aggression was �.30, whereas the regression coefficient

controlling for narcissism was �.39). We conducted Sobel tests

to determine if these apparent suppression effects were statis-

tically significant (MacKinnon, Krull, & Lockwood, 2000). For

the total AQ scale and all four subscales, the effects of self-

esteem were significantly stronger when narcissism was in-

cluded in the equation than when it was not included (all zs >

�4.90, ps < .05), and, similarly, all of the effects of narcissism

were significantly stronger when self-esteem was included in

the equations (all zs > 6.28, ps < .05). Thus, low self-esteem

and narcissism contribute independently to aggressive

thoughts, feelings, and behaviors, and in fact serve as mutual

suppressors.5

GENERAL DISCUSSION

In three studies, we found a robust relation between low self-

esteem and externalizing problems. This relation held for dif-

ferent age groups, different nationalities, and multiple methods

of assessing self-esteem and externalizing problems; after

controlling for potential confounding variables; and when we

delved beneath the broad construct of externalizing problems

and examined specific aggressive thoughts, feelings, and be-

haviors. Moreover, our results indicate that self-esteem may

foretell future externalizing problems; 11-year-olds with low

self-esteem tended to increase in aggression by age 13. Finally,

the effect of low self-esteem on aggression was independent of

narcissism; in fact, when healthy self-regard was disentangled

from narcissistic self-perceptions, the relation between low self-

esteem and aggression became even stronger. Thus, our results

support the concern (Baumeister et al., 1996) about the dangers

of narcissism but do not support the conclusion that low self-

esteem is unrelated to externalizing problems. In this section,

we discuss conceptual and methodological issues that may help

explain the inconsistencies in the literature on the association

between low self-esteem and externalizing problems.

Baumeister et al. (1996) suggested that inflated high self-

esteem (as captured by measures of narcissism) is a better

predictor of aggression than low self-esteem. This suggestion

seems to be based on the assumption that low self-esteem and

narcissism are opposite ends of the same continuum (self-hate

vs. self-love). For example, Baumeister et al. noted that ‘‘an

effective and valid [self-esteem] scale would identify the arro-

gant, conceited narcissist just as well as the person who holds an

unbiased appreciation of his or her own well-recognized good

qualities’’ (pp. 28–29). Accepting this view may result in the

need to pit the low-self-esteem hypothesis against the narcis-

sism hypothesis; that is, antisocial individuals have either low

self-esteem or its antithesis, narcissism. Moreover, conceptu-

alizing low self-esteem and narcissism as opposite ends of the

same continuum leads to the concern that ‘‘the societal pursuit

of high self-esteem for everyone may literally end up doing

considerable harm’’ (Baumeister et al., 1996, p. 29).

However, this concern may not be warranted because it is

possible to draw a distinction between healthy self-regard and

narcissistic self-views. For example, Rosenberg (1965) noted

that ‘‘when we deal with self-esteem, we are asking whether the

individual considers himself adequate—a person of worth—not

whether he considers himself superior to others’’ (p. 62). In

contrast, narcissists describe themselves as special, extraordi-

nary people who are better and more deserving than others.

Empirically, measures of self-esteem and narcissism typically

correlate only in the .20s to low .30s, which is far below the level

of convergent validity one would expect between two self-report

measures of the same construct. Thus, the precise relation be-

tween self-esteem and narcissism remains an open question.

TABLE 1

Results of Correlational and Regression Analyses Predicting Aggression From Self-Esteem and Narcissism (Study 3)

Predictor

Total Aggression Physical Aggression Verbal Aggression Anger Hostility

Zero-order correlation b

Zero-order correlation b

Zero-order correlation b

Zero-order correlation b

Zero-order correlation b

Self-esteem �.30n �.39n �.11n �.19n .02 �.09n �.26n �.33n �.48n �.51n

Narcissism .18n .30n .21n .27n .31n .34n .14n .24n �.05n .11n

Multiple R — .41n — .28n — .33n — .35n — .49n

Note. N 5 3,143. np < .05.

5To address the concern that self-esteem lacks predictive validity after con- trolling for neuroticism (e.g., Judge, Erez, Thoresen, & Bono, 2002), we repeated the analyses reported in Table 1 for the subsample of participants (n 5 2,516) who completed the Big Five Inventory Neuroticism scale (John & Srivastava, 1999). Controlling for neuroticism reduced the size of the coefficients for self- esteem in Table 1, but they remained statistically significant in all cases (e.g., for predictions of Total Aggression, b for self-esteem5 �.27 and b for narcissism 5 .30, ps < .05).

Volume 16—Number 4 333

M.B. Donnellan et al.

Several conceptualizations are currently being debated in the

self-esteem literature, including whether narcissism is an ex-

aggerated form of high self-esteem, a particular facet of self-

esteem, a highly contingent and unstable form of self-esteem, a

need to feel superior to others, or a defensive shell of inflated

self-esteem that compensates for unconscious feelings of in-

adequacy (e.g., Campbell, Rudich, & Sedikides, 2002; Kirk-

patrick et al., 2002; Morf & Rhodewalt, 2001; Tracy & Robins,

2003). Although resolving this issue is beyond the scope of this

article, our results indicate that self-esteem and narcissism

have independent effects on externalizing problems, thus

demonstrating their discriminant validity. Moreover, when

narcissism is partialed out of self-esteem, the regression coef-

ficient for self-esteem more closely captures Rosenberg’s (1965)

conceptualization of self-esteem and provides clear support for

the low-self-esteem hypothesis.

Another way to reconcile the inconsistent results in the lit-

erature is to consider methodological differences between our

work and previous laboratory research on self-esteem and ag-

gression. Although experimental measures of aggression have a

great deal of external validity (Anderson & Bushman, 1997),

they do not necessarily have the same correlates as measures of

real-world aggression and antisocial behavior. In fact, in their

review of the literature, Anderson and Bushman (1997) noted

that lab and real-world studies sometimes produce discrepant

results, and suggested that ‘‘rather than take the perspective

that one ‘side’ or the other is wrong, it may be more prudent to

try to locate the source of the discrepancies in psychological

processes that may differ in the two settings’’ (p. 33).

There are several possible sources of the discrepancy between

our findings and those of the previous lab studies. First, lab

studies typically examine a specific form of aggression, namely,

aggression provoked by a competitive task in which self-evalu-

ative processes have been activated. In contrast, real-world ex-

ternalizing problems occur in a wide range of contexts, and these

other forms of aggression may have distinct correlates.

Second, aggressive behavior in the lab does not lead to any

serious harm to the other person, whereas real-world aggression

often does; blasting someone with white noise does not have the

same consequences as hitting someone. The correlates of milder

forms of aggression, particularly those that have no clear neg-

ative repercussions, may differ from the correlates of other

forms of aggression.

Third, aggressive behaviors occurring in the lab are not anti-

social to the same extent as real-world aggression. One could

argue that it is socially appropriate to blast one’s opponent with

white noise in the context of an experiment that has been

sanctioned by the university. In contrast, the externalizing be-

haviors assessed in the present research are explicitly socially

undesirable, antisocial, and in most cases illegal. Thus, the

discrepancy between our findings and those of previous lab

studies may reflect the fact that individuals with narcissistically

high self-esteem are more likely to be aggressive when it is

socially desirable (e.g., lab paradigms for assessing aggression,

athletic events, some corporate settings), whereas individuals

with low self-esteem are more likely to be aggressive when it is

socially undesirable and contrary to social norms. Future re-

search should examine the specific motivational processes

underlying different forms of aggressive behavior in individuals

with low versus high self-esteem.

Finally, the relation between low self-esteem and aggression

was generally small to moderate in the present studies. This

result might provide the simplest explanation for inconsisten-

cies in the literature. If the true effect size is small, then it is not

surprising that some studies have reported null findings be-

cause of lack of power and fluctuations in observed effect sizes

across samples due to systematic and random factors. Moreover,

from a meta-analytic perspective, variation in effect sizes across

studies indicates the presence of moderator variables. Thus,

researchers need to develop theoretical models that generate

testable predictions about the boundary conditions on the effect

of low self-esteem.

Although much work on these exciting and controversial

topics remains to be completed, we believe it is reasonable to

conclude that both low self-esteem and narcissism contribute to

externalizing problems. Our findings provide strong support for

a replicable link between low self-esteem and externalizing

problems, and we recommend that the low-self-esteem hy-

pothesis not be discarded prematurely.

Acknowledgments—We thank Marc Braverman, Rand Con-

ger, Arthur Giraco, and Phillip Shaver for their helpful com-

ments; Keith Widaman for his statistical advice; HonaLee

Harrington for her assistance with the Dunedin study data; and

Richie Poulton for his work as unit director of the Dunedin

study. Studies 1 and 3 were supported by a grant from the Na-

tional Institute on Aging (AG022057-01). Study 2 was sup-

ported by grants from the Health Research Council of New

Zealand, the U.K. Medical Research Council (G0100527), and

the National Institute of Mental Health (MH45070, MH49414).

REFERENCES

Adler, A. (1956). The individual psychology of Alfred Adler: A system- atic presentation in selections from his writings (H.L. Ansbacher & R.R. Ansbacher, Eds.). New York: Harper.

Anderson, C.A., & Bushman, B.J. (1997). External validity of ‘‘trivial’’

experiments: The case of laboratory aggression. Review of General Psychology, 1, 19–41.

Armsden, G.C., & Greenberg, M.T. (1987). The inventory of parent and

peer attachment: Individual differences and their relationship to

psychological well-being in adolescence. Journal of Youth and Adolescence, 16, 427–454.

Baumeister, R.F., Bushman, B.J., & Campbell, W.K. (2000). Self-es-

teem, narcissism, and aggression: Does violence result from low

self-esteem or from threatened egotism? Current Directions in Psychological Science, 9, 26–29.

334 Volume 16—Number 4

Self-Esteem and Externalizing Problems