Personality and Individual Differences 54 (2013) 474–478
Contents lists available at SciVerse ScienceDirect
Personality and Individual Differences
journal homepage: www.elsevier.com/locate/paid
The implicit self and the specificity-matching principle: Implicit self-concept
predicts domain-specific outcomes
Melissa A McWilliams, Jason A. Nier ⇑, Jefferson A Singer
Department of Psychology, Box 5305, Connecticut College, 270 Mohegan Avenue, New London, CT 06320, United States
a r t i c l e
i n f o
Article history:
Received 1 May 2012
Received in revised form 21 September
2012
Accepted 21 September 2012
Available online 1 December 2012
Keywords:
Self-esteem
Self-concept
Implicit Association Test
Specificity matching
a b s t r a c t
According to the specificity-matching principle (Swann, Chang-Schneider, & McClarty, 2007), specific
aspects of self-concept should predict domain specific outcomes, rather than broader outcomes. The purpose of the current study was to determine whether this principle, which has thus far been examined
using explicit measures of the self, extends to the implicit self-concept. We tested this idea in the domain
of math achievement. We observed that explicit math self-concept was correlated with specific outcomes
(measures of math achievement), whereas explicit self-esteem was correlated with a broad outcome
(satisfaction with life). Thus, we replicated the specificity-matching principle using explicit measures of
self-esteem and self-concept. Moreover, we found that implicit self-concept was correlated with
domain-specific outcomes, but not a global outcome, as the specificity-matching principle would predict.
Furthermore, regression analyses indicated that implicit self-concept accounted for unique variance in the
domain-specific outcomes, for which the other measures of the self could not account. Taken together, we
conclude that the specificity-matching principle does indeed extend to the implicit self-concept.
Ó 2012 Elsevier Ltd. All rights reserved.
1. Introduction
For decades, the self has been an important topic of study in
psychology and constructs such as self-esteem and self-concept
have garnered significant attention within the field. Despite the
prominence of the self in the personality and social psychology literature, there have been significant conceptual and empirical debates over constructs such as self-esteem and self-concept. For
example, a number of researchers continue to harbor doubts
regarding the significance of self-esteem (Baumeister et al., 2005;
Dawes, 1996). Critics point to a variety of problems with the selfesteem construct, but perhaps the most frequent criticism of selfesteem has been its weak relationship with important outcomes,
such as academic achievement. This has led some researchers to
conclude that beliefs about the self may have little importance in
influencing performance and success (Baumeister, Campbell,
Krueger, & Vohs, 2003), and that the costs of pursuing high selfesteem may often outweigh its benefits (Crocker & Park, 2004).
In response to these critics, Swann, Chang-Schneider, &
McClarty, 2007 conducted a broad literature review, which asked
‘‘Do people self-view matter?’’ They concluded that self-views do
indeed matter, and argued that one of the major reasons for the
weak empirical relationship between self-esteem and meaningful
⇑ Corresponding author. Tel.: +1 860 439 5057; fax: +1 860 439 5300.
E-mail address: [email protected] (J.A. Nier).
0191-8869/$ - see front matter Ó 2012 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.paid.2012.09.014
outcomes was, at least partially, a result of a methodological problem. In particular, Swann et al. (2007) proposed the specificitymatching principle, which holds that a specific self-concept, such
as self-perceptions associated with math, should be used to predict
specific outcomes, such as math achievement. Conversely, global
constructs, such as self-esteem, should not be powerful predictors
of narrow outcomes, and should instead be more strongly linked to
broader outcomes, such as overall well-being. In short, the level of
specificity of the predictor should match the level of specificity of
the outcome being predicted. Indeed, when Swann et al. examined
the recent self literature, they concluded that many researchers
violated the specificity-matching principle as they focused ‘‘on
the capacity of global measures of self-esteem to predict specific
outcomes’’ (2007, pp. 87).
The difficulty of accurately measuring self-esteem may also
help explain why self-esteem is often a weak predictor of significant outcomes. For example, Buhrmester, Blanton, and Swann
(2011) outlined several shortcomings of traditional measures of
self-esteem. They reasoned, as have others (Greenwald & Farnham,
2000) that traditional self-report measures of self-esteem are unlikely to assess self-views of which one may not be aware, and that
self-presentation concerns may cause individuals to inflate their
self-reported self-evaluations. One way to possibly circumvent
these problems associated with traditional measures of selfesteem and self-concept is to rely upon implicit measures of
these constructs. These measures may tap into the automatic or
M.A McWilliams et al. / Personality and Individual Differences 54 (2013) 474–478
unconscious aspects of self-knowledge that people may be unable
or unwilling to report, and which explicit measures may not
accurately assess (Devos, Huynh, & Banaji, 2012). Yet despite the
potential advantages afforded by implicit measures of the self,
researchers have yet to fully explore their significance. Moreover,
a recent review by Buhrmester et al. (2011) suggests that the implicit self-esteem construct, in addition to suffering from conceptual problems (e.g., it does not seem to measure beliefs that arise
out of self-reflection), has weak reliability, and – like explicit
self-esteem – often fails to predict important outcomes. Thus, the
promise of implicit measures of the self has not yet been borne
out by strong empirical support.
1.1. Present study
In the present study, we examined whether the specificitymatching principle, which thus far has been observed in research
employing traditional measures of self-esteem and self-concept
(Swann et al., 2007), extends to the implicit self-concept. Given
the logic of the specificity-matching principle, we propose that implicit self-concept may have significant utility in predicting specific
outcomes. In the same way that explicit self-concept more strongly
predicts specific outcomes than explicit self-esteem, we hypothesized that implicit self-concept may more strongly predict specific
outcomes than implicit self-esteem? In particular, we examined
two questions about implicit self-concept that have yet to be
simultaneously tested. First, are the relationships between implicit
self-concept and outcome variables consistent with the specificitymatching principle? Second, might implicit self-concept account
for unique variability in domain-specific outcomes, for which explicit self-concept, as well as implicit and explicit self-esteem, cannot account? We tested these ideas in the domain of mathematical
achievement. We administered implicit and explicit measures of
mathematics self-concept, as well as implicit and explicit measures
of self-esteem. We then used these variables as predictors of mathematical achievement. We also included a measure of global wellbeing – the Satisfaction with Life Scale (Diener, Emmons, Larsen, &
Griffin, 1985).
Consistent with the specificity-matching principle, we predicted that our two global measures of the self (implicit and explicit self-esteem) would more strongly predict a global outcome
(satisfaction with life), relative to a specific outcome (math
achievement). Conversely, we predicted that the measures of specific aspects of the self (implicit and explicit measures of math selfconcept) would more strongly predict achievement in this specific
domain, relative to the degree to which they predicted a global
outcome. Furthermore, we also predicted that the measure of the
implicit self-concept would be able to account for unique
variance in the domain-specific outcomes, even after accounting
for implicit self-esteem and explicit beliefs about the self. Taken together, this pattern of results would provide support for the idea
that the specificity-matching hypothesis extends to the implicit
self-concept.
2. Method
2.1. Participants
Participants were160 college men (N = 40) and women
(N = 120) enrolled in psychology courses at a college in the Eastern
United States. The participants ranged in age from 18 to 22 years
and were primarily white (83%). Fifteen participants were excluded
for failing to complete the dependent measures, which resulted in
a final sample of 145 participants.
475
2.2. Measures
2.2.1. Explicit measures of the self
The Rosenberg Self-Esteem Scale (RSES; Rosenberg, 1989) was
used to assess explicit self-esteem. Participants indicated the degree to which they agreed with 10 self-descriptive statements
(e.g. ‘‘I feel that I have a number of good qualities’’) on a 0–3 scale.
Scores could range from 0 to 30, with higher scores indicating
greater explicit self-esteem (a = .79). To assess explicit math selfconcept, participants were asked to indicate the extent to which
they agreed with four items that indicated that math was a significant aspect of their self-concept (e.g., ‘‘I care about my mathematical abilities’’). The responses, indicated on a 7-point Likert scale,
were averaged together with higher scores indicating a greater explicit math self-concept (a = .81).
2.2.2. Implicit measures of the self
In order to measure implicit self-esteem, we employed the Implicit Association Test (IAT). Similar to Greenwald and Farnham
(2000), participants categorized words (self and non-self words,
and positive and negative words) into two categories (self versus
other and positive versus negative) using two response keys. Response latencies were analyzed using a procedure recommended
by Greenwald, Nosek, and Banaji (2003). The resulting score –
the D measure – reflected the degree to which participants had
more strong associations for the self with positive concepts, relative to negative concepts. Higher scores were indicative of more
positive implicit self-esteem.
To assess implicit math self-concept, the IAT was adapted
(again, similar to Greenwald & Farnham, 2000) to measure the extent to which participants associated the self with mathematical
concepts (e.g., ‘‘equation’’), relative to non-mathematical concepts
(e.g., ‘‘music’’). Participants had to categorize words (self and nonself words, and mathematical and non-mathematical words), into
categories (self versus other, mathematical skills versus other
skills) using two response keys. As with implicit self-esteem, the
D measure was then calculated, which reflected the degree to
which participants had stronger associations between the self
and mathematical concepts, relative to non-mathematical concepts. Higher scores were indicative of greater implicit math selfconcept.
2.2.3. Outcomes
The Satisfaction with Life Scale (SWLS; Diener et al., 1985) was
our measure of global life satisfaction. The SWLS contains five selfdescriptive statements (e.g. ‘‘In most ways my life is close to my
ideal’’) and participants indicated their agreement with each item
on a 7-point Likert scale. Scores ranged from 5 to 35, with higher
scores indicating greater life satisfaction (a = .80).
We used three measures of math achievement. The first measure, Math Engagement, was assessed by asking participants to
indicate which particular math courses they had completed, from
a list of all math courses offered at the institution. The number
of courses completed was summed together as a measure of Math
Engagement, such that higher scores indicated more courses taken
(i.e. greater Math Engagement).
Math Performance was assessed by asking participants to list
the letter grade they had received in each course they had taken.
The letter grades were converted to numerical quality point scores.
These quality points were weighted; courses that fulfilled requirements toward the mathematics major were given higher quality
point scores, relative to courses that did not fulfill any requirement
toward the mathematics major. Specifically, letter grades in
courses designed for non-mathematics majors were converted to
a 4.0–0.0 quality point scale. (‘‘A’’ equivalent to a score of 4.0, ‘‘B’’
equivalent to 3.0, etc). Each course that did fulfill requirements
476
M.A McWilliams et al. / Personality and Individual Differences 54 (2013) 474–478
towards the mathematics major was weighted (see Sadler & Tai,
2007) on a 5.0–0.0 quality point scale (‘‘A’’ = 5.0, ‘‘B’’ = 4.0,
‘‘C’’ = 3.0, ‘‘D’’ = 1.0, ‘‘F’’ = 0.0). These quality points scores were
then averaged together, with higher scores indicating better grades
in the math courses completed (i.e. better Math Performance). Participants who did not list any grades in any math course received a
score of zero.
The final measure, Overall Math Achievement, was a composite
of the previous two measures. For this measure, we weighted the
Math Performance scores by the number of Math courses completed. For example, a student who had a Math Performance score
of 3.5 and completed two courses would receive a score of 7.0. The
resulting score, commonly referred to as Total Quality Points, reflected both the number of math courses taken and the grades received in those courses.
2.3. Procedure
After indicating their consent and completing a demographics
questionnaire, participants completed the global self-report measures (the RSES and the SWLS). Participants then completed the
implicit self-esteem task. They then indicated their responses on
the self-report measure of explicit math self-concept, followed by
the implicit math self-concept task. Participants then completed
a questionnaire asking about their math grades.
3. Results
3.1. Was the pattern of intercorrelations supportive of the specificitymatching principle?
We first examined whether the pattern of intercorrelations was
consistent with the specificity-matching principle (see Table 1).
Recall that it was predicted that the global measures of the self
(implicit and explicit self-esteem) would be correlated with our
broad outcome (the SWLS), while the measures of specific selfconcept (implicit and explicit math self-concept), would be associated with achievement in math. As expected, explicit self-esteem
was correlated with SWLS (r = .27, p < .001), but was not significantly correlated with any of the math outcomes (r’s < .06,
p’s = n.s.). Next we tested whether these correlations were significantly different from one another, as the specificity-matching principle would predict. Given that the hypotheses derived from the
specificity-matching principle were all directional in nature, we
evaluated the significance of this hypothesis, and all subsequent
statistical procedures that tested the specificity-matching principle, with one-tailed tests. As predicted, the correlation between
explicit self-esteem and the SWLS was significantly stronger
than the correlations between explicit self-esteem and math
achievement, p’s < .05 (Meng, Rosenthal, & Rubin, 1992). Also as
hypothesized, implicit self-esteem was not significantly correlated
with any of math outcomes (r’s < 10, p’s = n.s.). However, contrary
to expectations, implicit self-esteem was not significantly correlated with the global measure of well-being, the SWLS (r = .07,
p = n.s.).
Next, the correlations between math self-concept and the outcome variables were examined to determine if they were consistent with the specificity-matching principle. Explicit math selfconcept was significantly correlated with Math Engagement
(r = .35), Math Performance (r = .19), and Overall Math Achievement (r = .30), p’s < .05. Moreover, explicit math self-concept was
not significantly correlated with the SWLS, (r = .02, p = n.s.), and,
as predicted, this correlation was significantly weaker than the correlations between the three math outcomes and explicit math selfconcept, p’s 6 .05. Thus these results indicated that we replicated
the specificity-matching principle using a measure of explicit math
self-concept.
For our final correlational analyses, we examined whether the
pattern of intercorrelations was consistent with the idea that the
specificity-matching principle extends to the implicit self-concept.We observed that implicit math self-concept was significantly
correlated with Math Engagement, (r = .27), Math Performance
(r = .20), and Overall Math Achievement(r = .29), p’s 6 .01. Implicit
math self-concept was not correlated with the SWLS (r = .02,
p = n.s.), and this correlation was significantly weaker than the
three correlations between each of the math outcomes and implicit
self-concept, p’s < .05. Taken together, the pattern of intercorrelations among the self variables and the outcomes largely supported
the predictions derived from the specificity-matching principle.
Though implicit self-esteem was not associated with responses
on the SWLS, all remaining correlations were consistent with
expectations.
3.2. Did implicit self-concept account for unique variance in domainspecific outcomes?
In the final group of analyses, we conducted a series of multiple
regressions to determine whether implicit self-concept could account for unique variance in domain-specific outcomes. In each
Table 2
Multiple regression analysis predicted satisfaction with life (SWLS) from self
variables.
Predictor Variable
b
t
p
Self-esteem
Explicit self-esteem
Implicit self-esteem
.28
.04
3.33
0.51
.001*
n.s.
Self-concept
Explicit math self-concept
Implicit math self-concept
.08
.07
0.88
0.83
n.s.
n.s.
r2 = .08.
p < .05.
*
Table 1
Descriptive statistics and intercorrelations.
*
**
Measure
M
SD
1
2
3
4
5
6
7
8
1.
2.
3.
4.
5.
6.
7.
8.
.46
21.3
.02
4.23
25.1
.75
2.01
2.62
.40
6.6
.53
1.3
5.3
.83
1.89
3.35
1.0
.09
.19*
.20*
.07
.05
.03
.08
1.0
.12
.10
.27**
.03
.05
.04
1.0
.33**
.02
.27**
.20*
.29**
1.0
.02
.35**
.19*
.30**
1.0
.09
.03
.07
1.0
.71**
.77**
1.0
.93**
1.0
Implicit self-esteem
Explicit self-esteem
Implicit math self-concept
Explicit math self-concept
Satisfaction with life
Math Engagement
Math Performance
Overall Math Achievement
p < .05.
p < .001.
477
M.A McWilliams et al. / Personality and Individual Differences 54 (2013) 474–478
Table 3
Multiple regression analyses predicting math outcomes from self variables.
Predictor variable
Self-esteem
Explicit self-esteem
Implicit self-esteem
Math self-concept
Explicit self-concept
Implicit self-concept
Model 1: Predicting Math Engagement
Model 2: Predicting Math Performance
B
p
b
t
p
n.s.
n.s.
.05
.02
0.59
0.27
n.s.
n.s.
<.01*
.01*
.14*
.15*
1.62
1.70
.05*
.04*
t
.03
.05
.29*
.19*
0.33
0.62
3.46
2.19
Model 3: Predicting Overall Math Achievement
t
b
.01
.03
.23*
.22*
p
0.11
0.36
2.65
2.58
n.s.
n.s.
<.01*
<.01*
r2 = .13 for Model 1; r2 = .06 for Model 2; r2 = .15 for Model 3.
*
p < .05.
of these regressions we used measures of the implicit and explicit
self to predict the outcome variables. We again used one-tailed significance tests, since the hypotheses derived from the specificitymatching principle were all directional in nature. In the first
regression, the SWLS was used as the criterion variable. The results
of this analysis (see Table 2), indicated that explicit self-esteem
was the only significant predictor of responses on the SWLS
(b = .28, p = .001). Consistent with the specificity-matching principle, neither of the self-concept variables predicted responses on
the SWLS. Similar to the results of the correlational analyses, implicit self-esteem was not a significant predictor of the SWLS.
In the remaining multiple regression analyses, we used each of
the measures of math achievement (i.e. Math Engagement, Math
Performance, and Overall Math Achievement) as a criterion variable (see Table 3). For Math Engagement, neither of the self-esteem
variables were significant predictors. However, explicit selfconcept was a significant predictor of Math Engagement (b = .29,
p = .001), and as predicted, implicit self-concept was also a significant predictor of Math Engagement (b = .19, p = .01). We obtained
similar results for Math Performance; neither implicit nor explicit
self-esteem were significant predictors of Math Performance.
However, both implicit self-concept (b = .15, p = .04) and explicit
self-concept (b = .14, p = .05) were significant predictors. The same
pattern of results was again observed for the composite measure,
Overall Math Achievement; only the self-concept variables were
reliable predictors of Overall Math Achievement. Specifically, implicit self-concept significantly predicted Overall Math Achievement (b = .22, p < .01) as did explicit self-concept (b = .23, p < .01)
Thus, the results of the regression analyses were largely consistent with expectations. Though implicit self-esteem was not a significant predictor of responses on the SWLS, the remaining
predictors were consistent with the specificity-matching principle.
Moreover, implicit self-concept accounted for unique variance in
all three measures of math achievement, for which the remaining
self variables could not account.
4. Conclusion
The purpose of the current study was to determine whether the
specificity-matching principle, which holds that specific self-concept should predict domain-specific outcomes, extends to the implicit self-concept. Correlational analyses indicated that we
replicated the specificity-matching principle using explicit measures of self-esteem and self-concept, such that global self-esteem
was significantly correlated with a global outcome (satisfaction
with life), while our measure of explicit math self-concept was significantly correlated with all three measures of math achievement.
Moreover, we found that implicit math self-concept was also correlated with each math achievement variable, but not satisfaction
with life, as the specificity-matching principle would predict. Furthermore, regression analyses indicated that implicit self-concept
accounted for unique variance in each of the math outcomes, for
which the other measures of the self could not account. Taken together, we conclude that the specificity-matching principle does
indeed extend to the implicit self-concept.
Although the predictions regarding implicit self-concept were
fully supported, the role of implicit self-esteem was less clear. In
particular, we observed that implicit self-esteem was not correlated with either global or specific outcomes. Thus implicit selfesteem appeared to lack predictive validity. Though this result
was not predicted and does not seem to be consistent with the
specificity-matching principle, similar findings have been observed
in previous research. For example, as pointed out in Buhrmester
et al.’s (2011) review of implicit self-esteem, numerous studies
have also found that implicit self-esteem often fails to predict
important outcomes. Given results such as these, they concluded
that ‘‘implicit self-esteem measures are at best weakly correlated
with typical self-esteem covariates, with many correlations hovering around zero’’ (Buhrmester et al., 2011, pp. 369–370). Our findings echo this conclusion and raise additional concerns about the
meaning and utility of the implicit self-esteem construct.
Despite the lack of support for the utility of global implicit selfesteem, our findings nonetheless strongly suggest that it is premature to abandon implicit measures of the self altogether. Although
it may be true that global self-esteem, by its very nature (i.e. traditionally conceptualized as a self-reflective judgment about
whether we think of ourselves positively or negatively), is a poor
candidate for implicit measurement, other implicit self-views
may still be effectively measured and predict important outcomes.
For example, specific self-esteem, defined as one’s self-evaluation
in a particular domain (Rosenberg, Schooler, Schoenbach, &
Rosenberg, 1995), may prove to be an important implicit construct
that may more strongly predict specific outcomes than does global
implicit self-esteem.
In any event, the results of the present study suggest that implicit self-concept is an important construct. Moreover, if implicit
self-concepts in various academic domains are indeed predictive
of one’s outcomes in those domains, independent of overall selfworth, this could influence how educators approach these domains
with students. Emphasizing evaluative outcomes, such as grades,
may be less effective than encouraging student engagement in particular academic domains, since consistent engagement with a particular topic may spur changes in implicit self-concepts. By placing
too much emphasis on students’ feelings of global self-worth in
relation to achievement, we may in fact be losing opportunities
for individuals to cultivate their more specific implicit affinities
for academic achievement in particular domains.
References
Baumeister, R. F., Campbell, J. D., Krueger, J. I., & Vohs, K. D. (2003). Does high selfesteem cause better performance, interpersonal success, happiness, or healthier
lifestyles? Psychological Science in the Public Interest, 4, 1–44.
Baumeister, R. F., Campbell, J. D., Krueger, J. I., & Vohs, K. D. (2005). Exploding the
self-esteem myth. Scientific American, 292, 84–91.
478
M.A McWilliams et al. / Personality and Individual Differences 54 (2013) 474–478
Buhrmester, M. D., Blanton, H., & Swann, W. (2011). Implicit self-esteem: Nature,
measurement, and a new way forward. Journal of Personality and Social
Psychology, 100, 365–385.
Crocker, J., & Park, L. E. (2004). The costly pursuit of self-esteem. Psychological
Bulletin, 130, 392–414.
Dawes, R. (1996). House of cards: Psychology and psychotherapy built on myth. New
York: Free Press.
Devos, T., Huynh, Q., & Banaji, M. R. (2012). Implicit self and identity. In M. R. Leary
& J. P. Tangney (Eds.), Handbook of self and identity (pp. 159–179). New York:
Guilford Press.
Diener, E., Emmons, R. A., Larsen, R. J., & Griffin, S. (1985). The satisfaction with life
scale. Journal of Personality Assessment, 49, 71–75.
Greenwald, A. G., & Farnham, S. D. (2000). Using the implicit association test to
measure self-esteem and self-concept. Journal of Personality and Social
Psychology, 79, 1022–1038.
Greenwald, A. G., Nosek, B. A., & Banaji, M. R. (2003). Understanding and using the
implicit association test: I. An improved scoring algorithm. Journal of Personality
and Social Psychology, 85, 197–216.
Meng, X., Rosenthal, R., & Rubin, D. B. (1992). Comparing correlated correlation
coefficients. Psychological Bulletin, 111, 172–175.
Rosenberg, M. (1989). Society and the adolescent self-image. Middletown, CT:
Wesleyan University Press.
Rosenberg, M., Schooler, C., Schoenbach, C., & Rosenberg, F. (1995). Global selfesteem and specific self-esteem: Different concepts, different outcomes.
American Sociological Review, 60, 141–156.
Sadler, P. M., & Tai, R. H. (2007). Weighting for recognition: Accounting for advanced
placement and honors courses when calculating high school grade point
average. NASSP Bulletin, 91, 5–32.
Swann, W., Chang-Schneider, C., & McClarty, K. (2007). Do people’s self-views matter?
Self-concept and self-esteem in everyday life. American Psychologist, 62, 84–94.