Full text of DOI:10.1207/S15327841Mpee0401_3

Measurement in Physical Education and Exercise Science
ISSN: 1091-367X (Print) 1532-7841 (Online) Journal homepage: http://www.tandfonline.com/loi/hmpe20
Confirmatory Factor Analysis of the Physical Self-
Efficacy Scale With a College-Aged Sample of Men
and Women
Robert W. Motl & David E. Conroy
To cite this article: Robert W. Motl & David E. Conroy (2000) Confirmatory Factor Analysis of
the Physical Self-Efficacy Scale With a College-Aged Sample of Men and Women, Measurement
in Physical Education and Exercise Science, 4:1, 13-27, DOI: 10.1207/S15327841Mpee0401_3
To link to this article: http://dx.doi.org/10.1207/S15327841Mpee0401_3
Published online: 18 Nov 2009.
Submit your article to this journal
Article views: 43
View related articles
Citing articles: 2 View citing articles
Full Terms & Conditions of access and use can be found at
http://www.tandfonline.com/action/journalInformation?journalCode=hmpe20
Download by: [McMaster University]
Date: 09 June 2016, At: 18:53

MEASUREMENT IN PHYSICAL EDUCATION AND EXERCISE SCIENCE, 4(1), 13–27
Copyright © 2000, Lawrence Erlbaum Associates, Inc.
Confirmatory Factor Analysis of the
Physical Self-Efficacy Scale With a
College-Aged Sample of Men and Women
Robert W. Motl
Department of Exercise Science
The University of Georgia
David E. Conroy
Department of Kinesiology
The Pennsylvania State University
The 2-factor model proposed by Ryckman, Robbins, Thornton, and Cantrell (1982)
for describing responses to the Physical Self-Efficacy Scale (PSES) was tested in a
sample of male and female college students (N = 313). The factor structure was tested
using confirmatory factor analysis with maximum likelihood estimation in LISREL
8.20 (Jöreskog & Sörbom, 1996a). The two-factor model for the PSES may not repre-
sent an entirely acceptable fit to the data (χ2[208, N = 313] = 410, p < .0001, root mean
square error of approximation = .06 [90% confidence interval = .05–.07], standard-
ized root mean square residual = .06, Non-Normed Fit Index = .80, Goodness of Fit
Index = .89). Examination of the standardized residuals and modification indexes
(MIs) in the factor-loading matrix identified possible problems with the proposed
2-factor model. The number of large standardized residuals suggested that the 2-fac-
tor model did not accurately predict the relations among several pairs of items, and the
MIs suggested that some items may cross-load or even load on a different factor. Be-
cause the 2-factor model provided by Ryckman et al. (1982) may not represent an en-
tirely satisfactory fit to the PSES responses of male and female college students, fur-
therresearchproposingandtestingvarioussolutionstoPSESresponsesisnecessary.
Key words: Physical Self-Efficacy Scale, confirmatory factor analysis, physical
activity
Requests for reprints should be sent to Robert W. Motl, Department of Exercise Science, The Univer-
sity of Georgia, 300 River Road, Athens, GA 30602. E-mail: rmotl@coe.uga.edu

14
MOTL AND CONROY
Nearly 20 years ago, Ryckman et al. (1982) developed the Physical Self-Efficacy
Scale (PSES) to measure the two constructs of perceived physical ability (PPA) and
physical self-presentational confidence (PSPC). The PPA subscale consists of
items such as “I have excellent reflexes” and “I am not agile and graceful,” and it as-
sesses the level of confidence in performing physical skills. The PSPC subscale
consists of items such as “I am rarely embarrassed by my voice” and “I am never in-
timidated by the thought of a sexual encounter,” and it appraises the level of confi-
dence in displaying physical skills in the presence of others. The PPA and PSPC
subscales are combined to form an overall measure of generalized physical self-ef-
ficacy (PSE).
To develop the PSES, Ryckman et al. (1982) initially generated a pool of 90
items and then administered the pool of items along with a measure of social desir-
ability to 363 undergraduate students. Responses to the 90 items were subjected to
a single exploratory factor analysis (EFA) using principal-axis factor extraction
with an orthogonal varimax rotation to extract an appropriate number of factors
and reduce the item content. Using the eigenvalue greater than 1.0 rule, the EFA
resulted in a three-factor solution, but the third factor was discarded based on an
insufficient number of substantial item loadings (i.e., > .35) and possible contami-
nation by a social-desirability response bias (i.e., r > .20). Items were also elimi-
nated from the first and second factors based on weak factor loadings (i.e., < .35),
cross- or dual-factor loadings (i.e., both > .35), and social-desirability contamina-
tion (i.e., r > .20). This process resulted in a two-factor model consisting of 22
items with 10 items loading on the PPA factor and 12 items loading on the PSPC
factor. Ryckman et al. (1982) then reported that composite scores from the two
factors were correlated (r = .26) and could be summed to yield an overall PSE
score. Conceptually, the two-factor correlated structure is analogous to a single
higher order PSE factor with two subfactors of PPA and PSPC as in a second-order
factor analysis. The two-factor model and the overall composite PSES score were
then supported by estimates of internal consistency in the initial sample and a sec-
ond, independent sample of 83 undergraduate students (Ryckman et al., 1982), but
the PSES was not further factor analyzed in the initial validation study.
The factor structure of the PSES has been examined in only one other published
study. Using confirmatory factor analysis (CFA), McAuley and Gill (1983) tested
the fidelity of the two-factor correlated model for describing responses of female,
collegiate gymnasts to the PSES. The CFA yielded an acceptable chi-square:de-
grees of freedom ratio (χ2:df = 1.53), but a poor Goodness of Fit Index (GFI) value
of .65. However, chi-square values are positively related to sample size (Bollen,
1989; Jöreskog & Sörbom, 1996a), and the seemingly acceptable chi-square:de-
grees of freedom ratio may be attributable to a small sample (N = 52). McAuley
and Gill (1983) reported coefficient alpha values of .76 and .42 for the PPA and
PSPC subscales and an alpha of .72 for the overall PSE score. The coefficient al-
pha value of .42 for the PSPC subscale was not acceptable according to the con-

PHYSICAL SELF-EFFICACY SCALE
15
ventional psychometric standard of .70 (DeVellis, 1991; Nunnally, 1978). Other
researchers have also reported internal consistency estimates for the PSPC
subscale that are not acceptable (Martin & Mack, 1996; McAuley & Burman,
1993; Ryckman, Robbins, Thornton, Gold, & Kuehnel, 1985).
Although the external validity of PSES scores has been established (McAuley
& Gill, 1983; Ryckman et al., 1982; Thornton, Ryckman, Robbins, Donolli, &
Biser, 1987), the structural aspects of the PSES have not been thoroughly exam-
ined and are not well supported. Further examination of the structural properties of
the PSES is warranted because scores from this measure may help in understand-
ing exercise and sport behaviors. The PSES has been employed by researchers ex-
amining self-presentational processes (e.g., Martin & Mack, 1996; McAuley &
Burman, 1993), self-esteem and physical acceptance (e.g., Baldwin & Courneya,
1997), and optimism and pessimism (e.g., Kavussannu & McAuley, 1995) in sport
and exercise environments. Other researchers have employed the PSES to exam-
ine its relation to measures of physical activity (e.g., Dishman, Darracott, & Lam-
bert, 1992; Yeung & Hemsley, 1997).
In light of the aforementioned psychometric issues and the significance of these
constructs, CFA was utilized to test the two-factor correlated model reported by
Ryckman et al. (1982) for describing the responses of male and female college-age
students to the PSES. We did not directly test a higher order model to the PSES,
which is statistically indistinguishable from a two-factor correlated model, be-
cause it cannot be identified in covariance modeling according to both the t rule
(i.e., necessary condition) and the three-indicator rule (i.e., sufficient condition)
provided by Bollen (1989). Furthermore, Schutz and Gessaroli (1993) indicated
that the magnitude of the correlation between two first-order factors from a
two-factor correlated model is the correct test of a higher order model consisting of
only two first-order factors, because a second-order factor perfectly explains the
correlation between two first-order factors.
METHODS
Participants
Participants (N = 313) included college students with a mean age of 22.2 years (SD
= 4.0). The sample consisted of 166 men and 146 women; one individual did not re-
port sex. The mean age of the men was 22.7 years (SD = 4.1), and the mean age of
the women was 21.7 years (SD = 3.9). The students were enrolled in a Liberal Edu-
cation lecture class (n = 104) and a variety of physical activity classes: weight train-
ing (n = 82), swimming (n = 61), aerobics (n = 30), aquatone (n = 22), and walking
(n = 14). The participants were recruited to represent college-age and physically ac-
tive groups.

16
MOTL AND CONROY
Instrument
The 22-item PSES was employed as originally developed by Ryckman et al.
(1982). Participants were asked to rate each item on a 6-point Likert scale ranging
from 1 (strongly agree) to 6 (strongly disagree). Items 1, 3, 4, 9, 11, 14, 17, 19, 20,
21, and 22 were reverse scored. The PPA subscale consisted of Items 1, 2, 4, 6, 8,
12, 13, 19, 21, and 22; the PSPC subscale consisted of Items 3, 5, 7, 9, 10, 11, 14, 15,
16, 17, 18, and 20.
Procedure
This study was conducted as part of a larger project examining self-presentational
processes in physical activity, and the procedures were approved by a university in-
stitutional review board. Course instructors were initially contacted to solicit their
cooperation and schedule a date for data collection. On the date of data collection,
instructors introduced the researchers and encouraged students to participate in the
project. The researchers subsequently described the study as “an inquiry into the at-
titudes of exercisers” and provided a brief overview of the methods. Students who
were willing to participate signed an informed consent document and then com-
pleted a demographic questionnaire and a battery of instruments that included the
PSES. The questionnaires were completed prior to class activities.
Data Analysis
The proposed two-factor structure of the PSES was tested using CFA with maxi-
mum likelihood estimation in LISREL 8.20 (Jöreskog & Sörbom, 1996a). Maxi-
mum likelihood was selected because it has resulted in accurate fit indexes with
ordered categorical data of varying degrees of kurtosis and minimally biased fit
indexes with ordered categorical data that were skewed and extremely
leptokurtic (Hutchinson & Olmos, 1998). The factor correlation matrix was
specified as a symmetric matrix such that all parameters were estimated (i.e.,
two factor variances and a single covariance). The factor-loading matrix was
specified to reflect simple structure such that item loadings were estimated for
only one latent factor. The uniqueness matrix was specified as a diagonal matrix
with a single uniqueness estimated for each indicator. The first item on each
subscale was constrained to 1.0 to establish the metric for each latent factor. The
CFA model was overidentified according to the necessary and sufficient guide-
lines of Bollen (1989). Sample size was appropriate to estimate the two-factor
model: (a) sample size was larger than 300, (b) ratio of sample size to number of
free parameters exceeded 5:1 and approximated 10:1, and (c) ratio of observed

PHYSICAL SELF-EFFICACY SCALE
17
variables per latent variable equaled or exceeded 10:1 in combination with a
sample size larger than 300 (Bentler & Chou, 1987; Bollen, 1989; Marsh, Hau,
Balla, & Grayson, 1998; Tanaka, 1987).
Multiple indexes were examined to determine whether the model represented a
reasonable fit to the data. The chi-square statistic is a classic test that assesses exact
fit of the specified model to the data. However, the chi-square value is positively
related to sample size, and the specified model may not be correct, suggesting that
other indexes should be employed to assess model fit (Bentler & Chou, 1987;
Bollen, 1989; Hu & Bentler, 1999; Jöreskog & Sörbom, 1996a; Ullman, 1996).
The root mean square error of approximation (RMSEA) represents closeness of fit,
with values less than .08 and .05 demonstrating reasonable and close fit, respec-
tively, of the model to the data (Browne & Cudeck, 1993; Jackson & Marsh, 1996).
RMSEA values approximating zero demonstrate exact fit of the model. The 90%
confidence interval (CI) around the RMSEA point estimate was calculated using
FITMOD (Browne, 1992), and it should include .05 and approximate zero to dem-
onstrate the possibilities of close and exact fit (Browne & Cudeck, 1993). The GFI
is a standalone index that represents the relative amount of observed variances and
covariances accounted for by the specified model (Bollen, 1989; Hu & Bentler,
1999). The Non-Normed Fit Index (NNFI) measures the proportionate improve-
ment in fit of the specified model compared to a baseline or null model in which all
observed variables are uncorrelated (Benter & Bonett, 1980; Bollen, 1989; Hu &
Bentler, 1999). NNFI values can be above one. Minimally acceptable fit was based
on conventional GFI and NNFI thresholds of .90 (Bentler & Bonett, 1980; Bollen,
1989) with values exceeding .95 indicative of good fit (Hu & Bentler, 1999). The
standardized root mean square residual (SRMR) is the average of the standardized
residuals computed between the obtained and specified elements in the vari-
ance–covariance matrices. The SRMR value should be less than .05 and close to
zero (Bollen, 1989; Ullman, 1996). The standardized residuals and modification
indexes (MIs) in the factor-loading matrix were examined to further characterize
the fit of the model and highlight possible problems with model specification
(Bollen, 1989; Jöreskog & Sörbom, 1996a; Ullman, 1996). The LISREL estimates
of item loadings, standard errors, t values, and squared multiple correlations
(SMCs) were also inspected for appropriate sign and magnitude (Bollen, 1989;
Jöreskog & Sörbom, 1996a; Ullman, 1996).
RESULTS
Descriptive Statistics
Table 1 contains the means, standard deviations, and skewness and kurtosis val-
ues for the 22 items composing the PSES. The mean scores for 15 of the 22

18
MOTL AND CONROY
TABLE 1
Mean Scores, Standard Deviations, and Skewness and Kurtosis Values for the 22 Items
Composing the Physical Self-Efficacy Scale
Variable
M
SD
Skewness
Kurtosis
Item 1
Item 2
Item 3
Item 4
Item 5
Item 6
Item 7
Item 8
4.78
4.35
4.44
4.32
3.70
3.65
4.27
4.49
3.54
4.97
4.76
4.34
4.32
3.47
2.93
4.91
3.42
5.04
3.70
4.09
4.61
4.29
1.09
1.35
1.66
1.26
1.58
1.65
1.71
1.45
1.72
1.33
1.43
1.46
1.56
1.43
1.49
1.41
1.57
1.43
1.67
1.55
1.23
1.41
–1.15
–0.71
–0.82
–0.75
–0.09
–0.21
–0.57
–0.70
–0.02
–1.21
–1.08
–0.73
–0.71
0.02
0.64
–1.16
0.03
–1.48
–0.33
–0.47
–0.86
–0.60
1.45
–0.39
–0.63
0.09
–1.23
–1.17
–1.07
–0.56
–1.36
0.47
Item 9
Item 10
Item 11
Item 12
Item 13
Item 14
Item 15
Item 16
Item 17
Item 18
Item 19
Item 20
Item 21
Item 22
0.10
–0.44
–0.62
–0.89
–0.51
0.36
–1.13
1.12
–1.14
–0.84
0.25
–0.43
items were relatively high (above 4 on a 6-point scale). The skewness values in-
dicated that most items were negatively skewed (M = –0.59, range = –1.48 to
0.64). The kurtosis values did not identify any items as extremely leptokurtic or
platykurtic (M = –0.39, range = –1.36 to 1.45). Examination of multivariate nor-
mality using PRELIS 2.20 (Jöreskog & Sörbom, 1996b) yielded the Mardia
(1970, 1974) coefficient values (normalized estimates) of 16.72 and 12.74 for
skewness and kurtosis. The normalized estimates of skewness and kurtosis are
interpreted similar to Z scores and were significant, suggesting that the data
were not multivariate normal.1 Mardia’s coefficient, however, is sample-size de-
1Because of an apparent lack of multivariate normality, we also performed the CFA utilizing
weighted least squares (WLS) estimation with a polychoric correlation matrix and an asymptotic weight
matrix. The two-factor model did not represent a reasonable fit to the PSES (χ2[208, N = 313] = 843, p <
.0001, RMSEA = .10 [90% CI = .09–.11], SRMR = .11, NNFI = .81, GFI = .91). The WLS estimation
with polychoric correlations and an asymptotic weight matrix is extremely data intensive (Bollen, 1989;
Jöreskog & Sörbom, 1996b), often requiring sample sizes larger than those employed in this study. We
satisfied a very conservative estimate of sample size based on the guideline of k(k – 1)/2, where k is the

PHYSICAL SELF-EFFICACY SCALE
19
pendent (Bollen, 1989; Mardia, 1970, 1974), and it was primarily employed for
descriptive purposes (Ulman, 1996).
Two-Factor Model Analysis
Model fit. The two-factor model may not represent an entirely satisfactory fit
to the PSES (χ2[208, N = 313] = 410, p < .0001, RMSEA = .06 [90% CI = .05–.07],
SRMR = .06, NNFI = .80, GFI = .89). The chi-square value was significant (p <
.0001), suggesting that the two-factor model should be rejected. The RMSEA point
estimate indicated that the model provided a reasonable, but not close or exact fit to
the PSES (Browne & Cudeck, 1993). The 90% CI around the RMSEA point esti-
mate included .05, but it did not approximate zero and suggested that the two-factor
model represents a possibility of close fit but does not represent an exact fit to the
PSES. The SRMR exceeded the acceptable threshold value (i.e., .05). The relative
fit indexes (i.e., NNFI and GFI) were below acceptable threshold values (i.e., .90).
NNFI and GFI values often exceed .90 even with seriously misspecified models
(Bollen, 1989; Hu & Bentler, 1999). The observed NNFI value of .80 was particu-
larly indicative of poor model fit. Rarely is this index below .90 in experimentally
collected data or Monte Carlo simulation data (Hu & Bentler, 1999), and Bentler
and Bonett (1980) reported that “in our experience, models with overall fit indexes
of less than .9 can usually be substantially improved” (p. 600).
Item-level analyses. As seen in Table 2, the LISREL unstandardized pa-
rameter estimates for the items were all significant and of the expected direction.
Interestingly, the t values for the parameter estimates were smaller for items on the
PSPC subscale than the PPA subscale. Items on the PSPC subscale also contained
relatively small SMCs (see Table 2). Only 2 of 12 SMCs were above .20 on the
PSPC subscale; 8 of 10 SMCs were above .20 on the PPA subscale. The small
SMCs reflect a high degree of error variance relative to the total item variance and
suggest that the indicators may not be good measures of PSPC (Bollen, 1989). In-
spection of the standardized residuals identified one possible cause of model
misspecification. The standardized residuals identify pairs of items that were either
over- or underpredicted by the model (Bollen, 1989; Jöreskog & Sörbom, 1996a).
There were 21 standardized residuals that exceeded 2.00 and 14 standardized resid-
uals that were less than –2.00. Seven of the positive standardized residuals ex-
number of observed variables. This guideline only serves as an absolute minimum for estimating the
sample size necessary to employ WLS with polychoric correlations and an asymptotic weight matrix
(Jöreskog & Sörbom, 1996b).

20
MOTL AND CONROY
TABLE 2
LISREL Unstandardized Estimates of Item Loadings, Standard Errors, T Values, and
Squared Multiple Correlations for the PPA and PSPC Items of the PSES Based on
Maximum Likelihood Estimation
Item Loadings
Variable
PPA
PSPC
SE
t
SMC
Item 1
Item 2
Item 4
Item 6
1.00
1.08
1.02
1.06
1.40
1.11
1.21
1.01
1.04
1.34
—
.14
.13
.17
.16
.15
.16
.17
.13
.16
—
.40
.43
.41
.40
.36
.34
.40
.42
.36
.37
.40
—
.36
.28
.29
.19
.41
.25
.26
.16
.31
.38
.07
.15
.14
.12
.23
.14
.11
.18
.22
.10
.15
.16
7.62
7.66
6.41
8.75
7.29
7.40
6.02
7.87
8.51
—
3.45
3.42
3.30
3.69
3.42
3.26
3.55
3.66
3.21
3.44
3.48
Item 8
Item 12
Item 13
Item 19
Item 21
Item 22
Item 3
Item 5
Item 7
Item 9
Item 10
Item 11
Item 14
Item 15
Item 16
Item 17
Item 18
Item 20
1.00
1.38
1.47
1.35
1.48
1.23
1.13
1.42
1.52
1.15
1.26
1.40
Note. Standard errors and t values are not reported for Items 1 and 3 because these items were
constrained to 1.0 to establish the metric for the latent variables. PPA = perceived physical ability; PSPC
= physical self-presentational confidence; PSES = Physical Self-Efficacy Scale; SMC = squared
multiple correlation.
ceeded 3.00, and only 1 of the standardized residuals was less than –3.00. The larg-
est positive standardized residual was 5.60 between Items 15 and 17. Multiple large
residuals were observed for Item 4 (5 exceeding ±2.00), Item 6 (4 exceeding
±2.00), Item 8 (4 exceeding ±2.00), Item 19 (4 exceeding ±2.00), Item 21 (4 ex-
ceeding ±2.00), and Item 22 (6 exceeding ±2.00). The number of residuals that ex-
ceeded 2.00 suggested that the two-factor model did not adequately predict the rela-
tions among multiple pairs of variables (Bollen, 1989; Jöreskog & Sörbom, 1996a)
and that the two-factor model might not adequately fit the data.
The MIs in the factor-loading matrix identified another possible problem with
the model. MIs were provided by LISREL for all parameters constrained to zero
and indicate when an item may cross-load or even load on a different factor. The

PHYSICAL SELF-EFFICACY SCALE
21
largest MI was obtained for Item 8 (MI = 14.90). Two other relatively large MIs
were found for Item 16 (MI = 6.31) and Item 20 (MI = 7.25). Respecifying these
items to cross-load or perhaps load on a different factor may improve the model fit.
The factor-loading matrix containing completely standardized factor loadings
is presented in Table 3. The completely standardized uniquenesses also are re-
ported in Table 3. Inspection of the factor-loading matrix indicated that the factor
loadings were smaller on the PSPC subscale than the PPA subscale. The
uniquenesses were generally larger on the PSPC subscale than on the PPA
subscale. Based on the formula described by Fornell and Larcker (1981), reliabil-
ity coefficients were computed for the PPA and PSPC latent factors with the com-
pletely standardized factor loadings and uniquenesses. The reliability coefficients
enabled a comparison of variance captured by the constructs (i.e., PPA and PSPC)
to variance attributable to measurement error. The formula involved squaring and
TABLE 3
Standardized Estimates of Factor Loadings and Uniquenesses for Each Item on the PPA
and PSPC Subscales of the PSES Based on Maximum Likelihood Estimation
Factor Loadings
Variable
PPA
PSPC
Uniquenesses
Item 1
Item 2
Item 4
Item 6
.60
.53
.54
.43
.64
.50
.51
.40
.56
.62
.64
.72
.71
.81
.59
.75
.74
.84
.69
.62
.93
.85
.86
.88
.77
.86
.89
.82
.78
.90
.85
.84
Item 8
Item 12
Item 13
Item 19
Item 21
Item 22
Item 3
Item 5
Item 7
Item 9
Item 10
Item 11
Item 14
Item 15
Item 16
Item 17
Item 18
Item 20
.26
.38
.38
.34
.48
.38
.33
.42
.47
.32
.38
.40
Note. PPA = perceived physical ability; PSPC = physical self-presentational confidence; PSES =
Physical Self-Efficacy Scale.

22
MOTL AND CONROY
then summing the factor loadings and also summing the unsquared uniquenesses
for items on the PPA and PSPC factors. The reliability coefficients were then com-
puted as the sum of squared factor loadings divided by the sum of squared factor
loadings plus the sum of unsquared uniquenesses. The computed reliability coeffi-
cients were .29 and .15 for PPA and PSPC, respectively. These values were less
than .50 and indicated that variance attributable to measurement error was larger
than variance captured by the PPA and PSPC constructs; this questions the validity
of the PSES items and the PPA and PSPC constructs (Fornell & Larcker, 1981).
The two factors were highly correlated (r = .73), and the interfactor correlation
was larger than the value (r = .26) reported by Ryckman et al. (1982). However,
Ryckman et al. (1982) computed an interfactor correlation after forcing factors
into an orthogonal structure, which is not a recommended statistical procedure
(Comrey & Lee, 1992; Gorsuch, 1983; Schutz & Gessaroli, 1993) and may have
attenuated the interfactor correlation. The interfactor correlation also is
disattenuated (i.e., measurement error is removed from the correlation) in CFA,
and it may appear larger than values obtained in traditional EFA depending on the
extent of measurement error. The large interfactor correlation between PPA and
PSPC also supported the notion of a higher order structure to the PSES (Schutz &
Gessaroli, 1993). As previously noted, however, the higher order model was not
estimated based on identification problems (i.e., t rule and three-indicator rule;
Bollen, 1989) and the correct test of a second-order model with only two
subfactors being the magnitude of the interfactor correlation between the two
subfactors (Schutz & Gessaroli, 1993). Overall, the absolute and relative fit in-
dexes, standardized residuals, and reliability coefficients (i.e., average variance
extracted) suggested that the two-factor model does not represent an entirely ac-
ceptable solution to the PSES in college-age men and women.2
One-Factor Versus Two-Factor Comparison
The high interfactor correlation between PPA and PSPC suggested that one factor
may represent a plausible solution to the PSES. Accordingly, the fit of the one-fac-
2Based on a reviewer’s advice, the factorial invariance of the two-factor model across sex was tested
with the routine outlined by Jöreskog and Sörbom (1996a). The routine involved initial CFAs on the
men’sandwomen’s data separatelyandthentestsofmultigroupequivalence of the variance–covariance
underlying item responses, factor structure, factor loadings, factor variances and covariance, and
uniquenesses. The CFAs provided little evidence of factorial invariance. The two-factor model did not
fit in separate analyses of men (χ2[208, N = 166] = 326, p < .0001, RMSEA = .06 [90% CI = .05–.07],
SRMR = .07, NNFI = .75, GFI = .85) or women (χ2[208, N = 146] = 369, p < .0001, RMSEA = .07 [90%
CI = .06–.08], SRMR = .09, NNFI = .62, GFI = .81). The variance–covariance underlying item re-
sponsesdemonstratedsomepossibleevidenceofinvariance(χ2[253, N=312]=338, p<.0005, RMSEA
= .04 [90% CI = .02–.05], SRMR = .07, NNFI = .84, GFI = .91). There was no support for multigroup

PHYSICAL SELF-EFFICACY SCALE
23
tor model was tested with CFA. The one-factor model also did not represent an ac-
ceptable fit to the PSES (χ2[209, N = 313] = 460, RMSEA = .07 [90% CI = .06–.08],
SRMR = .07, NNFI = .75, GFI = .88). With a chi-square difference test for nested
models, the two-factor solution represented a significantly better fit to the PSES
than the one-factor model (χ2diff[1, N = 313] = 50, p < .0001). This finding suggested
that two or more factors underlie responses to the PSES.
Internal Consistency
The internal consistency of the PSES and its subscales were estimated via coeffi-
cient alpha (Cronbach, 1951). The estimate of internal consistency for the PSE total
score was .79. The estimates of internal consistency for the PPA and PSPC
subscales were .78 and .63, respectively.
DISCUSSION
The two-factor solution reported by Ryckman et al. (1982) for describing the re-
sponses of male and female college-age students to the PSES was tested with CFA.
The two-factor model did not result in an entirely satisfactory fit to the PSES as il-
lustrated by absolute and relative fit indexes. The standardized residuals also sug-
gested that the two-factor model did not represent an acceptable fit. The results did
not support the conclusion of McAuley and Gill (1983), who suggested that the
PSES possessed factorial validity in a small sample of female collegiate gymnasts.
Based on the results of this study, the two-factor solution to the PSES may not dis-
play adequate factorial validity in samples of college-age men and women, and it
may not adequately represent the constructs involved in physical self-perception.
Several explanations might be offered for the questionable structural integrity
of the PSES. One explanation is that more than two factors may underlie responses
to the PSES. Although the two-factor model fit the data significantly better than a
invariance of factor structure (χ2[416, N = 312] = 695, p < .0001, RMSEA = .07 [90% CI = .06–.08],
SRMR = .09, NNFI = .69, GFI = .81), factor loadings (χ2[436, N = 312] = 711, p < .0001, RMSEA = .06
[90% CI = .05–.07], SRMR = .09, NNFI = .71, GFI = .81), factor variances and covariance (χ2[439, N =
312] = 713, p < .0001, RMSEA = .06 [90% CI = .05–.07], SRMR = .09, NNFI = .71, GFI = .81), or
uniquenesses (χ2[461, N = 312] = 753, p < .0001, RMSEA = .06 [90% CI = .05–.07], SRMR = .10, NNFI
= .71, GFI = .80). The invariance routine further suggested that the two-factor model was not acceptable
because the test of invariant factor structure (i.e., location of fixed, freed, and constrained parameters)
resulted in a larger chi-square value and worse fit indexes than the test of equivalent vari-
ance–covariance underlying item responses. The application of structure should result in an improved
fit rather than a worsened fit to the data (Jöreskog & Sörbom, 1996a).

24
MOTL AND CONROY
one-factor solution, the large number of standardized residuals suggested that the
two-factor model did not adequately estimate the relations between multiple pairs
of indicators (Bollen, 1989). Perhaps a three-factor solution, such as the one ini-
tially extracted by Ryckman et al. (1982) that consisted of PPA, PSPC, and a phys-
ical appearance dimension, may provide a better fit to the PSES. Another
explanation may be incorrect item-to-factor specification. As indicated by the MIs
and interfactor correlation, some items may cross-load (i.e., load on two or more
factors). Based on a post hoc inspection of item content stimulated by relatively
large MIs, some items even may be incorrectly specified to load on the a priori fac-
tor identified in the initial EFA performed by Ryckman et al. (1982). For example,
the content of Item 8 (i.e., “I don’t feel in control when I take tests involving physi-
cal dexterity”) may relate to both PSPC and PPA. The content of Item 20 (i.e., “I
find that I am not accident prone”) may relate more to PPA than PSPC. Items 8 and
20 both had relatively large MIs in the factor-loading matrix. The post hoc inspec-
tion of item content in combination with empirical findings indicated a potential
flaw in the design of the instrument.
The initial EFA procedure and factor-analytic solution may be the underlying
cause of the questionable structural integrity of the PSES. More specifically, the
questionable structural integrity may stem for the original EFA procedure, in
which items on the third factor in the three-factor solution and some items on the
PPA and PSPC factors were discarded, but the remaining items were not further
factor analyzed. Submitting the items to another EFA may have indicated that two
or more factors may underlie PSES responses and perhaps that some items may
load on a different factor, cross-load, or not load strongly on any factor. To test the
first possibility, future researchers should extract multiple solutions from the
PSES responses via EFA procedures and then test the fit of the solutions as nested
models using CFA in calibration and cross-validation samples. Testing the second
possibility may involve respecifying the structure of the PSES according to EFA
or MIs in the factor-loading matrix from CFA and then testing the model using
CFA in calibration and cross-validation samples. Another problem involves the
orthogonal factor rotation because correlation analyses were performed between
composite scores from the two orthogonal factors to determine the extent of
interfactor correlation and implicate a higher order model. This is not a recom-
mended procedure; factors forced into an orthogonal configuration should not be
employed to form a higher order, “global” PSE measure (Schutz & Gessaroli,
1993). Future researchers interested in examining and substantiating the structural
validity of the PSES may benefit from consulting the excellent discussions and ex-
amples of EFA and CFA provided by experts in factor-analytic methodology (e.g.,
Bollen, 1989; Comrey & Lee, 1992, Gorsuch, 1983; Jöreskog & Sörbom, 1996a;
Schutz & Gessaroli, 1993; Ullman, 1996).
The problems with the PSES also may be specific to the PSPC subscale rather
than the PPA subscale. This observation is based on the relatively small t values,

PHYSICAL SELF-EFFICACY SCALE
25
SMCs, and completely standardized loadings for PSPC items compared to the
PPA subscale. The estimate of internal consistency for the PSPC also was not ac-
ceptable according to the conventional psychometric criteria of .70 (DeVellis,
1991; Nunnally, 1978). Other researchers have also reported internal consistency
estimates for the PSPC subscale that were not acceptable (Baldwin & Courneya,
1997; Martin & Mack, 1996; McAuley & Burman, 1993; McAuley & Gill, 1983;
Ryckman et al., 1985).
Several explanations might be offered for the poor performance of the PSPC
subscale. The content of the PSPC items may not be appropriate for some popula-
tions (e.g., McAuley & Burman, 1993) or salient in exercise and sport environ-
ments. For example, Item 9 (i.e., “I am never intimidated by the thought of a sexual
encounter”) on the PSPC subscale pertains to self-presentational confidence dur-
ing sexual encounters. Many of the PSPC items also refer to general self-presenta-
tional confidence (i.e., laughing [Item 3] and talking [Item 18]) rather than
physical activity. Responses to these general self-presentation confidence items
may be quite variable and not pertinent to the study of behavior in physical activity
environments. Perhaps some of the PSES items could be refined and rewritten to
enhance the content validity of the PSES. Other items also could be generated to
enhance content validity. In light of the apparent limitations of the content validity
of the PSES, it may be profitable to test the item-content relevance of the original
PSES items and all newly developed and revised items according to the procedures
recently highlighted by Dunn, Bouffard, and Rogers (1999). Future researchers in-
terested in enhancing the content validity of the PSES items may benefit by em-
ploying these recommendations and also consulting the scale development
guidelines outlined by DeVellis (1991).
In summary, the two-factor model reported by Ryckman et al. (1982) does not
seem to represent an entirely acceptable fit to the PSES responses of college-age
men and women. Given the significance of PSES scores in understanding physical
activity behaviors, further research testing the factorial validity of the PSES is
warranted. The most fruitful line of research involves proposing and testing vari-
ous models to determine the latent variables underlying responses to the PSES. It
also may be profitable to refine existing PSES items and generate new items to en-
hance content validity. Once a structurally sound model of PSES scores has been
identified, further examination of the construct validity of PSES scores will be
necessary. PSES scores have been correlated with measures of social anxiety
(Martin & Mack, 1996; McAuley & Burman, 1993; Ryckman et al., 1982), physi-
cal acceptance and global self-esteem (Baldwin & Courneya, 1997), optimism and
pessimism (Kavussannu & McAuley, 1995), and self-reported levels of physical
activity (Baldwin & Courneya, 1997; Dishman et al., 1992; Kavussannu &
McAuley, 1995). However, PSES scores were not related to an objective electro-
mechanical measure of physical activity (i.e., movement counts using a Caltrac®
motion sensor) in the study by Dishman et al. (1992). The conflicting results of

26
MOTL AND CONROY
Dishman et al. (1992) illustrate the need for additional research examining the ex-
ternal validity of PSES scores in physical activity environments. Research estab-
lishing the factorial and external validity of the PSES will further highlight the
many possible roles of this measure in understanding behavior in exercise and
sport environments.
ACKNOWLEDGMENTS
We would like to express appreciation to Dr. Rod K. Dishman, Ms. Erica M. Jack-
son, and three anonymous reviewers for providing helpful comments and sugges-
tions on previous versions of this manuscript.
REFERENCES
Baldwin, M. K., & Courneya, K. S. (1997). Exercise and self-esteem in breast cancer survivors: An ap-
plication of the exercise and self-esteem model. Journal of Sport and Exercise Psychology, 19,
347–358.
Bentler, P. M., & Bonett, D. G. (1980). Significance tests and goodness of fit in the analysis of
covariance structures. Psychological Bulletin, 88, 588–606.
Bentler, P. M., & Chou, C. (1987). Practical issues in structural modeling. Sociological Methods and Re-
search, 16, 78–117.
Bollen, K. A. (1989). Structural equations with latent variables. New York: Wiley.
Browne, M. W. (1992). FITMOD: A computer program for calculating point and interval estimates of
fit of a model. Unpublished manuscript.
Browne, M. W., & Cudeck, R. (1993). Alternative ways of assessing model fit. In K. A. Bollen & J. S.
Long (Eds.), Testing structural equation models (pp. 136–162). Newbury Park, CA: Sage.
Comrey, A. L., & Lee, H. B. (1992). A first course in factor analysis (2nd ed.). Hillsdale, NJ: Lawrence
Erlbaum Associates, Inc.
Cronbach, L. J. (1951). Coefficientalphaandinternalstructureoftests. Psychometrika, 16, 297–334.
DeVellis, R. F. (1991). Scale development: Theory and applications. Newbury Park, CA: Sage.
Dishman, R. K., Darracott, C. R., & Lambert, L. T. (1992). Failure to generalize determinants of self-re-
ported physical activity to a motion sensor. Medicine and Science in Sports and Exercise, 24,
904–910.
Dunn, J. G. H., Bouffard, M., & Rogers, W. T. (1999). Assessing item content-relevance in sport psy-
chology scale-construction research: Issues and recommendations. Measurement in Physical Edu-
cation and Exercise Science, 3, 15–36.
Fornell, C., & Larcker, D. F. (1981). Evaluating structural equation models with unobserved variables
and measurement error. Journal of Marketing Research, 18, 39–50.
Gorsuch, R. L. (1983). Factor analysis (2nd ed.). Hillsdale, NJ: Lawrence Erlbaum Associates, Inc.
Hu, L., & Bentler, P. M. (1999). Cutoff criteria for fit indices in covariance structure analysis: Conven-
tional criteria versus new alternatives. Structural Equation Modeling, 6, 1–55.
Hutchinson, S. R., & Olmos, A. (1998). Behavior of descriptive fit indices in confirmatory factor analy-
sis using ordered categorical data. Structural Equation Modeling, 5, 344–364.
Jackson, S. A., & Marsh, H. W. (1996). Development and validation of a scale to measure optimal expe-
rience: The Flow State Scale. Journal of Sport and Exercise Psychology, 18, 17–35.

PHYSICAL SELF-EFFICACY SCALE
27
Jöreskog, K. G., & Sörbom, D. (1996a). LISREL®8: User’s reference guide. Chicago: Scientific Soft-
ware International, Inc.
Jöreskog, K. G., & Sörbom, D. (1996b). PRELIS™ 2: User’s reference guide. Chicago Scientific Soft-
ware International.
Kavussannu, M., & McAuley, E. (1995). Exercise and optimism: Are highly active individuals more op-
timistic? Journal of Sport and Exercise Psychology, 17, 246–258.
Mardia, K. V. (1970). Measures of multivariate skewness and kurtosis with application. Biometrika, 57,
519–530.
Mardia, K. V. (1974). Applications of some measures of multivariate skewness and kurtosis in testing
normality and robustness studies. Sankhya B, 36, 115–128.
Marsh, H. W., Hau, K. T., Balla, J. R., & Grayson, D. (1998). Is more ever too much? The number of in-
dicatorsperfactorinconfirmatoryfactoranalysis. MultivariateBehavioralResearch, 33, 181–220.
Martin, K. A., & Mack, D. (1996). Relationships between self-presentation and sport competition trait
anxiety: A preliminary study. Journal of Sport and Exercise Psychology, 18, 75–82.
McAuley, E., & Burman, G. (1993). The Social Physique Anxiety Scale: Construct validity in adoles-
cent females. Medicine and Science in Sports and Exercise, 25, 1049–1053.
McAuley, E., & Gill, D. (1983). Reliability and validity of the Physical Self-Efficacy Scale in a competi-
tive sport setting. Journal of Sport Psychology, 5, 410–418.
Nunnally, J. C. (1978). Psychometric theory (2nd ed.). New York: McGraw-Hill.
Ryckman, R. M., Robbins, M. A., Thornton, B., & Cantrell, P. (1982). Development and validation of a
Physical Self-Efficacy Scale. Journal of Personality and Social Psychology, 42, 891–900.
Ryckman, R. M., Robbins, M. A., Thornton, B., Gold, J. A., & Kuehnel, R. H. (1985). Physical self-effi-
cacy and actualization. Journal of Research in Personality, 19, 288–298.
Schutz, R. W., & Gessaroli, M. E. (1993). Use, misuse, and disuse of psychometrics in sport psychology
research. In R. N. Singer, M. Murphey, & L. K. Tennant (Eds.), Handbook of research on sport psy-
chology (pp. 901–917). New York: Macmillan.
Tanaka, J. S. (1987). “How big is big enough?”: Sample size and goodness of fit in structural equation
models with latent variables. Child Development, 58, 134–146.
Thornton, B., Ryckman, R. M., Robbins, M. A., Donolli, J., & Biser, G. (1987). Relationship between
perceived physical ability and indices of actual physical fitness. Journal of Sport Psychology, 9,
295–300.
Ullman, J. B. (1996). Structural equation modeling. In B. G. Tabachnick & L. S. Fidell (Eds.), Using
multivariate statistics (3rd ed., pp. 709–811). New York: HarperCollins.
Yeung, R. R., & Hemsley, D. R. (1997). Exercise behavior in an aerobics class: The impact of personal-
ity traits and efficacy cognitions. Personality and Individual Differences, 23, 425–431.