Journal of Counseling Psychology Copyright 1995 by the American Psychological Association, Inc.

1995, Vol. 42, No. 2, 190-198 0022-0167/95/S3.00

Mathematics Self-Efficacy and Mathematics Performances: The Need for

Specificity of Assessment

Frank Pajares M. David Miller

Emory University University of Florida

In this study, 391 students were asked to provide 3 types of mathematics self-efficacy
judgments: confidence to solve mathematics problems, confidence to succeed in math-related
courses, and confidence to perform math-related tasks. Criterial tasks were solution of math
problems and choice of math-related majors. As hypothesized, students' reported confidence
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

to solve the problems they were later asked to solve was a more powerful predictor of that
This document is copyrighted by the American Psychological Association or one of its allied publishers.

performance than was either their confidence to perform math-related tasks or to succeed in
math-related courses. Similarly, confidence to succeed in math-related courses was a stronger
predictor of choice of math-related majors than was either confidence to solve problems or
to perform math-related tasks. Results support A. Bandura's (1986) contention that, because
judgments of self-efficacy are task specific, measures of self-efficacy should be tailored
to the criterial task being assessed and the domain of functioning being analyzed to
increase prediction.

Social cognitive theorists contend that self-efficacy be- ported the predictive role of self-efficacy (Maddux, Norton,
liefs, or "people's judgments of their capabilities to organize & Stoltenberg, 1986), many studies are plagued by assess-
and execute courses of action required to attain designated ments of self-efficacy that do not conform to Bandura's
types of performances" (Bandura, 1986, p. 391), strongly guidelines regarding specificity of measurement (Multon et
influence the choices people make, the effort they expend, al., 1991). This mismatch between self-efficacy and criterial
the strength of their perseverance in the face of adversity, task assessment is a recurring theme in educational research,
and the degree of anxiety they experience. In part, these often producing confounded relationships and ambiguous
self-perceptions can be better predictors of behavior than findings.
actual capability because such self-beliefs are instrumental When the efficacy beliefs assessed do not reflect with any
in determining what individuals do with the knowledge and sense of specificity the criterial task with which they are
skills they have. The mediational role these beliefs play also compared, their predictive value is diminished. Multon et al.
helps explain why people's performance attainments may (1991) found only 68 published and unpublished papers
differ even when they have similar knowledge and skills. written between 1977 and 1988 on the relationship between
Bandura (1986) cautioned that because judgments of self- self-efficacy and academic performance or persistence. Of
efficacy are task and domain specific, "ill-defined global these, only 36 studies with 38 samples, encompassing 4,998
measures of perceived self-efficacy or defective assess- participants, assessed academic performance and met their
ments of performance will yield discordances" (p. 397). For criteria for inclusion in a meta-analysis, which were that the
this reason, measures of self-efficacy should be specifically study contain a measure of self-efficacy and academic per-
tailored to the criterial task being assessed and the domain formance and provide sufficient information to calculate
of functioning being analyzed. This caution has also been effect size estimates. They computed that efficacy beliefs
voiced by other social cognitive researchers (e.g., Lent & were related to performance (r u = .38) and accounted for
Hackett, 1987; Multon, Brown, & Lent, 1991) but has often approximately 14% of the variance in students' academic
gone unheeded in educational research, where assessments performance. However, effect sizes depended on specific
of self-efficacy frequently bear little resemblance to the characteristics of the studies, notably on the types of effi-
criterial task with which they are compared but, instead, cacy and performance measures used.
reflect generalized, or sometimes even unrelated, attitudes Multon et al. (1991) discovered that the strongest effects
about capabilities. Although findings have generally sup- were obtained by researchers who compared specific effi-
cacy judgments with basic skills measures of performances,
who developed highly concordant self-efficacy and perfor-
Frank Pajares, Department of Educational Studies, Emory Uni-
mance indices, and who administered them at the same
versity; M. David Miller, Department of Foundations, University
of Florida. time. Recall that Bandura (1986) cautioned that self-effi-
Correspondence concerning this article should be addressed to cacy judgments be specifically rather than globally as-
Frank Pajares, Department of Educational Studies, Emory Univer- sessed, correspond directly to the criterial task, and be
sity, Atlanta, Georgia 30322. Electronic mail may be sent via measured as closely as possible in time to that task. Signif-
Internet to [email protected]. icant relationships are obtained even with generalized self-
190
 MATHEMATICS SELF-EFFICACY 191

efficacy indices, a phenomenon that Multon et al. described specification of the domain of math-related behavior" (p.
as reinforcing "the theoretical and practical value of attend- 331) and that these domains included solution of math
ing to students' self-efficacy beliefs" (p. 35) but that also problems, math behaviors used in daily life, and, of partic-
may produce confounded and misleading results. In fact, if ular relevance to college students, satisfactory performance
global and generalized self-efficacy assessments can predict in a college course. They operationalized mathematics self-
performances that are not specifically related, the relation- efficacy as the composite score of the three subscales.
ship between properly assessed self-efficacy and perfor- Since its creation, the MSES has been used in studies
mance should certainly increase. focusing both on academic performance and on major and
career choice. For example, Hackett (1985) used the MSES
in a path analysis to discover the mediational role of self-
efficacy in predicting undergraduate students' choice of
Overview of Mathematics Self-Efficacy Findings
math-related majors. With the composite score of the three
Self-efficacy research in academic settings has focused MSES subscales as the self-efficacy assessment, Hackett
primarily on two major areas. The first has investigated the reported a correlation of .50 between perceived efficacy and
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

relationships among efficacy beliefs, related psychological choice of major, as well as a moderate direct effect of
This document is copyrighted by the American Psychological Association or one of its allied publishers.

constructs, and academic motivation and achievement (see self-efficacy on choice (.24). Hackett and Betz (1989) later
Schunk, 1989, 1991). The second has attempted to explore administered the MSES to undergraduates and reported that
the link between efficacy beliefs and college major and a hierarchical regression analysis demonstrated that the full-
career choices (see Lent & Hackett, 1987). Studies in this scale MSES score was strongly predictive of a choice of
second area have important implications for counseling and college major between science and nonscience fields (jB =
vocational psychology theory and practice, given that find- .45).
ings have provided insights into the career development of Hackett and Betz (1989) also investigated the relationship
young men and women and can be used to develop appro- between math self-efficacy and mathematical problem solv-
priate career intervention strategies (Lent & Hackett, 1987; ing. They obtained a correlation of .44 between undergrad-
Multon et al., 1991). uates' full-scale MSES scores and performance scores on an
Early studies investigated confidence in learning mathe- alternate-forms test of the MSES Problems scale—a corre-
matics, a conceptual forerunner to math self-efficacy that lation the researchers noted was not as large as they had
has consistently been found to predict math-related behavior expected. In contrast, Pajares and Miller (1994) obtained a
and performance (Reyes, 1984). This confidence was glo- correlation of .70 when they compared scores on one of the
bally assessed by asking students general questions about forms of Dowling's (1978) MCS with performance scores
their perceived mathematics capabilities. Mathematics self- on the same test on which self-efficacy was assessed. Pa-
efficacy has more recently been assessed in terms of indi- jares and Miller also found that—with math self-concept,
viduals' judgments of their capabilities to solve specific perceived usefulness of mathematics, math background, and
math problems, to perform math-related tasks, and to suc- gender as part of a path analysis model—self-efficacy had
ceed in math-related courses (Betz & Hackett, 1983). the strongest direct effect on problem-solving performance
Dowling (1978) was the first researcher to create a con- 0 = .545).
fidence measure to specifically correspond with a perfor- Other researchers have used full-scale MSES scores as a
mance assessment in which students were asked to solve the generalized math self-efficacy assessment. Randhawa,
same or similar math problems on which their confidence Beamer, and Lundberg (1993) adapted the MSES for use
was based. Drawing from math problems created for the with high school students and used LISREL procedures to
National Longitudinal Study of Mathematical Abilities find that the full-scale, generalized self-efficacy score me-
(NLSMA), she developed the Mathematics Confidence diated the effect of various math attitudes on math problem
Scale (MCS). Students were first asked to provide judg- solving. The criterial task, the solving of math problems,
ments of confidence to solve these math problems and later was conceptually related only to the Problems subscale.
were asked to solve an alternate-forms test of the problems Moreover, the problems on the performance measure dif-
on which their confidence was assessed. Dowling obtained fered from those presented on the self-efficacy assessment.
a correlation of .54 between self-efficacy and performance, Consequently, although the math attitude measures had a
a significantly higher figure than that reported by most strong direct effect on self-efficacy (.64), they also had a
previous researchers assessing math confidence globally stronger direct effect on performance (.44) than did gener-
and that is consistent with the results of Multon et al. alized self-efficacy (.32).
(1991). Studies that report a lack of relationship between math
Subsequently, Betz and Hackett (1983) created the Math- self-efficacy and performance often suffer from similar con-
ematics Self-Efficacy Scale (MSES). They incorporated as ceptual or measurement flaws. Benson (1989) found that the
one subscale a measure similar to Dowling's (1978) MCS path from what she called math self-efficacy to performance
and added two additional subscales, one to assess students' was not significant, whereas that between math self-concept
confidence to perform certain math-related tasks and an- and performance was. However, self-efficacy was assessed
other to assess their confidence to earn an A or B in certain with three global items dealing with expected success in a
math-related courses. Betz and Hackett created these sub- statistics class (e.g., "No matter how hard I study, I will not
scales in the belief that math self-efficacy required "detailed do well in this class"). Self-concept was assessed with seven
 192 FRANK PAJARES AND M. DAVID MILLER

items specific to feelings of math self-worth (e.g., "I feel Purpose of the Study
insecure in a math class"). Performance was the midterm
exam grade in a statistics course. Benson concluded that Hackett and Betz (1989) defined mathematics self-effi-
"additional studies need to be conducted to verify the rela- cacy as "a situational or problem-specific assessment of an
tionship between math self-concept and self-efficacy and to individual's confidence in her or his ability to successfully
explore why self-efficacy did not influence [performance] perform or accomplish a particular [mathematics] task or
nor statistical test anxiety" (p. 259). Perhaps the answer is problem" (p. 262). In line with Bandura's (1986) guidelines
that a comparison of general confidence to succeed in a regarding consistency of self-efficacy and performance as-
class with a statistics midterm grade is not likely to produce sessment, we hypothesized that individuals' judgments to
the sort of correspondence that social cognitive theory solve math problems, to perform math-related tasks, or to
hypothesizes. succeed in math-related courses entail substantively differ-
Other studies are more problematic. Smith, Arnkoff, and ent judgments of mathematics capability. Although they all
are judgments about certain mathematical abilities, their
Wright (1990) used hierarchical multiple regression to test
predictive value regarding these abilities should largely
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

the predictive power of three theoretical models—cogni-

depend on the nature of the specific capabilities with which
This document is copyrighted by the American Psychological Association or one of its allied publishers.

tive-attentional, cognitive-skills, and social learning—on they are compared. Consequently, students' judgments to
academic performance. The researchers concluded that vari- solve math problems should be more strongly predictive of
ables within each model predicted performance to some their ability to solve those problems than should their con-
degree but that self-efficacy was a weak predictor. Self- fidence to perform other math-related tasks or to succeed in
efficacy, however, was operationalized as study skills or math-related courses. Similarly, their judgments to succeed
test-taking self-perceptions and measured with items such in math-related courses should be more strongly predictive
as "Rate how certain you are that you can study at a time of their choice to enroll in such courses than should their
and place where you won't get distracted." These self- confidence to solve specific problems or perform mathe-
perceptions were then compared with academic outcomes matics tasks.
such as course and exam grades and grade point average. In essence, our argument is that, although researchers
Cooper and Robinson (1991) found a low but significant have tended to use generalized measures or multiple-scale
correlation between math self-efficacy and performance, math self-efficacy scores to predict math-related outcomes,
but a regression model with math anxiety, the quantitative the appropriateness of the match between self-efficacy as-
score on the American College Test, and prior math expe- sessment and the outcome is crucial to optimal prediction of
rience revealed that self-efficacy did not account for a that outcome. Figure 1 illustrates the traditional method of
significant portion of the variance in math performance. using multiple-scale math self-efficacy assessments with
Here again, self-efficacy and performance assessments were dependent measures such as solving math problems or
mismatched—scores from the Courses subscale of the choosing math-related majors. However, the more parsimo-
MSES were compared with scores on a performance mea- nious model in Figure 2 more accurately reflects the role of
sure that consisted of solving problems from the Missouri the same self-efficacy assessments in predicting those
Mathematics Placement Test. outcomes.

Mathematics Problems
Self-Efficacy

Mathematics
Problem-solving
Performance

Math-Related Tasks
Self-Efficacy

Selection of
Math-Related
Major

Math-Related Courses
Self-Efficacy

Figure 1. Multiple mathematics self-efficacy assessments to predict mathematics problems and

selecting math-related majors.
 MATHEMATICS SELF-EFFICACY 193

Mathematics Problems
Self-Efficacy

Mathematics
Problem-solving
Performance

Math-Related Tasks ^ \
Self-Efficacy )

Selection of
Math-Related
Major
Math-Related Courses
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

Self-Efficacy
This document is copyrighted by the American Psychological Association or one of its allied publishers.

Figure 2. Mathematics self-efficacy assessment matched to related outcomes.

In this study, we asked students to provide judgments of math-related major is equally as reasonable, as this choice
their confidence to solve specific math problems, to perform provides a logical reflection of both interest and intention.
math-related tasks, and to succeed in math-related courses. In some ways, choice of major has greater practical utility to
We compared these judgments of confidence with two the counseling and career literature (hence, its use by Betz,
outcome measures: their ability to solve the problems on Hackett, and their colleagues). It is interesting that our
which their confidence was assessed and the math-related- results in this study are consistent with those of Lent at al.
ness of their academic majors. Social cognitive theory holds (1993).
that judgments of capability to solve specific problems
should be more predictive of capability to solve those prob-
lems. Similarly, judgments of confidence to succeed in Method
math-related courses should be more predictive of choice of
math-related majors that involve enrolling in such courses. Participants and Procedure
If this is the case, findings have important implications for
future assessment of self-efficacy both in the area of math- Participants for the study consisted of 391 undergraduates (247
women and 144 men; 26 freshmen, 74 sophomores, 172 juniors,
ematics and other academic areas.
and 119 seniors) enrolled in various upper division courses at three
We acknowledge from the outset that confidence to suc- large public universities, two in the South (ns = 212 and 92) and
ceed in mathematics courses does not relate as directly to one in the Southwest (n = 87). The students represented over 30
choice of major as confidence to solve math problems does majors throughout their respective universities. Testing procedures
to actual problem solving, and we expected the latter to were carried out in individual, upper division classes in colleges of
provide stronger results. Clearly, however, the choice of a education and of arts and sciences at the three universities. In all
math-related major is predicated on the need to succeed in cases, colleagues volunteered the use of their classes at our re-
certain mathematics courses. Consequently, there is a strong quest. Participation by the students was voluntary and no remu-
neration was provided, but most instructors provided course credit.
logical correspondence between judgments of confidence to
Appropriate informed consent was obtained, and students were
succeed in math courses and choice to pursue a major that assured anonymity. Students were first explained the procedures.
requires enrolling in such courses. This correspondence is Subsequently, they were asked to complete a descriptive question-
stronger than that between choice of major and either tasks naire and the MSES—Revised (MSES-R). Students were in-
or problems self-efficacy. One might argue that a better formed that they would be asked to solve the problems on which
criterial task could have been selected to compare with their problems self-efficacy was assessed. However, they were
judgments to succeed in math-related courses. The logical asked not to attempt to solve the problems as they provided
task would be a measure of success in such courses. How- confidence judgments, and these directions were also clearly
ever, most undergraduates do not take most of the courses printed on the efficacy instrument. The test administrator moni-
on the efficacy instrument—low-efficacy students, for ex- tored the session to ensure that students followed directions
carefully.
ample, avoid such courses. Lent, Lopez, and Bieschke
(1993) resolved this issue by comparing courses self-effi- After these instruments were collected, students were adminis-
tered a test that asked them to solve the problems on which their
cacy scores with three outcomes: intention to take the
self-efficacy had been assessed. Bandura (1986) suggested that
courses listed on the instrument, grades obtained in math- efficacy and performance be assessed within as close a time period
related courses that students took during the subsequent as possible and that efficacy assessment precede performance
term, and interest in the math courses listed on the instru- assessment. Therefore, all instruments were administered during
ment. All three seemed to us reasonable choices. Choice of one class period. This procedure also ensured that student absences
 194 FRANK PAJARES AND M. DAVID MILLER

would not create a situation whereby some students completed one Mathematics outcome measures. Self-efficacy is assessed by
measure but not another. Pilot testing had demonstrated that the asking individuals to provide a judgment of whether they can
50-min periods provided ample time in which to complete all accomplish a specific task and subsequently asking them to ac-
measures. complish the same, or a similar, task (see Bandura & Schunk,
1981; Schunk, 1981, 1982a, 1982b). In this study, the tasks on
which confidence was assessed were consistent with those on
Instrumentation which our two performance outcomes were measured.
The first outcome measure asked students to solve the same
Mathematics Self-Efficacy Scale—Revised. Bandura (1977) problems on which their math self-efficacy was assessed. The
wrote that, because the definition of self-efficacy is straightfor- problems performance measure used in this study, like the math
ward, efficacy questionnaires in academic areas are adequate op- problems efficacy scale, was one of the two final forms developed
erational measures of "the conviction that one can successfully by Dowling (1978). This was an 18-item, multiple-choice instru-
execute the behavior required to produce the outcomes" (p. 93). ment constructed with midrange difficulty items from the
Norwich (1986) also found that traditional self-report assessments NLSMA, developed specifically for use with college undergradu-
of self-efficacy had enough reliability and validity to warrant their ates, and composed of three nonorthogonal subscales consisting of
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

use. As we noted earlier, the MSES was developed by Betz and mathematics components, cognitive demand, and problem context.
This document is copyrighted by the American Psychological Association or one of its allied publishers.

Hackett (1983) to assess the math self-efficacy of college students. Dowling reported K-R 20 coefficient of reliability of .79 and mean
The instrument has 52 items and three subscales representing three item difficulty of .29. Dowling constructed the final efficacy and
domains of math-related behavior: solution of math problems, performance measures from four preliminary forms. For an item to
completion of math tasks used in everyday life (e.g., balancing a be included in the final instrument, several criteria had to be met:
checkbook, computing income taxes), and satisfactory perfor- (a) percentage correct between .30 and .70, (b) point-biserial
mance in college courses that require knowledge and mastery of correlation coefficient greater than .50, (c) discrimination index
mathematics (e.g., calculus, statistics, and biochemistry). greater than .40, and (d) significant corrected phi coefficient (p <
The Problems subscale of the MSES was adopted from the .01). Pajares and Kranzler (1994) reported a K-R 20 of .83 on a
mathematics confidence scales created by Dowling (1978), who similar instrument with a high school sample. We obtained a K-R
selected midrange difficulty items from the NLSMA and devel- 20 of .86 in the present study.
oped an instrument to specifically assess the math confidence of The second performance measure consisted of the students'
college students. The problems she selected represent three com- declared majors, which were rated along Goldman and Hewitt's
ponents of mathematics (arithmetic, algebra, and geometry), three (1976) science-nonscience continuum adapted for mathematics.
levels of cognitive demand (computation, comprehension, and The rating on this continuum provides a 5-point continuous scale
application), and two problem contexts (real and abstract). Betz that includes majors in the fine arts (1), humanities (2), social
and Hackett (1983) reported item-total score correlations ranging sciences (3), natural-biological sciences and lower order mathe-
from .29 to .63 for tasks, .33 to .73 for courses, and .24 to .66 for matics (4), and physical sciences and higher order mathematics
problems. Coefficient alphas for the three subscales were .90 for (5). This scale is the same used by Hackett (1985), Betz and
Tasks, .93 for Courses, and .92 for Problems. Betz and Hackett Hackett (1981, 1983), and Lent et al. (1991, 1993) in their studies
also reported a correlation of .66 between the MSES and the Math of self-efficacy and choice of major. To test the underlying con-
Confidence subscale of the Fennema-Sherman (1976) scales. tinuum of this variable, we subjected the scale to trend analysis,
Hackett and Betz (1989) reported on unpublished raw data by full model F(l, 389) = 41.95, p < .001. Results showed a
Hackett and O'Halloran (1985) that demonstrated moderate significant linear trend, F(2, 388) = 8.19, p < .0003.
2-week test-retest reliabilities for the total scale (.88) and for each
of the three subscales (.79 for Tasks, .91 for Courses, and .82 for
Problems). Since its creation, the MSES or one of its scales has Data Analysis
been used in a number of studies (e.g., Betz & Hackett, 1983;
Hackett, 1985; Hackett & Betz, 1989; Langenfeld & Pajares, 1993; The hypotheses of this study were that there would be a stronger
Lapan, Boggs, & Morrill, 1989; Lent, Lopez, & Bieschke, 1991, relationship between scores on the math problems self-efficacy
1993; Pajares & Miller, 1994; Randhawa et al., 1993). scale and those on the problem-solving performance measure than
between scores on the outcome measure and those of either the
We made two alterations to the MSES for use in this study. math tasks or courses self-efficacy scales, and that there would be
When Dowling (1978) constructed two final forms of math con- a stronger relationship between scores on the courses self-efficacy
fidence and performance, she constructed and tested preliminary scale and choice of majors than between that outcome measure and
scales, selecting items according to predetermined criteria. Her those of either the math problems or math tasks scales. Four
final forms reflected the results of this process. Betz and Hackett analyses were performed for each hypothesis: Zero-order correla-
(1983) selected a preliminary scale as the Problems subscale of the tions were obtained to discover the strength of relationships be-
MSES. We selected one of Dowling's final forms as the Problems tween variables, the Williams T 2 statistic was used to determine
subscale of the MSES—R. Dowling reported a correlation of .57 whether the correlations were significantly different, and a multi-
between the MCS and the Confidence subscale of Fennema and ple regression analysis was conducted to further inform the rela-
Sherman's (1976) Mathematics Attitude Scales. Langenfeld and tionships. Finally, the difference between the models in Figure 1
Pajares (1993) conducted a factor-analytic study with 520 under- and Figure 2 were tested using LISREL 7 (Joreskog & Sorbom,
graduates using the MSES-R with a 5-point Likert scale and 1989). This procedure provides a joint test for the significance of
reported that the instrument suffered no loss of internal consis- the four paths missing in Figure 2.
tency (alpha coefficients = .94 for Tasks, .91 for Courses, and .91
for Problems). Pajares and Kranzler (1994) reported Cronbach's
alpha coefficient of .92 using a version of the Problems subscale Results
with 329 high school students. We also used a 5-point scale and
obtained alpha coefficients of .90 for Problems, .92 for Courses, Table 1 contains the means, standard deviations, and
and .91 for Tasks in the present study. Pearson product-moment correlations for the three
 MATHEMATICS SELF-EFFICACY 195

Table 1
Correlations Between Math Self-Efficacy Scales and Math Performances
Variable M SD 1 2 3 4 5 6
1. Problems self-efficacy 73.8 10.3 —
2. Tasks self-efficacy 72.1 11.4 .66* —
3. Courses self-efficacy 54.0 11.8 .67* .66* —
4. Full-scale math self-
efficacy 199.9 29.6 .87* .88* .89* —
5. Problem-solving
performance 14.0 2.8 .69* .45* .53* .63* —
6. Choice of math-related
major 3.0 0.9 .39* .34* .52* .48* .28* —
' p < .0001.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
This document is copyrighted by the American Psychological Association or one of its allied publishers.

MSES-R scales, the total scale, and the two outcome mea- ing performance as the dependent variable in one analysis
sures (problem solving and choice of majors). Problems and choice of majors in another, as shown in Tables 2 and
self-efficacy scores ranged from 36 to 90. Possible scores on 3, help clarify the relationships. Both models were signifi-
this 18-item subscale ranged from 18 to 90; hence, although cant. The model with problem-solving performance ac-
some students expressed maximum confidence in their counted for 48% of the variance, F(3, 387) = 120.89, p <
problem-solving ability, even the least confident student .0001, adjusted R2 = .48; the model with choice of majors
averaged 2 on the 5-point Likert scale. Also, the mean score accounted for 27%, F(3, 387) = 47.66, p < .001, adjusted
of 4.1 (much confidence) per item suggested that students R2 = .26. Although problems self-efficacy and courses
were largely confident about their ability to solve the prob- self-efficacy both achieved significance in the problem-
lems. The same was true for the 18-item Tasks self-efficacy solving performance model, the difference between them
subscale—scores ranged from 30 to 90. When asked was pronounced. Students' confidence in their capability to
whether they could complete a math-related course with a solve the problems had a much stronger influence on their
final grade of B or A, however, students showed less con- problem-solving performance than did their confidence to
fidence. Scores on this 16-item scale ranged from 22 to 80, succeed in math-related courses. Confidence to perform
with a mean score of 3.4 per item. Total math self-efficacy math-related tasks was nonsignificant in both models.
scores ranged from 108 (minimum possible = 52) to 260 Finally, a test for the differences between the models
(maximum possible), with a mean score of 3.8 per item. On presented in Figures 1 and 2 yielded a nonsignificant dif-
the whole, students were confident of their math self-effi- ference, ^ ( 4 , N = 391) = 3.58, suggesting that the four
cacy. Scores on the 18-item math performance outcome additional paths in Figure 1 do not help in the explanation of
measure ranged from 4 to 18, with a mean of 14.0. These the data. Because of these results and those of the Williams
results are consistent with those of Hackett and Betz (1989) T 2 analysis, we concluded that it is not only theoretically
and Pajares and Miller (1994). sound but empirically warranted to use math problems
As expected, the efficacy subscales were strongly related self-efficacy as the appropriate self-efficacy assessment
with each other. Also, as hypothesized, problems self-effi- when problem-solving performance is the criterial task and,
cacy had a stronger relationship with problem-solving per- similarly, to use courses self-efficacy when a related out-
formance than did either tasks or courses self-efficacy, come variable such as choice of math-related majors is the
although all were significant. We used Williams's (1959) T 2 criterial task.
modification of Hotelling's T1 to discover whether the dif-
ferences between these correlations were significant. Criti- Discussion
cal value for the Williams T 2 was f(388, a < .05) = 1.96.
Results showed that the correlation between problem-solv- Results from this study confirm that Bandura's (1986)
ing performance and problems self-efficacy was signifi- warnings regarding self-efficacy and performance assess-
cantly stronger than that between either performance and
courses self-efficacy, T 2 = 5.00, or tasks self-efficacy, T 2 =
7.76.
Table 2
A similar phenomenon occurred between the three scales Multiple Regression Analysis of the Effect
and choice of majors, whose strongest correlation was with of the Math Self-Efficacy Subscales on Math
courses self-efficacy. Again, results of Williams T 2 showed Problem-Solving Performance
that the correlation between courses self-efficacy and choice
of majors was significantly stronger than that between either Parameter Standard Probability
choice of majors and problems self-efficacy, T 2 = 3.51, or Variable estimate error t > \t\
tasks self-efficacy, T 2 = 4.80. Problems self-efficacy 0.170 0.015 11.70 .0001
Results of two multiple regression analyses with the three Tasks self-efficacy -0.019 0.013 -1.46 .1460
subscales as independent variables and math problem-solv- Courses self-efficacy 0.041 0.013 3.19 .0015
 196 FRANK PAJARES AND M. DAVID MILLER

Table 3 more closely matching the performance task. Hackett and

Multiple Regression Analysis of the Effect of Betz (1989) suggested that teachers and counselors would
the Math Self-Efficacy Subscales on Choice of be well advised to pay as much attention to students' self-
Undergraduate Major efficacy beliefs as they presently pay to actual test scores,
Parameter Standard Probability
for it is these beliefs that may better reveal the choices and
Variable estimate error ; > \t\ decisions students will make about matters such as choice of
college courses or majors and directions for career devel-
Problems self-efficacy 0.009 0.006 1.56 .1205 opment. This is sound advice, if it is tempered by the
Tasks self-efficacy -0.003 0.005 -0.57 .5698
Courses self-efficacy 0.036 0.005 7.45 .0001 warning that professionals would also be well served by
focusing on the self-efficacy beliefs most predictive of the
outcomes they wish to predict.
Some researchers have noted the need to explore the
merit are well founded. There are different ways of assess- generality of self-efficacy beliefs—that is, the extent to
ing self-efficacy, but the most theoretically appropriate and which they relate to, or transfer across, different perfor-
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

empirically warranted is one in which the self-efficacy mance tasks or domains (Lent & Hackett, 1987; Multon et
This document is copyrighted by the American Psychological Association or one of its allied publishers.

measure assesses the same or similar skills required for the al., 1991). Lent and Hackett (1987) wrestled with this issue
performance task. Students' confidence to solve mathemat- when they argued that the level of specificity of an efficacy
ics problems was a more powerful predictor of their ability assessment should ultimately depend on the complexity of
to solve those problems than was their confidence to per- the performance criteria with which it is compared (pp.
form math-related tasks or their confidence to earn As or Bs 365-366). Math-related outcomes, particularly in investiga-
in math-related courses. Similarly, their confidence to suc- tions of career choices and decisions, are seldom as precise
ceed in such courses was more predictive of their choice of and specific as one's capability to solve mathematics prob-
majors that required them to take many of the math-related lems. Lent and Hackett rightly observed that specificity and
courses on which they expressed that confidence. As we precision are often purchased at the expense of external
outlined in the introduction, the self-efficacy-performance validity and practical relevance.
correspondence between confidence to succeed in math- Our findings suggest that research questions should be
related courses and choice of major is not as specific as that formulated with an eye to enhancing the correspondence
between confidence to solve problems and actual capability between self-efficacy and performance assessments, espe-
to solve those problems. Consequently, we expected the cially if the aim is to help predict or explain academic
latter relationship to be stronger, and indeed it was. This outcomes. Bandura (1986) cautioned against the use of
finding also supports Bandura's (1986) theoretical argument omnibus tests to assess "global dispositions" (p. 396). It
and our empirical hypothesis. may be argued that the use of the composite score of a
Researchers have observed that even generalized or less multiscale test that provides a general judgment of capabil-
closely related indexes will correlate significantly with spe- ity for an entire academic area is, in essence, assaying a
cific outcomes, and we found this phenomenon as well (see global disposition (in this case, one's generalized mathe-
Multon et al., 1991). That is, each subscale, as well as the matics confidence) with an omnibus test. Nonetheless, we
full scale, correlated significantly with each outcome. As we agree that full-scale scores from reliable, multiscale aca-
found, however, prediction is enhanced as self-efficacy and demic self-efficacy instruments tap into a general factor of
performance correspondence more closely matches. The academic or subject-specific capability. This general self-
clear implication of these results is that researchers are well perception provides teachers and counselors with informa-
advised to consider the nature of the performance task when tion regarding students' general mathematics confidence
choosing or developing an appropriate self-efficacy mea- (see Kranzler & Pajares, 1994), and results may be useful in
sure. As findings showed, this advice is both theoretically predicting complex math-related outcomes that do not eas-
and empirically sound. ily lend themselves to microanalytic analysis.
It is not altogether easy to see what value full-scale scores As we have argued regarding the full-scale MSES, we
provided by multiple-scale instruments such as the MSES or admit that we see limited value in a scale that asks students
the MSES-R may have if one wishes to predict specific to provide confidence judgments to perform 18 varied,
mathematics outcomes. Such global scores decontextualize math-related tasks. Each judgment is logically related only
efficacy beliefs and transform the construct of self-efficacy to the task on which that judgment is made. Computing
into a generalized personality trait rather than the context- one's income taxes for a year or figuring out how much
specific judgment Bandura (1986) suggested it is (pp. lumber one needs to buy in order to build a set of book-
5-12). If, as Bandura argued, self-efficacy assessment must shelves, two of the tasks in the Tasks self-efficacy subscale,
conform to the criterial task to be useful and predictive, clearly have "figuring" in common. However, it is difficult
what criterial task can be compared with a global score that to see what one's judgments of confidence to accomplish
comprises judgments of confidence to succeed in math these two tasks may predict other than one's ability to
courses, to complete math-related tasks, and to solve spe- accomplish the specific tasks themselves. Neither judgment
cific math problems? Moreover, one might question the may prove particularly useful in understanding a myriad of
practical utility of administering a 52-item instrument when other math-related tasks, choosing a math-related major, or
greater prediction may be had from a shorter instrument predicting other mathematical capabilities. Completing cer-
 MATHEMATICS SELF-EFFICACY 197

tain IRS forms, for example, may tax the minds and wills of References
even our finest psychometricians and nuclear scientists.
Bandura, A. (1977). Self-efficacy: Toward a unifying theory of
We expect that the implications from our findings will not
behavioral change. Psychological Review, 84, 191-215.
be viewed as limited to investigations of mathematics self- Bandura, A. (1986). Social foundations of thought and action: A
efficacy. The mismatch between self-efficacy and perfor- social cognitive theory. Englewood Cliffs, NJ: Prentice Hall.
mance assessment is also prevalent in investigations in other Bandura, A., & Schunk, D. H. (1981). Cultivating competence,
academic areas. Shell, Murphy, and Bruning (1989) re- self-efficacy, and intrinsic interest through proximal self-moti-
ported significant correlations between measures of reading vation. Journal of Personality and Social Psychology, 41, 586-
and writing self-efficacy and related outcomes. Their self- 598.
efficacy instrument was composed of two scales—one mea- Benson, J. (1989). Structural components of statistical test anxiety
in adults: An exploratory study. Journal of Experimental Edu-
sured students' confidence that they could perform specific
cation, 57, 247-261.
reading or writing tasks and another that they possessed Betz, N. E., & Hackett, G. (1981). The relationship of career-
specific skills. The criteria for assessing reading and writing related self-efficacy expectations to perceived career options in
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

performance were similar only to the skills on which the college women and men. Journal of Counseling Psychology, 28,
This document is copyrighted by the American Psychological Association or one of its allied publishers.

skills self-efficacy scale were based. Although the research- 399-410.

ers reported that the skills subscale scores had stronger Betz, N. E., & Hackett, G. (1983). The relationship of mathematics
correlations with performance than did the task subscale's self-efficacy expectations to the selection of science-based col-
scores, they used the composite score as the self-efficacy lege majors. Journal of Vocational Behavior, 23, 329-345.
assessment. Using procedures and instruments similar to Cooper, S. E., & Robinson, D. A. G. (1991). The relationship of
mathematics self-efficacy beliefs to mathematics anxiety and
those of Shell et al., Pajares and Johnson (1994) also found
performance. Measurement and Evaluation in Counseling and
that writing skills self-efficacy was a stronger predictor of Development, 24, 4-11.
essay scores holistically assessed in terms of writing skills Dowling, D. M. (1978). The development of a mathematics con-
than was writing tasks self-efficacy. Consequently, we sug- fidence scale and its application in the study of confidence in
gest that investigators in academic areas other than mathe- women college students. Unpublished doctoral dissertation,
matics also think carefully about the match between the Ohio State University, Columbus.
appropriateness of self-efficacy and outcome assessment. Fennema, E., & Sherman, J. A. (1976). Fennema-Sherman Math-
ematics Attitude Scales: Instruments designed to measure atti-
Our findings also have implications for the type of math-
tudes toward the learning of mathematics by females and males.
ematics research that focuses on international and multicul- JSAS Catalog of Selected Documents in Psychology (No. 1225),
tural comparisons of math confidence. For example, results 6, 31.
from the Second Study of Mathematics conducted by the Goldman, R. D., & Hewitt, B. N. (1976). The scholastic aptitude
International Association for the Evaluation of Educational test "explains" why college men major in science more often
Achievement (Robetaille & Garden, 1988) revealed that than college women. Journal of Counseling Psychology, 23,
although Japanese students outperformed American stu- 50-54.
dents in most areas of mathematics, American students Hackett, G. (1985). The role of mathematics self-efficacy in the
exhibited greater confidence in their mathematical ability. choice of math-related majors of college women and men: A
path analysis. Journal of Counseling Psychology, 32, 47-56.
This confidence, however, was assessed with global, decon-
Hackett, G., & Betz, N. E. (1989). An exploration of the mathe-
textualized items of the sort that have plagued self-efficacy matics self-efficacy/mathematics performance correspondence.
research (e.g., "I am not so good at mathematics"). Robe- Journal for Research in Mathematics Education, 20, 261-273.
taille and Garden (1988) posited that the negative responses Hackett, G., & O'Halloran, S. (1985). [Test-retest reliability of
of Japanese students may have been due to a culturally career self-efficacy measures]. Unpublished raw data.
based reluctance to boast about their academic ability but Joreskog, K. G., & Sorbom, D. (1989). LISREL 7 user's guide.
argued that a genuine lack of self-confidence was a more Mooresville, IN: Scientific Software.
likely explanation. Regardless of which explanation is more Kranzler, J., & Pajares, F. (1994). An exploratory factor analysis
likely, it would be instructive to assess mathematics self- of the Mathematics Self-Efficacy Scale—Revised. Manuscript
submitted for publication.
efficacy in international and multicultural settings in a con- Langenfeld, T. E., & Pajares, M. F. (1993, April). The Mathemat-
textual manner consistent with the guidelines of social cog- ics Self-Efficacy Scale (MSES): Refining the construct. Paper
nitive theory and with the findings from this investigation. presented at the meeting of the American Educational Research
If, for example, the criterial task in such international stud- Association, Atlanta, GA.
ies involves the solving of math problems, the confidence Lapan, R. T., Boggs, K. R., & Morrill, W. H. (1989). Self-efficacy
assessment should consist of students' judgments of their as a mediator of investigative and realistic general occupational
confidence to solve those problems rather than of global themes on the Strong-Campbell Interest Inventory. Journal of
Counseling Psychology, 36, 176-182.
confidence statements infused with personal judgments of
Lent, R. W., & Hackett, G. (1987). Career self-efficacy: Empirical
self-worth. The results would be a much less culturally status and future directions. Journal of Vocational Behavior, 30,
biased assessment, a clearer and more specific judgment of 347-382.
perceived capability, and a better understanding of the Lent, R. W., Lopez, F. G., & Bieschke, K. J. (1991). Mathematics
relationship between self-efficacy beliefs and academic self-efficacy: Sources and relation to science-based career
outcomes. choice. Journal of Counseling Psychology, 38, 424-430.
 198 FRANK PAJARES AND M. DAVID MILLER

Lent, R. W., Lopez, F. G., & Bieschke, K. J. (1993). Predicting Robetaille, D. F., & Garden, R. A. (Eds.). (1988). The IEA study of
mathematics-related choice and success behaviors: Test of an mathematics: 2. Contexts and outcomes of school mathematics.
expanded social cognitive model. Journal of Vocational Behav- Oxford, England: Pergamon Press.
ior, 42, 223-236. Schunk, D. H. (1981). Modeling and attributional effects on chil-
Maddux, J. E., Norton, L. W., & Stoltenberg, C. D. (1986). dren's achievement: A self-efficacy analysis. Journal of Educa-
Self-efficacy expectancy, outcome expectancy, and outcome tional Psychology, 73, 93-105.
value: Relative effects on behavioral intentions. Journal of Per- Schunk, D. H. (1982a). Effects of effort attributional feedback on
sonality and Social Psychology, 51, 783-789. children's perceived self-efficacy and achievement. Journal of
Multon, K. D., Brown, S. D., & Lent, R. W. (1991). Relation of Educational Psychology, 74, 548-556.
self-efficacy beliefs to academic outcomes: A meta-analytic
Schunk, D. H. (1982b). Verbal self-regulation as a facilitator of
investigation. Journal of Counseling Psychology, 38, 30-38.
children's achievement and self-efficacy. Human Learning, 1,
Norwich, B. (1986). Assessing perceived self efficacy in relation
265-277.
to mathematics tasks: A study of the reliability and validity of
assessment. British Journal of Educational Psychology, 56, Schunk, D. H. (1989). Self-efficacy and achievement behaviors.
180-189. Educational Psychology Review, 1, 173-208.
Schunk, D. H. (1991). Self-efficacy and academic motivation.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

Pajares, F., & Johnson, M. J. (1994). Confidence and competence

Educational Psychologist, 26, 207-231.
This document is copyrighted by the American Psychological Association or one of its allied publishers.

in writing: The role of self-efficacy, outcome expectancy, and

apprehension. Research in the Teaching of English, 29, 313— Shell, D. F., Murphy, C. C , & Bruning, R. H. (1989). Self-efficacy
331. and outcome expectancy mechanisms in reading and writing
Pajares, F., & Kranzler, J. (1994). Role of self-efficacy, self- achievement. Journal of Educational Psychology, 81, 91-100.
concept, and general mental ability in mathematical problem- Smith, R. J., Arnkoff, D. B., & Wright, T. L. (1990). Test anxiety
solving. Manuscript submitted for publication. and academic competence: A comparison of alternative models.
Pajares, F., & Miller, M. D. (1994). The role of self-efficacy and Journal of Counseling Psychology, 37, 313-321.
self-concept beliefs in mathematical problem-solving: A path Williams, E. J. (1959). The comparison of regression variables.
analysis. Journal of Educational Psychology, 86, 193-203. Journal of the Royal Statistical Society (Series B), 21, 396-399.
Randhawa, B. S., Beamer, J. E., & Lundberg, I. (1993). Role of
mathematics self-efficacy in the structural model of mathematics
achievement. Journal of Educational Psychology, 85, 41-48. Received March 28, 1994
Reyes, L. H. (1984). Affective variables and mathematics educa- Revision received July 18, 1994
tion. The Elementary School Journal, 84, 558-581. Accepted August 1, 1994 •