Sunteți pe pagina 1din 13

Personality and Social Psychology Review Copyright (C 2002 by

2002, Vol. 6, No. 1, 59-71 Lawrence Erlbaum Associates, Inc.

Self-Esteem and Socioeconomic Status: A Meta-Analytic Review


Jean M. Twenge
Department of Psychology
San Diego State University

W. Keith Campbell
Department of Psychology
University of Georgia

Socioeconomic status (SES) has a small but significant relationship with self-esteem
(d = .15, r = .08) in a meta-analysis of 446 samples (total participant N = 312,940).
Higher SES individuals report higher self-esteem. The effect size is very small in
young children, increases substantially during young adulthood, continues higher un-
til middle age, and is then smallerfor adults over the age of60. Gender interacts with
birth cohort: The effect size increased over time for women but decreased over time
for men. Asians and Asian Americans show a higher effect size, and occupation and
education produce higher correlations with self-esteem than income does. The results
are most consistent with a social indicator or salience model.

Self-esteem and socioeconomic status (SES) are teem/SES link, using moderators such as age, gender,
two of the most important and frequently examined race and culture, and SES measurement to confirm or
psychological and sociological variables. At their in- disconfirm each model.
tersection we find the relationship between the individ- Before we begin, it is important to define both
ual's view of self (self-esteem) and society's primary self-esteem and SES. Self-esteem is usually defined as
view of the individual (SES). Yet the empirical rela- "the extent to which one prizes, values, approves, or
tionship between these two variables is very unclear, likes oneself' or "the overall affective evaluation of
with some studies finding a positive relationship, some one's own worth, value, or importance" (Blascovich &
negative, and some no relationship at all (e.g., Tomaka, 1991, p. 115). This article focuses on self-es-
Gray-Little & Appelbaum, 1979; Kohr, Coldiron, teem within this definition-that is, as an overall eval-
Skiffington, Masters, & Blust, 1988; Mullis, Mullis, & uation of the self. SES can also be a difficult term to
Normandin, 1992; Rosenberg & Pearlin, 1978; define. In this analysis, any study measuring a respon-
Trowbridge, 1972; Wiltfang & Scarbecz, 1990; Wylie, dent's (or their parents' or family's) educational attain-
1974; Zirkel & Moses, 1971). The inconsistent find- ment, occupational status, income, or a composite of
ings suggest the existence of powerful moderators, these is included in the definition of SES. Although
which may be examined using a meta-analytic strategy. these indicators of SES are separable and different,
For example, is SES more important to self-esteem they tend to be correlated. (For example, occupations
with increasing age and occupational achievement? with higher status generally offer more salary than
Have there been changes in the importance of SES to those with low status.) More important, these indica-
the self-esteem of women? A meta-analysis opens the tors can be subsumed in the same theoretical rubric,
possibility of delineating the sources of variation in the measuring social class and social status. In many ways,
research to date and thus the theoretical underpinnings SES categories are the quintessential example of "sta-
of the relationship between self-esteem and SES. tus groups"-categories of people with unequal
In this article, we used meta-analytic techniques to amounts of prestige (Gerth & Mills, 1946).
summarize 446 studies of self-esteem and SES. Taken Moderator variables might help explain the process
together, these studies include almost a third of a mil- and causation between SES and self-esteem. Age differ-
lion participants (total N = 312,940). We present and ences show how the source of SES might matter for
then test several theoretical models of the self-es- self-esteem: Children's SES is ascribed from their par-
ents, whereas adult SES is derived from the individual
and is presumably earned (e.g., Rosenberg & Pearlin,
Requests for reprints should be sent to Jean M. Twenge, Depart-
ment of Psychology, San Diego State University, 5500 Campanile 1978). Birth cohort, as a proxy for the larger social envi-
Drive, San Diego, CA 92182-4611. E-mail: jeant@umich.edu or ronment, may also affect the link between SES and
jtwenge@ mail.sdsu.edu self-esteem (Caspi, 1987; Elder, 1974, 1981; Gergen,

59
TWENGE & CAMPBELL

1973; Nesselroade & Baltes, 1974; Twenge, 2000, 2001; searchers found that Black Americans reported self-es-
Twenge & Campbell, in press). In particular, women's teem equal to or higher than that of Whites (e.g., Porter
work roles have changed tremendously over the last 40 & Washington, 1979; however, the model may apply
years, a shift that is likely to affect women's individual differently to SES groups compared to racial groups).
characteristics (Stewart & Healy, 1989; Twenge, 2001). These findings spurred a reappraisal of minority
In her social role theory, Eagly (1987) argued that work self-esteem centered on self-protective mechanisms
and family roles cause gender differences in personality (Crocker & Major, 1989). Each model makes specific
and self-views. Thus, women's changing work roles predictions for the overall effect size and four primary
might affect the correlation between SES and self-esteem moderator variables (age, gender interacting with birth
for women; gender and birth cohort may interact. In addi- cohort, race and culture, and SES measurement).
tion, culture and race may affect the relationship. Cul- These predictions are summarized in Table 1.
tures and racial groups vary in their perspectives, and
these perspectives affect personality, emotion, and cogni-
tion (e.g., Choi, Nisbett, & Norenzayan, 1999; Heine & Social Indicator or Salience Model
Lehman, 1997; Markus & Kitayama, 1991; Markus, Because SES is an indicator of status within social
Kitayama, & Heiman, 1996; Markus, Kitayama, & groups, elevated self-esteem should result from ele-
VandenBos, 1996; Nisbett & Cohen, 1996; Suh, Diener, vated SES (Rosenberg & Pearlin, 1978). If an indi-
Oishi, & Triandis, 1998). Values placed on education and vidual aspires to success in the form of social status
income clearly differ among cultures, a factor that could and wealth and achieves these goals, elevated self-es-
moderate the SES/self-esteem effect. The way SES is teem should result (e.g., James, 1890/1950; Pelham,
measured might also be an important moderator. The 1995; Pelham & Swann, 1989). Conversely, the indi-
most common measures of SES are education, occupa- vidual who does not achieve social status may suffer
tion, and income; some of these might be more important from lowered self-esteem. This might be especially
to self-esteem than others. true in the United States, where SES is perceived as
changeable and earned (Hare, 1977). Thus, this
model predicts a positive correlation between self-es-
Models Relevant to SES teem and SES.
and Self-Esteem However, there is reason to expect that SES does not
have the same meaning and salience to every group.
We delineate three models we judged most relevant Thus, this model predicts that the relationship between
to SES and self-esteem: social indicator or salience, re- self-esteem and SES should be moderated by several
flected appraisals, and self-protective mechanisms. variables. As French and Kahn (1962) argued, people
The social indicator or salience model has been used to differ in what they consider important to their self-es-
explain the link between status and self-esteem (e.g., teem: Dimensions of the self may differ in their psy-
Rosenberg & Pearlin, 1978), taking the varying psy- chological "centrality." In the case of SES, individuals
chological centrality of status into account (French & probably differ in the emphasis they place on SES in
Kahn, 1962). The reflected appraisals model, some- their overall equation for self-esteem.
times called "internalization of stigma," was used in
the mid-20th century to theorize that oppressed groups Age. To explain age differences in the effect
should have low self-esteem (see, e.g., Cartwright, size, the social indicator or salience model notes the
1950; Erikson, 1956; Scott, 1997). This model fell following. In those college age or younger, SES is
somewhat out of favor during the 1980s when re- measured not by the individual's own income or ac-

Table 1. A Summary of Predictions


Overall Gender and
Model Effect Size Age Birth Cohort Race and Culture SES Measurement
Social Indicator or Positive Linear upward, then Increase over time for Based on salience Education and
Salience decrease in older women, decrease occupation higher
adults for men
Reflected Appraisals Positive Constant in children, Increase over time for Unclear for race; None
higher for adults. women, no change based on treatment
Possibly highest for men of others in culture
during middle age
Self-Protective Neutral or small Most negative for No changes More negative for None
Mechanisms positive school age, then less minorities
negative or neutral
Note: SES = socioeconomic status.

60
SELF-ESTEEM AND SOCIOECONOMIC STATUS

complishments, but by their parents'. Thus, for chil- more salient than another does, racial or national dif-
dren and adolescents, SES is an ascribed status, one ferences in the effect size may appear. However, it is
they have not earned or worked toward (Demo & difficult to make a priori predictions about each
Savin-Williams, 1983; Rosenberg & Pearlin, 1978). group's valuation of SES. Thus, we can only say that
For adults, however, SES reflects their own, earned racial and national differences would not be inconsis-
status, such as their education, their occupation, and tent with this model, although it is difficult to predict
their income. Thus, SES is a social indicator of the pattern of these differences. We can make one pre-
earned status only for adults. In addition, the salience diction: If culture influences the salience of SES to
or psychological centrality of SES to the self-concept self-esteem, we would expect to see similarities be-
may change with age (Rosenberg & Pearlin, 1978). tween samples of recent U. S. immigrants and samples
Young adults who have not yet attained their full from their home countries. For example, many Asian
earning power may not find SES as salient as mid- Americans are relatively recent immigrants; the effect
dle-age individuals, who are more likely to be at the size for Asian Americans should be similar to that of
peak of their careers and their income. Similarly, regional Asian samples. Research suggests that both
older adults and the elderly, often retired, may no lon- Asians and Asian Americans emphasize education
ger view their occupation and income as their pri- more than Westerners do (e.g., D'Ailly, 1992; Goyette
mary source of self-esteem. (Because SES is still an & Xie, 1999; Whang & Hancock, 1994). This suggests
earned status for older adults, it is not likely that the that Asian groups might show a stronger link between
effect size will return to the levels of childhood.) SES and self-esteem.
Thus, the salience or centrality model predicts that
the correlation between self-esteem and SES will
SES measurement. In addition, the way SES is
vary considerably depending on the age of the indi-
measured may also moderate the effect size based on sa-
vidual. It will begin very low in elementary school lience. For example, occupational and educational sta-
children, increase slowly until high school and col- tus may be more salient to personal self-esteem than in-
lege, rise considerably for young adults, increase come because they are better indicators of social status
again for the middle-aged, and then decrease for and importance. If this is true, the salience or centrality
older adults and the elderly.
model predicts higher effect sizes for occupational sta-
tus and educational level than for income.
Gender and birth cohort. This model also pre-
dicts gender differences across generations. In these
discussions, we refer to birth cohort or generation as an Reflected Appraisals
effective proxy for changes in the larger sociocultural
environment (see Twenge, 2000, 2001, 2002, for dis- According to the reflected appraisals model, we in-
cussions). Most relevant for changes in the ternalize others' perceptions of ourselves (Cooley,
SES/self-esteem link, women's labor force participa- 1902; Mead, 1934). This process is sometimes referred
tion has changed considerably over the years. In addi- to as the "looking-glass self' or the "internalization of
tion, women have increasingly pursued professional stigma." This paradigm applies to SES because SES in-
careers; the percentage of law, medicine, and doctoral fluences the way others treat us and is thus eventually
degrees awarded to women has quadrupled since the reflected in self-esteem. If others see us as lower class
early 1 960s (Statistical Abstract of the United States, and lower status, we are likely to see ourselves that way
U. S. Bureau of the Census, 1998). The changes for and therefore experience lower self-esteem. For exam-
men over this time have been more subtle. Men's la- ple, children and adults living in poor neighborhoods,
bor-force participation has not changed. However, they using food stamps, or wearing worn clothes may be os-
do not dominate the professions as they once did. In ad- tracized and demeaned by more advantaged people,
dition, men have relinquished sole ownership of the with obvious consequences for the self-concept
breadwinner role, as fewer men are now solely respon- (Symonds, 1968; Wiltfang & Scarbecz, 1990). Con-
sible for supporting their families. Using the social in- versely, people with high education, high occupational
dicator or salience model, we reason that these gender prestige, and high income are often treated with respect
role changes have increased the psychological central- and consideration, perhaps leading to higher self-es-
ity of SES to women's self-esteem and decreased it for teem. Although the internalization of stigma model has
men's. Thus, this model predicts that the effect size received little support from studies of race differences
linking SES and self-esteem will increase over time for in self-esteem (Twenge & Crocker, in press), it may
women and decrease over time for men. still be an effective model for SES because SES is
viewed as changeable and earned, whereas race is not
Race and culture. The salience or centrality (Hare, 1977). Like the social indicator or salience
model may also have implications for racial and cul- model, the reflected appraisals model predicts a posi-
tural groups. If one culture or subculture considers SES tive correlation between self-esteem and SES.
61
TWENGE & CAMPBELL

Age. The reflected appraisals model predicts age how others treat us. This is difficult to say; in their own
differences, but only after young adulthood. Children ways, the SES indicators of education, occupation, and
old enough to complete a self-esteem measure are income are all visible to others. Thus, this model pre-
also old enough to internalize others' opinions, so dicts no significant differences in SES measurement.
there should be few differences between elementary,
junior high, and high school samples. After young
adulthood, the effect size should increase because Self-Protective Mechanisms
outsiders' judgments of low-SES adults are likely to
be harsher than their judgments of low-SES children The self-concept possesses a wide range of power-
(just as their judgments of high SES are likely to be ful self-protective strategies. These strategies may ef-
more positive for adults). Judgments of SES are fectively shield the self from external feedback due to
likely to be strongest for middle-aged adults because SES (e.g., the external feedback described in the re-
this is the age when income is expected to peak. Re- flected appraisals model previously discussed). In ad-
tired adults who are poor might be judged less dition, individuals in low SES groups may use a
harshly. On the other hand, judgments could be simi- self-serving bias (Campbell & Sedikides, 1999); that
lar for all adult age groups. People might think that is, they might take credit for the SES they have
young adults and retirees are just as responsible for achieved but blame any lowered SES on factors exter-
their SES as middle-aged adults are. nal to the self. Individuals may also strategically use
downward social comparisons. They may maintain
Gender and birth cohort. Because the reflected self-esteem by always comparing themselves to those
appraisals model is concerned with the perceptions of who are less fortunate, no matter what level of SES
others, change over time in attitudes toward women is they have obtained (Wills, 1981).
relevant for the gender and birth cohort moderator. An excellent example of such a strategy was pro-
Most evidence indicates that attitudes toward women's vided by Crocker and Major (1989), who argued that
working have grown steadily more positive over time belonging to a low-status group does not necessarily
(Thornton, Alwin, & Camburn, 1983; Twenge, 1997). lead to lowered self-esteem. Instead, group members
Thus, the correlation between SES and self-esteem may protect their self-esteem by attributing negative
should increase over time for women: High-SES feedback to prejudice, comparing themselves only to
women have been judged more positively over time, others in their group, or devaluing the dimensions in
and this should be internalized in the form of self-es- which the group performs poorly. Such self-protective
teem. There is little work on changes in attitudes to- strategies may help explain why Blacks actually have
ward men's working over time; it seems likely that higher self-esteem than Whites do (Gray-Little &
these attitudes have not changed much, as men's work Hafdahl, 2000; Twenge & Crocker, in press).
patterns have not substantially altered. Thus, the corre- However, it is not clear that self-protective mecha-
lation between SES and self-esteem should not change nisms operate for SES in the same way they do for
over time for men. race. As Hare (1977) pointed out, economic classes
lack the solidarity, mutual support, and protection
Race and culture. The reflected appraisals that racial groups have. Individuals are less likely to
model does not make clear predictions for the influ- see class as a primary identity and are not as likely to
ence of race on the effect size. It is not clear that people attribute negative reactions to prejudice against their
treat low-SES people differently according to their class group. Some individuals, especially adults, may
race; in some cases, race may cause further prejudice, also feel at least partially responsible for their mem-
whereas in others it may soften the impact of low SES. bership in a lower social class, an attribution that mit-
It is also not clear if the treatment of SES varies within igates the use of self-protective mechanisms
racial groups (that is, in how members of the group (Crocker, Cornwell, & Major, 1993; Crocker & Ma-
treat each other). jor, 1989). The idea that people judge SES by social
However, within nations and cultures, there may be comparison (comparing themselves to similar others
a moderating effect similar to that predicted by the so- rather than larger ideals) has received support in
cial indicator or salience model. If some nations or cul- some studies (e.g., Hagerty, 2000) but not in others
tures place more importance on SES, they may treat in- (e.g., Diener, Diener, & Diener, 1995). If self-protec-
dividuals differently depending on SES. This is the tive mechanisms operate for class as they do for race,
external version of the cultural differences predicted there should be a negative relationship between SES
by the salience model. and self-esteem. However, if they are less powerful
(but still operative) for class-the most likely propo-
SES measurement. To determine if SES mea- sition self-protective mechanisms may reduce or
surement influences the effect size under this model, we nullify (though not reverse) any positive relationship
must know which aspects of SES are most relevant to predicted by the other models.
62
SELF-ESTEEM AND SOCIOECONOMIC STATUS

Age. This model predicts specific age-based dif- pation, education, and income). Thus, this model pre-
ferences in the effect size. Given the psychological skill dicts no differences based on SES measurement.
necessary to use self-protective mechanisms, elemen-
tary school students would not be very adept at using this
strategy. However, as the other models pointed out, they Overview
would also not be very affected by the social indicator or
salience or reflected appraisals mechanisms. Thus, they In this article, we examine the relationship between
should show little to no effect. Older children and ado- self-esteem and SES and several moderators (primar-
lescents would take fullest advantage of self-protective ily: age, gender interacting with birth cohort, race,
mechanisms: They have the maturity to use the tech- country or culture, and SES measurement; second-
nique and, most important, would not feel responsible arily: self-esteem measurement and published vs. un-
for their lower status (because their SES is ascribed from published data source). We interpret the results based
their parents). Individuals who feel that their lower sta- on three possible models: social indicator or salience,
tus is ascribed and not earned are much more likely to reflected appraisals, and self-protective mechanisms,
employ self-protective mechanisms (Crocker et al., each of which makes specific predictions for the effect
1993; Crocker & Major, 1989). Thus, junior high size and moderators (see Table 1).
school, high school, and college students would be most
likely to show a null effect (if self-protective mecha-
nisms reduce any positive relationship). The same rea- Method
soning predicts a positive effect size for adults (young
adults through the elderly) who might feel more respon- To locate studies for the meta-analysis, we searched
sible for low SES and thus would not have the fullest use three databases: PsycINFO (journal articles in
of self-protective mechanisms. psychology, sociology, and related fields searched 1967
to 1998; American Psychological Association,
Gender and birth cohort. Women's use of 1967-1998), ERIC (journal articles in education andre-
self-protective mechanisms might depend on birth co- lated fields searched 1966 to 1998; ERIC, 1966-1998),
hort and time. Because prejudice based on gender did and Dissertation Abstracts International (dissertations
not receive much attention before the late 1960s, and master's theses, a major source of unpublished data,
women might not have used self-protective mecha- searched 1960 to 1998; University Microfilms Interna-
nisms until later. Thus, the lower effect size predicted tional, 1960-1998). We searched these sources using
by this model would grow closer to zero for women the key words self-esteem or self-concept and socioeco-
over time as they attributed their low SES to sexism. nomic or income or class or SES. This search yielded
However, it is also possible that the feminist movement both studies reporting primary data and those reporting
has actually decreased women's use of self-protective data from national, large-scale studies; data from na-
mechanisms; if women believe that gender prejudice tional studies representing the same samples from the
no longer exists, they may actually be less likely to same year were entered only once. We relied on these
make prejudice-based attributions, causing the effect database searches and did not search the reference lists
size to rise. Given the unclear and possibly contradic- of articles because the vast majority of studies did not
tory nature of self-protective mechanisms based on concern themselves primarily with the SES differences
gender, this model in the end predicts no changes in the (thus reference lists would have yielded few useful data
effect size for women. Because these movements have points; two other recent meta-analyses on self-esteem
not affected men significantly, the correlation for men also did not search reference lists-Kling, Hyde,
is expected to remain constant. Showers, & Buswell, 1999; Gray-Little & Hafdahl,
2000). We excluded articles written in a language other
than English at the database stage, given the difficulty in
Race and culture. This model might also predict coding effect sizes from such articles.
some moderating effect for race. Because Blacks may Studies using any measure of SES and any measure
use self-protective mechanisms to guard their self-es- of general self-esteem were retained for further exami-
teem against any racially based threats, they could be nation. For self-esteem measures, this included classic
more likely to employ these mechanisms against global scales such as the Rosenberg Self-Esteem Scale
SES-based threats as well. Thus, the effect size would be (Rosenberg, 1965), adjective scales, and omnibus mea-
closer to zero among Black samples (and possibly in sures summing over several areas of competence (e.g.,
Hispanic and Asian samples). This model does not pre- the Coopersmith Self-Esteem Inventory [Coopersmith,
dict any regional differences in the effect size. 1967], the Tennessee Self-Concept Scale [Fitts, 1965],
and the Piers-Harris Self-Competence Scale for
SES measurement. Self-protective mechanisms Children [Piers, 1969]). It did not include scales tap-
are likely to apply equally to all aspects of SES (occu- ping only narrow domains of self-esteem (e.g., aca-
63
TWENGE & CAMPBELL

demic self-esteem, body image, etc.) Studies examin- (Grades 6-8; ages 11-13); high school (Grades 9-12,
ing abnormal populations were excluded (e.g., ages 14-17); and college (ages 18-22). We also created
psychiatric or hospital patients, people with alcohol- three adult groups: young adult (23-39), middle age
ism, people with drug addictions, children in fos- (40-59), and older adults or the elderly (60 and up).
ter-home care, juveniles labeled as delinquent, chil- Samples were assigned to groups based on the average
dren with hyperactivity, gang members, or abuse age of the sample. An effect size for an age group was
survivors). Studies including fewer than 15 respon- computed by weighting for sample size if, for example,
dents were also eliminated from the initial search. a study reported means separately for different grades.
For self-esteem measures with subscales, we used Studies reporting statistics separately for male and
the total score to calculate the effect size. In most cases, female samples were counted as two separate data
the relationship between self-esteem and SES was re- points, as were studies reporting statistics for different
ported as an r; in other cases, t tests, F tests, or means and age groups, SES measurement, countries, self-esteem
standard deviations were reported. We converted all of measures, racial groups, or years of data collection.
these effect sizes to d, as that is the most universal metric However, studies reporting separate statistics for
for effect sizes derived from different statistical sources. mother's and father's education, occupation, or both
When means were reported for more than two SES were averaged to obtain a mean value.
groups, the groups were recombined to form two groups The ds grouped by moderating variables were then
(this is necessary because computing d requires com- tested for heterogeneity within the group (whether, for
paring two groups; when groups were combined they example, all studies of high school students showed a
were weighted by sample size). When there were an odd homogeneous effect size) and between groups
number of SES groups (e.g., three or five), we compared (whether age groups differed in their effect sizes, a pro-
lower SES groups to combined middle and higher SES cedure analogous to an analysis of variance). These
groups. Formulas specified by Wolf (1986) were em- heterogeneity statistics (Q) were calculated using the
ployed to convert statistics into the effect size d, or the formulas provided by Hedges and Becker (1986). The
difference between two groups in terms of standard de- distribution follows an approximate chi-squared distri-
viations (although we also used techniques not men- bution, with k - 1 degrees of freedom (where k is the
tioned there, e.g., including studies that reported group number of studies) for the within-groups computation
means and an Fbut not a standard deviation). Studies re- and p - 1 degrees of freedom (where p is the number of
porting no effect (k = 7) were coded as d = .00. Because groups being compared) for the between-groups analy-
there were so few ofthese studies, their inclusion did not sis. Analyses with continuous variables were per-
significantly change any of the study results. formed using linear regressions weighted by n - 3.
For the analyses, we converted d to r and then con- This method yielded 446 effect sizes, involving a to-
verted r to Z using Fisher's r to Z transformation. This tal of 312,940 respondents (median n for individual
is the best technique for analyzing data when many of studies = 197). There were 287 studies (i.e., articles or
the data points are in terms of r (Hedges & Olkin, dissertations); there are more samples than studies be-
1985). The analyses were then weighted by n - 3 cause articles often reported statistics for different age,
(Hedges & Olkin, 1985; Shadish & Haddock, 1994). gender, racial, and measurement groups (see preced-
This ensures that samples with more respondents (as- ing). The studies were published between 1966 and
sumed to be better estimates of the population mean) 1998, with a median data collection year of 1981. The
are correspondingly weighted in computing the aver- median age of the study participants was 17 (M = 24),
age effect size. These Zs were then converted back to r with a range from 5 to 76. Most samples (n = 391) were
and then d (as noted previously, we relied on d as the collected in the United States. Published studies to-
best universal indicator of effect size). The effect size taled 244, whereas dissertations and masters' theses
in terms of r is also reported for most analyses. yielded 202 data points.
Studies providing enough information were then
coded for moderating variables: measure of SES used Results
(occupational status, education, income, or a composite
measure); year of data collection (assumed to be 2 years Overall Effect Size
prior to publication unless otherwise specified; see, e.g.,
Oliver & Hyde, 1993); measure of self-esteem used; age Is SES related to self-esteem? If so, what is the na-
of respondents; gender composition of sample; country; ture of the relationship? Across all 446 samples, the
racial composition of sample; urban, rural, or suburban weighted mean effect size (d) for the relationship be-
location; and type of study (published versus disserta- tween SES and self-esteem was .15. A test for
tion or master's thesis). The samples were categorized heterogeneity indicated that the samples were signifi-
into age groups, with the younger groups categorized cantly heterogeneous, Quotal = 2,542.20, p < .001. The
using the system employed in Kling et al. (1999): ele- mean unweighted d was .19, with a median of .1 8. The
mentary school (Grades 1-5; ages 6-10); junior high 95% confidence interval for the weighted d was .14 to
64
SELF-ESTEEM AND SOCIOECONOMIC STATUS

.16; because this confidence interval did not include sion also showed a significant curvilinear pattern for
zero, the effect was statistically significant. age (R2= .05,R = .22,p< .01).
Expressed in terms of a correlation (r), the weighted We also performed comparisons between specific
effect was .08. This meant that individuals with higher age groups. First, we compared school-age individuals
SES also scored higher on measures of self-esteem. (elementary school through college) to adults, thus as-
This is a small effect size in absolute terms (Cohen, certaining the difference between ascribed SES (one's
1977), very similar to the d for sex differences in parents' SES) and self-earned SES. The average effect
self-esteem (d = .16; Kling et al., 1999). size for the school-age groups was . 13, compared to .23
for the adult groups (QBet.een = 136.70, p < .001). We
also compared high school and college students to
Moderator Variables adults; the effect was still highly significant (QB =
We proceeded to try to explain some of the variance in 1 1 1.43, p < .001). High school and college students did
the effect sizes by examining relevant moderator variables. not differ significantly in effect size. The slightly
smaller effect size in college students might have been
Age. Age is an important moderator variable for caused by reduced variance in SES in this group.
determining the theoretical underpinnings of the We also made individual comparisons within the
SES/self-esteem link. The results showed that age was adult age groups. The effect size was significantly
a strong moderator. The effect size increased progres- higher in middle age than in young adulthood (QB =
sively at nearly every stage of development, reaching a 8.35,p < .01) and significantly higher in middle age than
peak at middle age but then declining in adults over the in the older adults and the elderly (QB = 12.1 3,p < .00 1).
age of 60 (see Table 2 and Figure 1). Age was also a Thus, the effect size increased from college into young
significant effect when retained as a continuous vari- adulthood and middle age and then decreased after re-
able, r(446) = . 18, p < .001 (in a bivariate linear regres- tirement age. The results for age are most consistent
sion weighted by n - 3). This correlation was inde- with the social indicator or salience model and the re-
pendent of birth cohort, as it remained the same with flected appraisals model. They are less consistent self-
the year in the regression equation. A quadratic regres- protective mechanisms (see Table 1 for predictions).

Table 2. Moderator Variables in the Socioeconomic Status and Self-Esteem Effect Size
95% CI
Variable and categories QB k d for d r QW
Age group 172.20*
Elementary school (5-10) 69 .08 .05 to .11 .04 277.38*
Junior High (11- 13) 65 .12 .10 to .14 .06 319.80*
High School (14-18) 99 .14 .13 to.15 .07 475.20*
College (18-22) 50 .10 .08 to.12 .05 120.50*
Young Adults (23-39) 58 .21 .19 to .23 .11 321.90*
Middle Aged (40-59) 77 .25 .23 to .27 .13 730.73*
Older Adults/Elderly (60+) 28 .17 .13 to .21 .08 116.76*
Racial group (United States Only) 51.45*
White 105 .14 .12 to.16 .07 533.40*
Black 74 .18 .15 to .21 .09 245.68*
Hispanic 24 .05 .00 to .10 .02 155.76*
Asian 7 .46 .35 to .57 .23 19.11*
Country or Culture
Western (United States, Canada, Europe, Australia) 60.98* 424 .15 .14 to .16 .08 2,391.36*
Asia 15 .29 .25 to .33 .15 38.70*
Africa 7 .04 .00 to .08 .02 50.68*
SES Measurement 58.16*
Occupation 109 .20 .18 to .22 .10 452.35*
Education 95 .18 .16 to .20 .09 712.50*
Income 109 .11 .09 to .13 .06 633.29*
Composites 133 .14 .13 to.15 .07 696.97*
Self-Esteem Measurement 141.80*
Rosenberg Self-Esteem Scale 178 .17 .16 to.18 .08 1,101.82*
Coopersmith Self-Esteem Inventory 45 .14 .11 to .17 .07 232.20*
Piers-Harris Self-Competence Scale for Children 30 .26 .20 to .32 .13 60.60*
Tennessee Self-Concept Scale 30 .19 .14 to .24 .09 135.90*
All Other Scales 163 .03 .01 to .05 .02 862.27*
Note: QB = between-groups heterogeneity; QW = within-groups heterogeneity; CI = confidence interval.
*p< .001.
65
TWENGE & CAMPBELL

0.35 self-esteem increased in successive birth cohorts of


0.30
women, with women born later showing a higher effect
size. The opposite effect occurred for men, who
0.25 showed a negative correlation between the effect size
and birth year, r = -.37, p < .001, k = 56. Men born in
0.20 later years demonstrated a smaller effect size than
those born earlier. As Figure 2 displays, the regression
0.15L lines cross; whereas the effect size was once stronger
0 10
for men, it is now stronger for women. It is also inter-
esting to note that the regression lines cross at the
0.05 beginning of the Baby Boom generation (circa
1940-1945), demonstrating the generational effect.
0lementary Junior High High School College Young adult Middle age Older adult The interaction between birth cohort and gender
Age group
also occurs when all samples (mixed sex as well as sin-
gle sex) are included. In this analysis, gender was
Figure 1. Age and effect size.
coded as the percentage of the sample that was female
(e.g., 0%, 50%, 80%, 100%). The interaction term of
Gender and birth cohort. Several models pre- Birth Cohort x Gender was a significant predictor of
dicted that gender and birth cohort would interact. To the effect size, r = .18, p < .01, k = 197. These results
compute the birth year of each sample, we subtracted are most consistent with the social indicator or salience
the sample's average age from the year the data were model; the reflected appraisals model predicted the in-
collected. In these analyses, birth cohort is a proxy for crease for women, but not the decrease for men.
changes in the sociocultural environment. Using a re-
gression model weighted by n - 3, we examined the
correlation between birth year and d separately for Race and culture. Several models predicted dif-
samples of adult women and men in the United States. ferences based on race and country or culture. We first
We limited the analysis to adults because we expected examine overall differences between racial groups
that only earned SES would be affected by historical within the United States. Analyses showed that racial
changes. Only U.S. samples were included to focus on group was a significant moderating variable (see Table
the social change within one culture; social change 2). Hispanics had the smallest effect size (d = .05; QB
may have taken different paths in other cultures. compared to Whites = 13.75, p < .001). Asian Ameri-
Has the effect size for self-esteem and SES changed cans showed the largest effect size; with d = .46, it ap-
over time for women and men? The results show that it proached half a standard deviation (QB compared to
has (see Figure 2 for a plot of the regression lines). For Whites = 32.45, p < .001). Although Asian American
the sake of simplicity, we examined the effect first for samples were predominantly adult (71 % compared to
single-sex samples (effect sizes reported for either 48% in the overall sample), this age distribution was
all-female or all-male samples). For all-female sam- not responsible for the larger effect size: The two sam-
ples (k = 55), the correlation between d and birth year ples of Asian American children actually had a larger
was .42 (p < .001). Thus, the link between SES and effect size (d = .56) than the five samples of adults (d =
.39). The effect size for Blacks was not significantly
different than that for Whites.
Females Regional differences between countries were con-
0.6 --- Males sistent with the White-Asian U.S. racial differences
0.5 (see Table 2). Asian samples showed a higher effect
0.4 j size than Western samples (QB for the two regions =
a1) 37.45, p < .001). This suggests that Asian culture sees
0.3
SES as more salient to self-esteem. African nations
LL
0.2 show the opposite result, with an effect size smaller
0.1
than that of Western nations (QB = 18.17, p < .00 1). Af-
rican cultures may see SES as less salient; alterna-
0.0 tively, there may be less variance in SES in these coun-
-0.1 tries, making it less relevant to self-esteem or lowering
-0.2 any correlation between the two variables. These re-
1920 1930 1940 1950 1960 1970 1980 sults were best predicted by the social indicator or sa-
Birth year
lience model, partially predicted by the reflected ap-
Figure 2. Birth cohort changes in the effect size, interaction by praisals model, and not predicted by the self-protective
sex. mechanisms model (see Table 1).
66
SELF-ESTEEM AND SOCIOECONOMIC STATUS

SES measurement and other moderators. The Table 3. Multiple Regression Analysis: Predicting the
way SES was measured influenced the effect size Effect Size From Moderator Variables
(see Table 2). Occupation and education correlated Variable
more highly with self-esteem than did income. Com-
posites of SES displayed an intermediate effect size. Age 39**
Age Squared (Quadratic Term for Age) .22**
These were the results predicted by the social indica- Birth Cohort x Gender Interaction .11 *
tor or salience model. Year -.01
Self-esteem measurement had a small but signif- White -.02
icant influence on the effect size (see Table 2). The Black .03
Rosenberg scale (Self-Esteem Scale; Rosenberg, Hispanic -.1 1*
Asian .10*
1965) and Coopersmith scale (Self-Esteem Inven- Occupation as SES Measure .07
tory; Coopersmith, 1967) were the two most com- Education as SES Measure .02
mon self-esteem measures; the effect sizes pro- Income as SES Measure -.15**
duced by these scales did not differ. The Note: SES = socioeconomic status. A negative beta does not mean
Piers-Harris scale (Self-Competence Scale for that the variable produces a negative effect size; rather, a negative
Children; Piers, 1969) showed a larger effect size, beta indicates that higher levels of the moderator variable are associ-
whereas miscellaneous, primarily unstandardized ated with smaller effect sizes.
measures showed smaller ones. The miscellaneous *p < .05. **p < .01. ***p < .001.
scales may have produced a smaller effect size due
to lower reliability and validity. cant, beta = .22. The Birth Cohort x Gender interaction
A few other potential moderating variables did not was significant; the beta was lower here than in the
reach significance. Location of sample (urban, subur- bivariate correlation, most likely because this analysis
ban, rural) and type of study (published vs. unpub- included all age groups (the bivariate analysis included
lished) did not show any significant relationships with only adults, where SES is more salient). Asians showed
effect size. a significantly higher effect size and Hispanics a signifi-
cantly lower one than did Whites and Blacks. (The effect
sizes for Whites and Blacks did not differ significantly.)
A Regression of Moderators Measuring SES using income led to a lower correlation
between self-esteem and SES.
To give a more complete picture of our findings, we
entered previously significant moderator variables into
a multiple regression equation with the effect size as Discussion
the dependent variable and n - 3 as a regression weight.
Non-U.S. samples were excluded to clarify the analy- We began this article with the goal of clarifying the
sis for race and the birth cohort and gender interaction. previously confusing literature on the relationship be-
Dummy variables were necessary in some cases; in tween SES and self-esteem. Our main result can be sum-
general, samples with specific characteristics (e.g., marized as follows: There is a small but statistically sig-
from a specific racial group) were compared to sam- nificant positive relationship between self-esteem and
ples without this information (e.g., with samples of SES across 446 samples including 312,940 respon-
mixed or unknown racial origin). For SES measure- dents. Although small, this effect is not trivial. The ef-
ment, specific variables (income, education, and occu- fect size is similar in magnitude to sex differences in
pation) were compared to composite measures. The self-esteem (Kling et. al., 1999); Black and White dif-
gender and birth cohort interaction term was entered ferences in self-esteem (Gray-Little & Hafdahl, 2000;
using Birth Cohort x Percentage Female of each sam- Twenge & Crocker, in press); the effect of parental di-
ple. We also included the quadratic term for age (age vorce on children's well-being (Amato & Keith, 1991);
squared) to test the curvilinear pattern of age. Age was and sex differences in verbal ability, spatial visualiza-
centered by subtracting the mean to reduce the correla- tion, and social leadership (Ashmore, 1990).
tion between age and age squared. The data summarized here are very heterogeneous.
Results from the multiple regression are reported in Thus, it is difficult to discuss the effect size without ref-
Table 3. A positive beta in this analysis meant that the erence to moderators because the effect size varies so
moderator variable led to a higher effect size; a negative much among groups (particularly age groups). There
beta meant that it led to a lower effect size (not necessar- was a smaller association between SES and self-es-
ily a negative one). The multiple R was .35, p < .001 teem in childhood and adolescence (when SES reflects
across 391 samples. The largest effect was for age, parents' earned status rather than children's earned sta-
which rose to .39 with all variables included (compared tus). The SES/self-esteem link increased significantly
to .18 for the bivariate correlation between age and the in young adulthood, further increased in middle age,
effect size). The quadratic term for age was also signifi- and then decreased after retirement age.
67
TWENGE & CAMPBELL

Gender and birth cohort interact as moderators. more salient to self-esteem in Asians and less salient in
Over time, the correlation increases for women and de- Africans and U. S. Hispanics. On the other hand, the
creases for men, producing a crossover effect. Asian effect may also be caused by the treatment of people in
and Asian American samples showed larger effect these cultures based on SES; the reflected appraisals
sizes, suggesting the existence of cultural differences model would predict that some cultures and subcul-
in the salience of SES for self-esteem. Finally, educa- tures judge people differently based on SES and others
tion and especially occupation showed the largest cor- do not. If these external opinions were internalized,
relations, with income producing a smaller effect. then differences in the correlation between self-esteem
and SES would result based on culture. Thus, either
model potentially explains the racial and cultural dif-
Theoretical Models ferences. The self-protective mechanisms model, how-
ever, would predict a lower effect size for all minori-
Which model best explains the pattern of data uncov- ties. This did not occur, as Asian minorities showed a
ered in the present meta-analysis? The overall effect larger effect size compared to Whites, Hispanics a
size, which is positive, is most consistent with the social smaller one, and Blacks the same as Whites.
indicator or salience and reflected appraisals models. Finally, the social indicator or salience model best
However, the positive effect is small; perhaps self-pro- predicts the moderating effect of SES measurement.
tective mechanisms dampen the effect produced by so- This model predicted that education and occupation,
cial indicator or salience and the reflected appraisals. arguably the most internally salient aspects of SES,
These results can be clarified by examining the would show the largest effects. This was indeed the
moderator variables. The results for the moderator case. The reflected appraisals and self-protective
variables were most consistent with the social indicator mechanisms models did not predict any differences.
or salience model (e.g., French & Kahn, 1962; Rosen- Thus, the social indicator or salience model fits the
berg, 1979; Rosenberg & Pearlin, 1978). Age demon- data the most closely. The reflected appraisals model
strated the pattern predicted by this model: It increased also made several correct predictions. All three models
until high school, rose during young adulthood and may explain the direction and size of the overall effect,
middle age, and decreased after retirement age. The re- but the moderators are clearly best explained by the sa-
sults for adults also support the salience model. SES is lience model. This does not imply that the reflected ap-
the most salient to the self-esteem of middle-age adults praisals and self-protective mechanism models of
and somewhat less salient for young adults and retired self-esteem are wrong; they simply do not explain dif-
people. The reflected appraisals model also predicted ferences based on SES as well as the salience model
larger effect sizes for adults compared to children, but does. This may reflect the weaker overall influence of
it did not predict the curvilinear pattern evident in the these models; alternatively, these models may operate
adult data. However, as noted in the introduction, it is only in limited domains.
possible that people judge middle-age adults most These results provide an interesting comparison to
stringently on their SES. Thus the results are not incon- results on racial differences in self-esteem. Studies
sistent with reflected appraisals. Self-protective mech- have consistently found that Black Americans report
anisms should work better for children than adults higher self-esteem than Whites (Gray-Little &
(consistent with the results), but no differences among Hafdahl, 2000; Twenge & Crocker, in press); however,
adult age groups were expected. Thus the social indica- other racial groups report lower self-esteem compared
tor or salience model best explains the age differences. to Whites. Twenge & Crocker (in press) found that
The social indicator or salience model also made the these differences were best explained by cultural views
most accurate predictions for the interaction between and tendencies, especially those surrounding individu-
gender and birth cohort. As predicted by this model, alism (Oyserman, Coon, & Kemmelmeier, 2002).
the effect size increased for women over time and de- These cultural views have some similarities with the
creased for men, based on the salience of SES to salience model, which essentially proposes that the sa-
self-esteem. The reflected appraisals model expected lience of SES depends on stage of life and culture. The
the increase for women, but not the decrease for men, reflected appraisals model, referred to as internaliza-
whereas self-protective mechanisms predicted no tion of stigma in Twenge and Crocker (in press), did
changes. However, the reflected appraisals model not fit the data on race differences in self-esteem. For
might help explain the decrease in the correlation for SES, however, it conformed to the data reasonably
men; perhaps societal expectations have shifted away well, explaining the overall effect size and aspects of
from men valuing the breadwinner role above all else some moderators. This suggests that low SES is inter-
and men have internalized this view. nalized as a stigma in a way that race is not. Observers
The differences based on race and culture are best may view a poor person as responsible for their fate,
explained by either the social indicator or salience particularly if that person is an adult. Even the person
model or the reflected appraisals model. SES may be may see it this way, rendering self-protective mecha-
68
SELF-ESTEEM AND SOCIOECONOMIC STATUS

nisms useless (Crocker et al., 1993; Crocker & Major, variables would be useful in mapping the social
1989). Whether people are responsible for their own sources of self-esteem (e.g., parenting, peer relations,
poverty is a question that lies beyond the scope of this etc.; see, e.g., Leary, Tambor, Terdal, & Downs, 1995).
article. However, it seems that SES may be a stigma Additionally, research could examine whether attribut-
similar to being overweight: People often perceive the ing negative feedback to prejudice based on low SES
poor and the fat as responsible for their conditions, protects self-esteem (as it does for race prejudice; cf.
whether they are or not (Crocker et al., 1993). Crocker & Major, 1989).
Second, our findings regarding racial and cultural
differences in the SES/self-esteem link should be con-
Strengths and Limitations sidered tentative. We noted that the SES/self-esteem
link was strongest for Asians, lowest for Hispanics, and
The data presented here have several obvious moderate for Whites and Blacks. These results are con-
strengths. First, we summarized a broad field of data, sistent with a salience model only if SES is somewhat
including over 450 samples and almost a third of a mil- more important to the self-esteem of Asians than other
lion respondents. This offers tremendous benefits for racial groups (some research does suggest that educa-
statistical reliability, an especially important quantity tion is more important to Asians; e.g., D'Ailly, 1992;
given the inconsistency of past findings. It also helps Goyette & Xie, 1999; Whang & Hancock, 1994). In ad-
ensure that the age differences in particular are solid, dition, we did not include studies written in a language
given their basis in the reports of thousands of respon- other than English; this necessarily limited the number
dents from each group. of international samples. Although the samples we did
We were able to examine many variables that indi- find did not differ in overall effect size from the U. S.
vidual studies usually cannot. For example, individual samples, this probably occurred because the effect size
studies usually cannot include such a complete range was higher in some regions (Asia) than in others (Af-
of ages; here, we summarized data from samples with rica). Further research might explore more fully how the
average ages from 5 to 76. We were also able to include SES/self-esteem link differs between cultures.
many possible moderators in a regression equation, de- Last, we did not explore all possible models and ex-
termining which variables produce unique variance in planations for the SES/self-esteem link. For example, it
the effect size. In addition, our examination of the in- is possible that having higher self-esteem leads to higher
teraction between gender and birth cohort is effectively SES. The age effect might be the result of this causality.
impossible for an individual study, given the difficul- Children can do little to change their SES, whereas
ties of collecting longitudinal data for more than one adults (particularly middle-age adults) might be able to
cohort. Using meta-analytic techniques to summarize translate high self-esteem into SES success. For retired
data collected over a 32-year period (1964 to 1996) adults, it is probably too late for high self-esteem to lead
permitted us to complete this analysis. This long time to success in education or jobs. The interaction between
span also allowed us to conclude that the age effect was gender and birth cohort might also be explained this
not due to birth cohort (Schaie, 1965) because year of way: Women could have developed higher self-esteem
data collection does not affect the correlation between first, which lead them to pursue higher education and
age and the effect size. However, controlling for year betterjobs (although this seems unlikely given that both
does not entirely eliminate the possibility of an age by genders increased in self-esteem from the 1970s to the
cohort interaction. Thus, we cannot definitively say 1990s; Twenge & Campbell, 2002). In addition, a third
that birth cohort does not influence the age effect, but variable (e.g., extraversion) might be correlated with
only that the effect is most likely due to age. both SES and self-esteem (Jackson & Gerard, 1996).
Our conclusions must include several caveats. First, Another limitation of the data we examined is their
the effect size found in this research is small. This does emphasis on one-time measurement. Some ofthe adults
not mean the results are theoretically unimportant; in the samples could have had a very different SES dur-
these results, as noted, are similar to those of other ing their childhood (either lower or higher than their
meta-analyses involving self-esteem. Nevertheless, adult status). If self-esteem is formed early, there might
these data show that SES, an important measure of so- be a mismatch between SES and self-esteem for adults
cial standing, accounts for only a small percentage of because their SES is measured for their adult status. This
the variance observed in individual self-esteem. This might help explain why the effect size is fairly small.
relationship is stronger when SES is salient and earned, Another explanation we did not explore in depth is so-
as with middle-age adults. Even in this group, however, cial dominance theory (Sidanius & Pratto, 1999). This
the size of the effect implies that there must be several theory posits that people in power (here, those high in
other variables causing (or being affected by) self-es- SES) create myths to legitimize their power. High
teem. That is, if SES does not explain a large percent- self-esteem might be the outcome of these legitimizing
age of the variance in self-esteem, what variables do? myths; people high in SES might have higher self-es-
Future research that simultaneously examines a host of teem because they believe their wealth and power must
69
TWENGE & CAMPBELL

mean they are better people. This can be a justification References


for their social position. However, the three theoretical
models we presented seemed to represent the most Amato, P. R., & Keith, B. (1991). Parental divorce and the well-being of
likely explanations for the observed correlation be- children: A meta-analysis. Psychological Bulletin, 110, 26-46.
American Psychological Association. (1967-1998). PsycINFO [Online].
tween the two variables (and indeed the social indicator Washington, DC: Author [Producer]. Ovid Technologies [Distribu-
or salience model fits the data very well). tor].
Ashmore, R. D. (1990). Sex, gender, and the individual. In L. A.
Pervin (Ed.), Handbook ofpersonality theory and research (pp.
Implications and Conclusions 486-526). New York: Guilford.
Blascovich, J., & Tomaka, J. (1991). Measures of self-esteem. In J. P.
Robinson, P. R. Shaver, & L. S. Wrightsman (Eds.), Measures
The results have several implications for research of personality and social psychological attitudes (pp.
and practice in a variety of areas. First, our findings 1 15-160). New York: Academic.
point to the importance of including age as a variable Campbell, W. K., & Sedikides, C. (1999). Self-threat magnifies the
when examining self-esteem. If we had limited our in- self-serving bias. Review of General Psychology, 3, 23-43.
Cartwright, D. (1950). Emotional dimensions of group life. In M. L.
quiry to only a certain age group, we would have Raymert (Ed.), Feelings and emotions (pp. 439-447). New
reached very different conclusions regarding the York: McGraw-Hill.
self-esteem/SES link. Indeed, the significant age effect Caspi, A. (1987). Personality in the life course. Journal of Personal-
may help account for the disagreement and inconsis- ity and Social Psychology, 53, 1203- 1213.
tency in the literature to date. Choi, I., Nisbett, R. E., & Norenzayan, A. (1999). Causal attribution
across cultures: Variation and universality. Psychological Bul-
Second, these results highlight the importance of letin, 125, 47-63.
examining birth cohort, a variable often unwisely ig- Cohen, J. (1977). Statistical power analysis for the behavioral sci-
nored in psychological research (Caspi, 1987; Gergen, ences (rev. ed.). New York: Academic.
1973). As seen in other research, birth cohort provides Cooley, C. H. (1902). Human nature and the social order New York:
a useful proxy for examining the effects of the larger Scribner's.
Coopersmith, S. (1967). The antecedents of self-esteem. New York:
sociocultural environment. These data in particular Freeman Press.
demonstrate the importance of examining the interac- Crocker, J., Cornwell, B., & Major, B. (1993). The stigma of over-
tion between birth cohort and gender (Stewart & Healy, weight: Affective consequences of attributional ambiguity.
1989; Twenge, 2001). Our findings regarding birth co- Journal of Personality and Social Psychology, 64, 60-70.
hort and gender would not have been observed if we Crocker, J., & Major, B. (1989). Social stigma and self-esteem: The
self-protective properties of stigma. Psychological Review, 96,
did not examine the two variables together. 608-630.
A final and related implication involves the differ- D'Ailly, H. H. (1992). Asian mathematics superiority: A search for
ence between ascribed and earned SES. First, the corre- explanations. Educational Psychologist, 27, 243-261.
lation between self-esteem and SES is weaker in chil- Demo, D. H., & Savin-Williams, R. C. (1983). Early adolescent
dren than in adults. There is clearly intuitive appeal in self-esteem as a function of social class: Rosenberg and Pearlin
revisited. American Journal of Sociology, 88, 763-774.
the prediction that low SES results in children with low Diener, E., Diener, M., & Diener, C. (1995). Factors predicting the
self-esteem. However, this conclusion is not well sup- subjective well-being of nations. Journal ofPersonality and So-
ported by the data. As noted in the introduction, the cial Psychology, 69, 851-864.
smaller correlation for children may occur because their Elder, G. H. (1974). Children of the Great Depression: Social change
SES is not earned. In contrast, the earned nature of SES and life experience.Chicago: University of Chicago Press.
Elder, G. H. (1981). Social history and life experience. In D. H.
for adults may account for the higher effect size in these Eichorn, J. A. Clausen, N. Haan, M. P. Honzik, & P. H. Mussen
groups. The notion of SES as earned may also help ex- (Eds.), Present and past in middle life (pp. 181-203). New
plain the increase in the SES/self-esteem correlation for York: Academic Press.
women over time. At one time, a woman's SES was ERIC database (1966-1998). University of Iowa. Distributor: First Search.
equivalent to her husband's; in later birth cohorts, mea- Erikson, E. (1956). The problem of ego-identity. Journal of the
American Psychoanalytic Association, 4, 56-121.
sured SES was more likely to include the woman's own Fitts, W. H. (1965). Tennessee Self-Concept Scale manual. Nash-
occupation or income (both due to measurement effects ville, TN: Counselor Recordings and Tests.
and to women's greater labor force participation). Thus French, J. R. P., & Kahn, R. L. (1962). A programmatic approach to
the correlation for women became higher when SES studying the industrial environment and mental health. Journal
measures incorporated women's earned SES. The con- of Social Issues, 18(4), 1-47.
Gergen, K. J. (1973). Social psychology as history. Journal of Per-
verging evidence from these two moderators (age and sonality and Social Psychology, 26, 309-320.
gender interacting with birth cohort) provides a view of Gerth, H. H., & Mills, C. W. (1946). From Max Weber: Essays in so-
how self-esteem operates within the individual. For ciology. New York: Oxford University Press.
self-esteem, it is not enough to simply enjoy the perks of Goyette, K., & Xie, Y. (1999). Educational expectations of Asian
high status; one must have had something to do with it. American youths: Determinants and ethnic differences. Sociol-
ogy of Education, 72, 22-36.
This invites the conclusion that self-esteem is more de- Gray-Little, B., & Appelbaum, M. I. (1979). Instrumentality effects
pendent on willful actions and choices than passively in- in the assessment of racial differences in self-esteem. Journal of
herited (or reflected) status. Personality and Social Psychology, 37, 1221-1229.

70
SELF-ESTEEM AND SOCIOECONOMIC STATUS

Gray-Little, B., & Hafdahl, A. R. (2000). Factors influencing racial Rosenberg, M. (1965). Society and the adolescent self-image.
comparisons of self-esteem: A quantitative review. Psychologi- Princeton, NJ: Princeton University Press.
cal Bulletin, 126, 26-54. Rosenberg, M. (1979). Conceiving the self. New York: Basic Books.
Hagerty, M. R. (2000). The social comparisons of income in one's Rosenberg, M., & Pearlin, L. I. (1978). Social class and self-esteem
community: National surveys of income and happiness. Jour- among children and adults. American Journal of Sociology, 84,
nal of Personality and Social Psychology, 78, 764-771. 53-77.
Hare, B. R. (1977). Racial and socioeconomic variations in Schaie, K. W. (1965). A general model for the study of developmen-
preadolescent area-specific and general self-esteem. Interna- tal problems. Psychological Bulletin, 64, 92-107.
tional Journal of Intercultural Relations, 1, 31-51. Scott, D. M. (1997). Contempt and pity: Social policy and the image
Hedges, L. V., & Becker, B. J. (1986). Statistical methods in the meta-anal- of the damaged Black psyche 1880-1996. Chapel Hill: Univer-
ysis of research in gender differences. In J. S. Hyde & M. C. Linn sity of North Carolina Press.
(Eds.), The psychology of gender: Advances through meta-analysis Shadish, W. R., & Haddock, C. K. (1994). Combining estimates of effect
(pp. 14-50). Baltimore: Johns Hopkins University Press. size. In H. Cooper & L. V. Hedges (Eds.), The handbook of research
Hedges, L. V., & Olkin, I. (1985). Statistical methodsfor meta-anal- synthesis (pp. 261-281). New York: Russell Sage Foundation.
ysis. New York: Academic Press. Sidanius, J., & Pratto, F. (1999). Social dominance. Cambridge, Eng-
Heine, S. J., & Lehman, D. R. (1997). Culture, dissonance, and self-affir- land: Cambridge University Press.
mation. Personality and Social Psychology Bulletin, 23, 389-400. Stewart, A. J., & Healy, J. M. (1989). Linking individual develop-
Jackson, L. A., & Gerard, D. A. (1996). Diuranal types, the "Big ment and social changes. American Psychologist, 44, 30-42.
Five" personality factors, and other personal characteristics. Suh, E., Diener, E., Oishi, S., & Triandis, H. C. (1998). The shifting basis
Journal of Social Behavior and Personality, 11, 273-283. of life satisfaction judgments across cultures: Emotions versus
James, W. (1950). The principles of psychology. New York: Dover. norms. Journal ofPersonality and Social Psychology, 74,482-493.
(Original work published 1890) Symonds, M. (1968). Disadvantaged children growing in a climate
Kling, K. C., Hyde, J. S., Showers, C. J., & Buswell, B. N. (1999). of hopelessness and despair. American Journal of Psychoanaly-
Gender differences in self-esteem: A meta-analysis. Psycholog- sis, 28, 15-24.
ical Bulletin, 125, 470-500. Thornton, A., Alwin, D. F., & Camburn, D. (1983). Causes and con-
Kohr, R. L., Coldiron, R., Skiffington, E. W., Masters, J. R., & Blust, sequences of sex-role attitudes and attitude change. American
R. S. (1988). The influence of race, class, and gender on self-es- Sociological Review, 48, 211-227.
teem for fifth, eighth, and eleventh grade students in Pennsylva- Trowbridge, N. (1972). Self-concept and SES in elementary school
nia schools. Journal of Negro Education, 57, 467-481. children. American Educational Research Journal, 9, 525-537.
Leary, M. R., Tambor, E. S., Terdal, S. K., & Downs, D. L. (1995). Twenge, J. M. (1997). Attitudes toward women, 1970-1995: A
Self-esteem as an interpersonal monitor: The sociometerhypoth- meta-analysis. Psychology of Women Quarterly, 21, 35-51.
esis. JournalofPersonalityandSocialPsychology, 68,518-530. Twenge, J. M. (2000). The age of anxiety? Birth cohort change in
Markus, H. R., & Kitayma, S. (1991). Culture and the self: Implica- anxiety and neuroticism, 1952-1993. Journal of Personality
tions for cognition, emotion, and motivation. Psychological Re- and Social Psychology, 79, 1007-1021.
view, 98, 224-253. Twenge, J. M. (2001). Changes in women's assertiveness in response to
Markus, H. R., Kitayama, S., & Heiman, R. J. (1996). Culture and status and roles: A cross-temporal meta-analysis, 1931-1993.
"basic" psychological principles. In E. T. Higgins, & A. W. Journal of Personality and Social Psychology, 81, 133-145.
Kruglanski (Eds.) Social psychology: Handbook of basic prin- Twenge, J. M. (2002). Birth cohort, social change, and personality:
ciples, pp. 857-913. New York: Guilford. The interplay of dysphoria and individualism in the 20th cen-
Markus, H. R., Kitayama, S., & VandenBos, G. R. (1996). The mu- tury. In D. Cervone & W. Mischel (Eds.), Advances in Personal-
tual interactions of culture and emotion. Psychiatric Services, ity Science (pp. 196-218). New York: Guilford.
47,225-226. Twenge, J. M., & Campbell, W. K. (2001). Age and birth cohort dif-
Mead, G. H. (1934). Mind, self, and society. Chicago: University of ferences in self-esteem: A cross-temporal meta-analysis. Per-
Chicago Press. sonality and Social Psychology Review, 5, 735-748.
Mullis, A. K., Mullis, R .L., & Normandin, D. (1992). Cross-sec- Twenge, J. M., & Crocker, J. (in press). Race and self-esteem:
tional and longitudinal comparisons of adolescent self-esteem. Meta-analyses comparing Whites, Blacks, Hispanics, Asians,
Adolescence, 27, 51-61. and American Indians, including a commentary on Gray-Little
Nesselroade, J. R., & Baltes, P. B. (1974). Adolescent personality devel- and Hafdahl. 2000 Psychological Bulletin.
opment and historical change, 1970-1972. Monographs of the Soci- University Microfilms International (1800-1998). Dissertation Ab-
etyfor Research in Child Development, 39 (1, Serial No. 154). stracts International [Online]. Ann Arbor, MI: Proquest.
Nisbett, R. E., & Cohen, D. (1996). Culture of honor: The psychol- U. S. Bureau of the Census. (1998). Statistical abstract of the United
ogy of violence in the South. Boulder, CO: Westview Press. States. Washington, DC: U.S. Government Printing Office.
Oliver, M. B., & Hyde, J. S. (1993). Gender differences in sexuality: Whang, P. A., & Hancock, G. R. (1994). Motivation and mathemat-
A meta-analysis. Psychological Bulletin, 114, 29-51. ics achievement: Comparisons between Asian-American and
Oyserman, D., Coon, H. M., & Kemmelmeier, M. (in press). Re- non-Asian students. Contemporary Educational Psychology,
thinking individualism and collectivism: Evaluation of theoreti- 19, 302-322.
cal assumptions and meta-analyses. Psychological Bulletin. Wills, T. A. (1981). Downward comparison principles in social psy-
Pelham, B. W. (1995). Self-investment and self-esteem: Evidence chology. Psychological Bulletin, 90, 245-271.
for a Jamesian model of self-worth. Journal of Personality and Wiltfang, G. L., & Scarbecz, M. (1990). Social class and adoles-
Social Psychology, 69, 1141-1150. cents' self-esteem: Another look. Social Psychology Quarterly,
Pelham, B. W., & Swann, W. B. (1989). From self-conceptions to 53, 174-183.
self-worth: On the sources and structure of global self-esteem. Wolf, F. M. (1986). Meta-analysis: Quantitative methods for re-
Journal of Personality and Social Psychology, 57, 672-680. search synthesis. Beverly Hills, CA: Sage.
Piers, E. V. (1969). Manualfor the Piers-Harris Children's Self-Con- Wylie, R. C. (1974). The self-concept. Lincoln: University of Ne-
cept Scale. Nashville, TN: Counselor Recordings & Tests. braska Press.
Porter, J. R., & Washington, R. E. (1979). Black identity and self-es- Zirkel, P. A., & Moses, E. G. (1971). Self-concept and ethnic group
teem: A few of studies of black self-concept, 1968-1978. An- membership among public school students. American Educa-
nual Review of Sociology, 5, 53-74. tional Research Journal, 2, 253-265.

71

S-ar putea să vă placă și