Professional Documents
Culture Documents
The construct of self-efficacy has received increasing attention in the literature (Gist & Mitchell, 1992). Bandura (1986)
defined perceived self-efficacy as follows:
Cynthia Lee, College of Business Administration, Northeastern University; Philip Bobko, Department of Management, Rutgers University.
Preparation of this article was facilitated by the Riesman Award to
Cynthia Lee from the College of Business Administration at Northeastern University.
We thank Tom Lee, Edwin Locke, Karyll Shaw, and several anonymous reviewers for their comments. We also thank Sharon Beckstrom,
Joseph Czajka, Jeffrey Mello, Leticia Pena, and Frank Spital for their
assistance in data collection.
Correspondence concerning this article should be addressed to Cynthia Lee, College of Business Administration, Northeastern University,
304 Hayden Hall, Boston, Massachusetts 02115.
364
SELF-EFFICACY
zations and to present a validation of a partial nomological network incorporating these self-efficacy operationalizations. We
conducted two studies, using two different tasks, with students
as subjects. Study 1 was conducted in a natural classroom setting in which the subject matter (introduction to organizational
behavior) served as the complex, but somewhat novel task.
Wood and Locke's (1987) academic self-efficacy scale (composed of multiple task elements) was used to examine the convergent and predictive validities of the self-efficacy magnitude,
strength, and composite measures. In Study 2 we assessed the
generalizability of Study 1 by using a simple and novel brainstorming task. Our self-efficacy measure was consistent with the
one used by Locke et al. (1984), who also used the brainstorming task in their study. Thus, we arrived at the following
hypothesis.
Hypothesis 1: There will be positive relationships among all selfefficacy operationalizations (some of this will be because two measures are components of the composite measures). More important, we expect that the single-item confidence rating will show the
least convergence with the other four measures, because of the potential for generally low reliability, low validity, or both from singleitem ratings.
Antecedents of Self-Efficacy
Because we did not intend to review all the determinants of
self-efficacy in the literature (cf. Bandura, 1986), we only examined past performance and affectivity as antecedents of selfefficacy. Of the major sources of self-efficacy, both ability and
past performance have consistently been found to be positively
related to self-efficacy (Bandura, 1986; Locke & Latham, 1990;
Podsakoff& Farh, 1989; Wood & Bandura, 1989). Furthermore, Kavanagh and Bower (1985) found that subjects in whom
a happy mood had been induced reported higher overall selfefficacy than those in whom no mood had been induced. Bandura (1986) further noted that if positive mood activates
thoughts of accomplishments, perceived self-efficacy will be
boosted. As noted in Study 2, we had the opportunity to collect
data on positive mood states. This led to our second hypothesis.
Hypothesis 2: There will be a positive relationship between both
past performance and positive affectivity with self-efficacy (Study
2 only). Consistent with Hypothesis 1, we expect the single-item
measure to have the lowest validity with these antecedents. Furthermore, we suggest that the third and fourth measures of selfefficacy will have the highest validities with antecedent measures,
because of their information richness and because they reflect the
original conceptualization of the self-efficacy construct.
Consequences of Self-Efficacy
The overall thrust of self-efficacy research is to provide both a
mechanism that mediates behavior change and a parsimonious
account of why and how different techniques effect change
(Bandura, 1986). Other consequences of self-efficacy include
the setting of higher personal goals and higher task performance
(Bandura, 1986; Locke & Latham, 1990; Wood & Bandura,
1989). With this in mind, we proposed our third hypothesis.
365
MEASURES
Study 1
In Study 1, we investigated the convergent and predictive validities of the various operationalizations of self-efficacy level,
strength, and composite operationalizations, using Wood and
Locke's (1987) academic self-efficacy measure.
Method
Respondents and Procedures
Participants (N = 207) were predominantly third-year undergraduates in a 5-year degree program from six introductory management
classes. There were 113 men and 83 women; 11 undergraduates did not
indicate their gender. The average age was 21 years. Students were asked
to participate voluntarily in a study on personality. Individuals who
agreed to participate (approximately 90% of all the students) completed
Wood and Locke's (1987) academic self-efficacy questionnaire after the
students had taken their first examination in the course but before they
had received their grades. This timing was purposive, so that subjects
would have some information about their ability to cope with the course
demands and yet would not base their ratings simply on their grades
(Wood & Locke, 1987).
Measures
Self-efficacy.
Because our sample consisted of students, we used
Wood and Locke's (1987) academic self-efficacy scale.
As did Wood and Locke, we excluded items that had extremely
skewed distributions with very little or no dispersion. As a result, only
two items out of five from each subscale were retained. For purposes of
comparison with Wood and Locke's results, we also combined six of
their seven subscales of class concentration, memorization, understanding, explaining concepts, discriminating concepts, and note taking for
subsequent analyses. The Cronbach coefficient alphas for each of these
subscales were .86, .82, .80, .86, .81, and .84, respectively. The use of
coefficient alphas to estimate the reliability of self-efficacy strength and
magnitude is somewhat deceiving because Gunman's scaling was used
to construct the self-efficacy measures. We report coefficient alphas for
the purposes of completeness because most empirical studies have used
the same reliability estimates. In Study 2, the test-retest correlations
were .77 and .78 for self-efficacy strength and magnitude, respectively.
Students were asked to indicate if they could achieve a certain level of
attainment (yes or no) as well as their degree of confidence in their ability to perform at that level (on a scale from 0 to 10). Self-efficacy level,
or magnitude, was defined as the total number of yess divided by the
total number of items.
We defined self-efficacy strength as the mean confidence rating, using
a scale ranging from completely unconfidenl (0) to completely confident
(10). We computed self-efficacy strength by summing all of the scores
across items and then dividing by the total number of items.
Self-efficacy composite measures of strength and magnitude were
computed in two ways. In one method we took the raw scores of selfefficacy strength and then summed these across self-efficacy levels that
were answered yes (total self-efficacy). The other method was to standardize the self-efficacy strength items by converting them to z scores
and then sum them across all self-efficacy levels that were answered yes.
366
Results
Table 1 shows the intercorrelations of the various operationalizations of self-efficacy. The mean correlation, after applying
Fisher's z transformation, was .62. (Note that the particularly
high correlations were somewhat due to common components
across measures.) Consistent with Hypothesis 1, the one-item
confidence rating generally demonstrated the lowest convergent
validity with the remaining measures. One of the noncombination measures (strength) also demonstrated relatively weaker
convergence. Table 2 shows the correlations of these operationalizations of self-efficacy with examination scores and self-set
goals. The results supported Hypothesis 3; that is, although the
self-efficacy operationalizations were all positively related to
self-set goals and task performance, the strongest and most consistent positive relationships tended to be from the two composite measures of self-efficacy.
Summary
The results provided support for the convergent validity of
the self-efficacy measures. The one-item confidence rating and
self-efficacy strength appeared to have the lowest correlation
with all other self-efficacy measures and tended to be relatively
weakly related to examination score and examination-score
goal. With the exception of the one-item confidence rating and
self-efficacy magnitude, Study 1 shows that the two self-efficacy
composites and self-efficacy strength operationalizations correlated highest with self-set goals. In interpreting the correlations
in Table 2, note that any two correlations within a column can
Table 1
Correlations Among Self-Efficacy
Measure
1.
2.
3.
4.
5.
SE magnitude
SE strength
SE Composite 1
SE Composite 2
Confidence rating
1
_
.54**
.94**
.90**
.29**
Table 2
Zero-Order Correlations Between Self-Efficacy
and Goals and Performance: Study 1
Exam score
Measure
SE magnitude
SE strength
SE Composite 1
SE Composite 2
Confidence rating
.28**
.13*
Self-set goal
r
xyi
.06
.15*
.21**
.25**
.32**
.13*
.12*
.20**
.20**
.08
.23*
.49*
.48*
.53*
.25*
Note. Values in the table are validity coefficients. The confidence rating consisted of one item. SE Composite 1 = raw scores of SE strength
summed across SE magnitude items that were answered yes; SE Composite 2 = standardized SE strength items summed across SE magnitude items that were answered yes. rxyi = correlation with first examination score; rx>a = correlation with second examination score; rfy3 =
correlation with second examination score's goal.
*/><.05. **p<.01.
be tested to see if they are statistically different by using Hotelling's test for dependent correlations (cf. Cohen & Cohen,
1983). Each test statistic is somewhat unique because of the
changing magnitudes of the intercorrelations. However, given
our sample sizes, differences in correlations over. 15 tend to be
statistically significant. Thus, for example, both of the composite measures had higher correlations with self-set goals than did
the single-item confidence rating of self-efficacy. The complete
set of statistical tests is available from Cynthia Lee.
As noted by Wood and Locke (1987), performance feedback
is usually task specific in laboratory studies. In contrast, an examination score is feedback from multiple tasks (e.g., studying,
class attendance, note taking, and memorizing). This performance feedback is only given for overall examination performance, not for the individual task components that bring the
examination score about. Thus, the self-efficacy ratings for such
tasks are probably not as accurate as they would be in singletask studies. Furthermore, Gist and Mitchell (1992) stated that
in complex task situations, the complexity of requiring individuals to estimate numerous skill and motivational parameters
may increase the error of assessment. Therefore, in Study 2 we
used a single, but novel, brainstorming task to further investigate the convergent validity of the various operationalizations
of self-efficacy magnitude, strength, and composite measures.
In addition, we again examined the relative validities of selfefficacy operationalizations to the hypothesized antecedents
and outcomes.
Study 2
.71**
(SE)
.86**
.28**
.28**
Method
Note. Ns = 163-205 for survey after the first exam. The confidence
rating consisted of one item. SE Composite 1 = raw scores of SE
strength summed across SE magnitude items that were answered yes;
SE Composite 2 = standardized SE strength items summed across SE
magnitude items that were answered yes.
*p<.05. **p<.0l.
Respondents
The subjects were 92 undergraduates from three introductory management classes of a 5-year degree program. As part of another study
(Lee & Bobko, 1992), each class was told that they would be experiencing a different form of learning (through the use of a brainstorming
exercise). Although students in each class were asked to volunteer, they
367
SELF-EFFICACY MEASURES
all participated. After the exercise was over, students were debriefed and
thanked. The mean age of these students was 22 years. There were 33
women and 57 men; 2 students did not indicate their gender.
Table 3
Correlations Among Self-Efficacy (SE) Measures
After Two Practice Trials: Study 2
Measure
Task
1.
2.
3.
4.
5.
Conditions
There were three assigned goal levels: (a) the somewhat difficult goal
(n = 35) of generating at least 8 uses for a single object in 3 min, (b) a
moderately difficult goal (n = 28) of generating 12 uses in 3 min, and (c)
a very difficult goal (n = 29) of generating 16 uses in 3 min. These goal
levels were achieved by 50%, 12%, and 5%, respectively, of the subjects
during a pilot study.
Procedure
At the beginning of the class period, the task was explained and the
subjects were then allowed two practice trials. After these practice trials,
subjects were given their assigned goals. Measures of perceived goal
difficulty and a self-efficacy scale were then administered. Subjects were
then asked for their personal goal for the next trial. Finally, subjects were
asked to brainstorm for two experimental trials.
Measures
We measured individual task performance by the total number of
uses given, deleting responses that were inappropriate (e.g., "break it"
for the object "brick") or duplications within the same trial.
We assessed a subject's self-set, or personal, goal by their responses to
a single item asking how many uses the subject intended to list for the
two practice trials and the two experimental trials.
Self-efficacy was assessed in a manner consistent with Study 1. Subjects were asked to indicate yes or no as to whether they could perform
at five different levels of proficiency (i.e., listing 4, 8, 12, 16, and 20
uses). Proficiency level (self-efficacy magnitude) was measured by the
total number of yes responses for listing the uses for each object divided
by the number of items to which subjects responded with a yes. For each
of the five proficiency levels, subjects also estimated their confidence
(self-efficacy strength) about their present capability to perform at that
level on a scale from no confidence at all (0) to totally confident (100).
Inspection of the individual items (five levels of proficiency) indicated
that two of the items had very little dispersion and were making negligible contributions to the overall measure. As in Wood and Locke (1987),
these items (of listing 4 and 8 uses of an object) were deleted from subsequent analyses. The self-efficacy magnitude, strength, and composite
scores were computed in the same manner as in Study 1. Likewise, we
again used a one-item confidence rating by asking respondents "How
confident are you that you will attain the goal of listing at least 'X' uses
in 3 minutes for the next trial?" The Cronbach alphas for the self-efficacy magnitude and strength measures obtained after the first two practice trials were .73 and .79, respectively, and their respective alphas after
the first experimental trial were .74 and .77.
We measured positive affectivity on a 10-item scale developed and
validated by Watson, Clark, and Tellegen (1988). Respondents indicated
on a 5-point Likert-type scale the extent to which they felt, for example,
"excited," "enthusiastic," "determined," and "attentive" over the past
year. The Cronbach alpha for this scale was .88.
SE magnitude
SE strength
SE Composite 1
SE Composite 2
Confidence rating
.59*
.76*
.73*
.31*
.90*
.50*
.28*
.85*
.40*
.41*
Results
Table 3 shows the correlations among the self-efficacy measures in Study 2. The correlations are consistent with those of
Study 1. The mean correlation of the self-efficacy measures, using Fisher's z transformation, was also .62. Thus, as expected,
the results supported Hypothesis 1 regarding the convergent validity of the self-efficacy measures. As hypothesized, the singleitem confidence rating demonstrated the lowest convergent validity with the remaining measures.
Part of Hypothesis 2 predicted positive relationships between
self-efficacy and the antecedent measures of past performance
and positive affectivity. Our data (see the positive affectivity and
past performance columns in Table 4) supported this hypothe-
Table 4
Correlations of Self-Efficacy
and Outcome Measures
Measure
PA
PA
Goal
1
Goal
2
Past
Present
.37**
.36**
.37**
.34**
.52**
.54**
.61**
.53**
.54**
.53**
.51**
.65**
.60**
.48**
.53**
.54**
.63**
.53**
.57**
.51**
.50**
.64**
.60**
.54**
.29**
.22*
.28**
.19*
.41**
.58**
.52**
.63**
.53**
.34**
.51**
.50**
.63**
.58**
.41**
.31**
.23**
.30**
.19*
.41**
.55**
.51**
.60**
.53**
.43**
.49**
.49**
.61**
.41**
.49**
Note. PA = positive affectivity; Goal 1 = self-set goal after the first two
practice trials; Goal 2 = self-set goal after the first experimental trial;
Past = past performance; Present = present performance; SE Composite 1 = raw scores of SE strength summed across SE magnitude items
that were answered yes; SE Composite 2 = standardized SE strength
items summed across SE magnitude items that were answered yes.
*p<.05. **p<.Q[.
368
Discussion
These two studies demonstrate that the five self-efficacy operationalizations are highly correlated. Of the five operationalizations of self-efficacy, the measures with respect to multiple performance levels (self-efficacy strength, self-efficacy magnitude,
and the two composite measures) showed higher convergent and
predictive validities than the one-item task-specific confidence
rating. These relationships appeared in both studies. On the basis of these two empirical studies and the increased information
in the other four measures, we thus recommend that researchers
refrain from using single-item measures of confidence as indexes of self-efficacy. Also, the self-efficacy magnitude and
strength measures appear to have generally weaker predictive
validities and correlations than the self-efficacy composites
have. Of the two composite measures, Self-Efficacy Composite
1 showed slightly stronger relationships with personal goals in
Study 2; but in most cases across both studies, the strength of
the relationships of these composite measures with antecedents
and outcomes were similar. Because the operationalization of
these two composites was consistent with Bandura's (1986) conceptualization (incorporating both magnitude and strength information) and because Composite 1 is less cumbersome to pperationalize, this may be the measure of choice.
It is interesting to speculate why the one-item confidence rating was more highly related to positive affectivity (and past performance) than were the other self-efficacy measures. The confidence rating asks about confidence "that you will obtain the
score that you are trying for in your next exam." As one reviewer noted, this may well be interpreted as an item assessing
confidence in outcome expectancy rather than self-efficacy. According to Bandura (1986), self-efficacy is a judgment of one's
capability to accomplish a certain level of performance,
whereas an outcome expectation is a judgment of the likely consequence such behavior will produce.
It is also suggestive that outcome expectancy does not occur
independent of an individual's past performance. Notice also
that Watson et aJ. (J 988) assessed their positive affectivity mea-
sure by using a past-tense orientation. So both positive affectivity and past performance differ from the goal measures and
present performance, with the latter having a forward-looking
orientation. Bandura (1986) maintained that "people see outcomes as contingent on the adequacy of their performances, and
care about those outcomes, that they rely on self-judged efficacy
in deciding which courses of action to pursue" (p. 392). This
distinction between past- and future-oriented measures may, in
part, explain the unexpected large correlations of the confidence rating with the positive affectivity and past performance
measures.
Bandura (1986) noted that a self-efficacy measure and the
behavioral test with which it is being correlated should be administered closely in time. In Study 1, self-efficacy was assessed
after students had taken their first examination but before they
had received their grade. In Study 2, self-efficacy was assessed
after the practice trials of a brainstorming task. However, because of the nature of the brainstorming task, immediate objective feedback on task performance was available in Study 2; this
was less so in Study 1, because subjects had not yet received
their examination scores. According to Kazdin (1978), in the
case of Study 2, completing self-efficacy items (especially when
the items are hierarchically presented) may sensitize subjects to
the number of uses they should list on the brainstorming task.
Alternatively, task performance (listing uses of an object) may
alter subsequent self-efficacy perceptions because these two
measures are administered at nearly the same time. Thus, the
net effect of this measurement context (self-efficacy items are
framed in performance levels) and procedure may result in the
observed higher correlations between self-efficacy and performance in Study 2. Also note that, in Study 1, the self-efficacy
subscales developed by Wood and Locke (1987) were not directly framed in performance levels (i.e., levels of examination
scores) but in terms of task requirements essential for successful
class performance (e.g., class concentration, memorization, understanding, and note taking). Because the level of self-efficacy
was inferred from a different set of items than those used for
behavioral or performance assessment, the correlation between
self-efficacy and behavioral performance would decline. Thus,
in addition to the timing of self-efficacy assessment, the measurement context and content of self-efficacy are also critical
and deserve closer attention from researchers.
In summary, it appears that there is convergent and predictive
validity across different types of self-efficacy measures and
different performance contexts. The composite self-efficacy operationalizations are consistent with Bandura's (1986) definition and conceptualization, and the two present studies provide
empirical evidence that these may be the measures of choice.
Future studies should replicate our results in longitudinal field
settings where performance feedback is perhaps less accurate,
or less timely, and where performance measures are often less
quantifiable. Not all recent operationalizations of self-efficacy
are equal. We suggest that researchers consider using and researching the relatively more information-rich (composite)
measures in future empirical settings.
References
Bandura, A. (1986). Social foundations of thought and action: A social
cognitive theory. Englewood Cliffs, NJ: Prentice Hall.
SELF-EFFICACY MEASURES
Bandura, A., & Jourden, F. J. (1991). Self-regulatory mechanisms governing the impact of social comparison on complex decision making.
Journal of Personality and Social Psychology, 60, 941-951.
Bandura, A., & Wood, R. (1989). Effects of perceived controllability
and performance standards on self-regulation of complex decision
making. Journal of Personality and Social Psychology, 56, 805-814.
Barling, J., & Beattie, R. (1983). Self-efficacy beliefs and sales performance. Journal of Organizational Behavior Management, 5,41-51.
Berry, J. M., West, R. L., & Dennehey, D. M. (1989). Reliability and
validity of memory self-efficacy questionnaire. Developmental Psychology, 25, 701-713.
Bores-Rangel, E., Church, A. T., Szendre, D., & Reeves, C. (1990). Selfefficacy in relation to occupational consideration and academic performance in high school equivalency students. Journal of Counseling
Psychology, 37, 407-418.
Brown, S. D., Lent, R. W, & Larkin, K. C. (1989). Self-efficacy as a
moderator of scholastic aptitude-academic performance relationships. Journal of Vocational Behavior, 35, 64-75.
Cervone, D., & Palmer, B. W. (1990). Anchoring biases and the perseverance of self-efficacy beliefs. Cognitive Therapy and Research, 14,
401-416.
Clement, S. (1987). The self-efficacy expectations and occupational
preferences of females and males. Journal of Occupational Psychology, 60, 257-265.
Cohen, J., & Cohen, P. (1983). Applied multiple regression/correlation
analysis for the behavioral sciences (2nd ed.). Hillsdale, NJ: Erlbaum.
Frayne, C. A., & Latham, G. P. (1987). Application of social learning
theory to employee self-management of attendance. Journal of Applied Psychology, 72, 387-392.
Gist, M. E., & Mitchell, T. R. (1992). Self-efficacy: A theoretical analysis
of its determinants and malleability. Academy of Management Revim, 17, 183-211.
Gist, M. E., Schwoerer, C. E., & Rosen, B. (1989). Effects of alternative
training methods on self-efficacy and performance in computer software training. Journal of Applied Psychology, 74, 884-891.
Gould, D., Hodge, K., Peterson, K.., & Giannini, J. (1989). An exploratory examination of strategies used by elite coaches to enhance selfefficacy in athletes. Journal of Sport and Exercise Psychology, 11,
128-140.
Holahan, C. K., & Holahan, C. J. (1987). Life stress, hassles, and selfefficacy in aging: A replication and extension. Journal of Applied Social Psychology, 17, 574-592.
Jones, G. R. (1986). Socialization tactics, self-efficacy, and newcomers'
adjustments to organizations. Academy of Management Journal, 29,
262-279.
Kavanagh, D. J., & Bower, G. H. (1985). Mood and self-efficacy: Impact
of joy and sadness on perceived capabilities. Cognitive Therapy and
Research, 9, 507-525.
Kazdin, A. E. (1978). Conceptual and assessment issues raised by selfefficacy theory. Advanced Behavioral Research Therapy, 1, 177-185.
Kerr, N. L. (1989). Illusions of efficacy: The effects of group size on
369