Beruflich Dokumente
Kultur Dokumente
Expanding on Shavelson et als theoretical work, Marsh (1992; Marsh & ONeill, 1984; Marsh, Ellis,
Parada, Richards & Heubeck, 2005) developed the Self Description Questionnaires to measure the academic
self concepts. These instruments proved to be psychometrically robust (Marsh, 1992; Marsh & ONeill,
1984; Marsh et al, 2005). Marshs research into academic self-concept provided support for the Shavelson et
al model, however, it also revealed that the domains or factors within the model are quite discreet from one
another lending stronger support for the multidimensionality of the model rather than its hierarchical
structure. These findings gave rise to a revision of the model (Marsh & Shavelson, 1985; Marsh, Byrne &
Shavelson, 1988) and Marshs (1990a, 1990b, 1993a) recommendations that researchers should measure
self-concepts specifically related to the issues under investigation, as well as general self-concept.
Accordingly, Marsh has developed a raft of instruments to measure each of the domains encompassed within
the multidimensional model self-concept, which includes the Physical Self Description Questionnaire
developed to measure Physical Self-Concept. (Marsh, 1996a, 1996b)
The Physical Self Description Questionnaire (PSDQ)
The PSDQ is comprised of 9 factors or scales specific to physical self-concept Activity, Appearance, Body
Fat, Coordination, Endurance, Flexibility, Health, Sport, Strength, and two global scales Global Physical
and Global Esteem. PSDQ reflects some SDQ Scales including Physical Ability, Physical Appearance and
Esteem, and recognised physical fitness components (Marsh, 1993b). Each scale includes 6 or 8 items; each
item is presented as a declarative statement to which subjects may respond using a 6-point true-false scale
(see Marsh et al, 1994 for the complete instrument).
The PSDQ was developed for adolescents however, research across a wide variety of Australian samples
confirms its appropriateness for wider application, for example, the PSDQ has been demonstrated as
appropriate for adolescent elite and non-elite athletes (Marsh, 1998 cited in Marsh, 2002), and in crosscultural settings with adult subjects (Marsh, Asci & Tomas, 2002; Marsh, Marco & Asci, in press cited in
Marsh, 2002)
Both within-network investigations (those focusing on the internal structure of the construct) and betweennetwork investigations (those exploring the relationship between the construct and other related external
criteria) have been conducted on the PSDQ. This is in keeping with the construct validity approach where
theory, measurement and empirical research are form a feedback loop furthering the understanding and
development of the construct of physical self-concept. (Marsh, 2002)
The PSDQ has been demonstrated to have good reliability (Marsh, 1996a; Marsh, Richards, Johnson, Roche
& Tremayne et al, 1994), a well defined, replicable factor structure as shown by confirmatory factor analysis
(CFA) (Marsh, 1996b; Marsh et al, 1994), a factor structure that is invariant over gender as shown by
multiple group CFA (Marsh et al, 1994), good test-retest stability over both the short and longer term
(Marsh, 1996b), convergent and discriminant validity as shown by an MTMM study of responses to three
Physical Self-concept instruments (Marsh et al, 1994).
The construct validity of PSDQ has also been confirmed in its correlations with related external criteria. In a
study involving 192 high school students, as predicted, all 23 external measures selected including physical
activity, body composition, strength, flexibility, were found to be significantly related to one of the PSDQ
scales (Marsh, 1996a). The SDQ Physical Ability self-concept has also been shown to be significantly
related to physical activity levels, sport participation, physical fitness and body mass and SDQ Physical
Appearance to body mass only (Marsh & Jackson, 1986; Marsh & Peart, 1988). These results support a
multidimensional perspective of self-concept, confirm the discreetness of the various domains such as
Physical Self-Concept and the need for domain specific rather than omnibus instruments to measure them.
The Present Investigation
The PSDQ has been proven to be psychometrically robust instrument with potential usefulness across a wide
variety of physical education, sport and exercise settings. A short version of the PSDQ (PSDQ-S) has been
developed in order to accommodate the needs of researchers who may administer the PSDQ as one of a
battery of tests, and thus require instruments that are as concise as possible without sacrificing reliability or
validity.
Smith, McCarthy and Anderson (2000) developed guidelines for the evaluation of short form instruments in
response to the lack of rigorous methodologies applied to their development. Key areas of concern include
the overestimation of the psychometric properties of instruments of this type and issue of loss of
psychometric strength in the trade-off for shortened length.
In response to Smith et als recommendations, in this investigation, we specifically address the following
questions : a) do the psychometric properties of the items PSDQ-S generalise to other samples; and b) are the
original factors in the long form of the PSDQ and corresponding factors on the PSDQ-S equivalent.
Method
Sample 1 (Normative Archive Sample)
The normative archive sample was based on responses by 986 high school students from the Sydney
metropolitan area who completed the long (70 item) version of the PSDQ. The questionnaires were
distributed during class time and instructions delivered verbally by a researcher and teachers who had been
previously briefed.
Sample 2 (Cross Validation Sample)
The cross validation sample was based on responses by 158 individuals from the general community in the
Hunter Region, NSW, who completed the short (47 item) version of the PSDQ (PSDQ-S). Participants
included members of local sporting clubs, community groups and employees of a health insurance company.
The age of participants ranged from 18 to 79, 87% were females. Questionnaires including full instructions
were distributed to the organisations involved. Representatives of the organizations were also briefed by a
researcher to facilitate administration of the survey. Completed surveys were collected from the
organizations by the researcher or mailed directly to the SELF Research Centre, UWS.
Criteria Applied in the Selection of Items for the PSDQ-S
The goals of the item selection procedure were to a) substantially reduce the length of the PSDQ, b) measure
and preserve the content of all 11 factors on the PSDQ-S, c) retain a minimum of 4 items per scale (original
scales had up to 8 items), d) maintain reliability estimates of at least .80, e) provide a factor structure in
which goodness of fit indexes meet acceptable standards, and f) retain the original context of each of the 11
PSDQ factors.
Selection of the items for the PSDQ-S (see Table 1), were made with reference to the following guidelines:
1. Corrected item-total correlations and the size of standardised factor loadings in CFA were used to
infer items that best measured the intended construct.
2. Items that had minimal cross-loadings as evidenced by LISRELs modification indexes, indicating
the extent to which the fit would be improved if an item were allowed to load on a factor other than
the one that it was intended to measure and the expected size of the cross loading.
3. Items that had minimal correlated uniqueness, particularly with other items in the same scale.
4. The number of times an item was left blank.
5. Subjective evaluations of the content of each item in order to maintain the breadth of content of the
original construct.
6. Sufficient items in each scale to maintain a coefficient alpha estimate of reliability of .80.
Scale
Item Number
Action
Appearance
3, 14, 20, 35
Body Fat
Coordination
1, 9, 18, 23, 31
Endurance
Flexibility
5, 16, 22, 43
I usually catch whatever illness (flu, virus, cold etc.) is going around.
Health
Sport
2, 12, 26, 41
10
Strength
4, 15, 21, 36
11
Global Physical
12
Global Esteem
13
14
15
16
My body is flexible.
17
18
19
20
21
22
23
24
I do physically active things (e.g. jog, dance, bicycle, aerobics, gym, swim) at least three times a week).
25
I am overweight.
26
27
28
29
Overall, I am no good.
30
31
32
33
34
35
I am good looking.
36
37
I can be physically active for a long period of time without getting tired.
38
39
40
41
42
43
44
I am good at endurance activities e.g. distance run, aerobics, swim, cross-country, ski.
45
46
I have to go to the doctor because of illness more than most people my age.
47
Statistical Analysis
A new methodological approach to the evaluation of short forms involving Confirmatory Factor Analysis
(CFA) and Structural Equation Modeling (SEM) implemented by Marsh, Ellis, Parada, Richards & Heubeck
(2005) in the development of the SDQII-S, has been adopted in the present investigation to evaluate the
psychometric robustness of the PSDQ-S. For detailed coverage of CFA as applied in the testing of the PSDQ
long version, please refer to Marsh, Richards, Johnson, Roche and Tremayne (1994).
The main questions addressed here are: a) whether the psychometric properties of the items on the PSDQ-S
generalise to other samples i.e. whether the 47 selected items are invariant for the normative archive group
that completed all 70 items on the long version of the PSDQ and the cross validation group that completed
on the 47 items on the PSDQ-S; and b) if the original factors in the long form of the PSDQ and
corresponding factors on the PSDQ-S are equivalent.
Coefficient alpha estimates were computed to determine the internal reliability of the scales on the PSDQ
long version (70 items), the 47 items selected for the PSDQ-S on the long version, and the PSDQ-S.
SEM and CFAs were conducted with LISREL (Version 8.54) using maximum likelihood estimation.
Goodness of fit was evaluated using the root-mean-square error of approximation (RMSEA), the TuckerLewis Index (TLI) and the Relative Non-centrality Index (RNI) as well as the normal theory chi-square test
statistic and an evaluation of parameter estimates. RMSEA values of less than .05 and .08 are considered to
reflect a close fit and a reasonable fit, respectively. The TLI and RNI values of greater than .90 and .95 are
considered to reflect acceptable and excellent fits to the data. The TLI and RMSEA contain penalties for
parsimony. No missing data was found in the responses. For full descriptions of SEM and CFA
methodologies, please refer to Marsh et al (2005) and Marsh et al (1994)
Results of Study 1: Cross-Validation of Psychometric Properties of Items Selected for the PSDQ-S
Table 2. Descriptive Statistics for Long (70) and Short (47) Items from the Physical Self Description
Questionnaire (PSDQ) Based on Responses from the Normative Archive Group who Completed the Long
Version and the Cross-validation Group who Completed the Short Version
Cross-Validation Group
Normative Archive
Normative Archive
n=156; 47 items
n=986; 70 items
n=986; 47 items
Short Form (47 items)
Short Form (51 items)
Long Form (70 items)
Variable
No.
Rel
Mean SD
Rel
Mean
SD
No.
Rel
Mean SD
Activity
4
.89
3.24
1.49
.82
4.26**
1.36
6
.86
4.22
1.30
Appearance
4
.88
3.71
1.16
.90
3.46*
1.34
6
.89
3.73
1.25
Body Fat
4
.90
3.37
1.57
.92
4.54**
1.52
6
.94
4.61
1.47
Coordination
5
.89
4.00
1.16
.81
4.37**
1.05
6
.84
4.34
1.04
Endurance
4
.89
2.66
1.24
.88
3.79**
1.43
6
.90
3.65
1.37
Flexibility
4
.88
3.83
1.26
.80
4.08*
1.20
6
.85
4.20
1.14
Health
5
.88
5.01
1.08
.81
5.00
1.04
6
.82
4.73
.98
Sport
4
.94
3.28
1.40
.91
4.26**
1.35
6
.93
4.18
1.29
Strength
4
.92
3.36
1.27
.88
4.00**
1.25
6
.89
4.18
1.16
Global Physical
4
.90
3.68
1.21
.91
4.50**
1.33
6
.93
4.44
1.29
Global Esteem
5
.77
4.41
.97
.80
4.69**
1.01
6
.86
4.74
1.00
Mean
.88
3.69
1.25
.86
4.27
1.26
.88
4.27
1.21
Note. No. = number of items; Rel = coefficient estimate of reliability. All responses in the original archive were based
on the long (70 item) version of the PSDQ. Responses items on both forms of the PSDQ varied from 1 to 6. As all 47
items in the short (47) form and the remaining unused items (47 discarded items) of the PSDQ come from the original
set of 70 items, it was possible to do parallel analyses on the long and short, and discarded sets of items. All subjects in
the new cross-validation sample group completed the 47-item version of the instrument and so it is not possible to
compute descriptive statistics based on all 70 items for this group.
* p < .05; ** p < .01 (mean difference between Archive and Cross-validation groups for scales based in items on PSDQ
Short)
Reliability estimates are all above .80 for all the scales on the PSDQ-S, with the exception of Global Esteem,
=.77 , however, this is still acceptable within Smith et als (2000) guidelines (see Discussion). The mean
coefficient alpha across the 11 PSDQ factors is .88 for both the PSDQ-S and PSDQ long version, and .86 for
the 47 items on the long version. These results confirm that each scale on the PSDQ and PSDQ-S is
measuring the factor it is claimed to measure.
Mean self-concepts (except for Appearance and Health) are all lower for cross-validation sample compared
to the Normative short form group. The group mean for physical self-concept for the cross validation group
was 3.69 compared to 4.27 for the normative short form group. However, mean group standard deviations
for these two groups are similar 1.25 and 1.26 respectively.
Results of Study 2: Goodness of Fit
Table3. Summary of Goodness of Fit for Models Based on Responses by Students in the Normative Archive (N =
986) and Cross-validation (N = 158) Groups
Model
DF
TLI
RNI
RMSEA
Description
As shown in Table 3, the a priori model of the normative archive sample (based on the 47 items on the long
form) provides close fit to the data RMSEA .0394. The a priori model of the cross validation sample
(based on the 47 items in the short version of the PSDQ PSDQ-S) provides a reasonable fit to the data
RMSEA .0723.
In the test of multiple group invariance, the fit of the two samples taken across the two groups show that the
factor structure, the correlations and the residuals are invariant when all three elements are constrained to be
equal.
In summary, the pattern of results from the invariance tests demonstrate that the PSDQ (long version) and
the PSDQ-S (short version) are measuring the same factors and confirm that the PSDQ-S is psychometrically
robust.
Discussion
The results of the CFA and SEM conducted on the cross validation sample and the normative archive sample
confirm that the psychometric properties of the items PSDQ-S generalise to other samples; and that the
original factors in the long form of the PSDQ and corresponding factors on the PSDQ-S are equivalent.
These finding are now discussed with reference to Smith et als (2000) guidelines for evaluating short form
instruments.
1. Start with a strong instrument.
The PSDQ long form has been demonstrated to have good reliability (Marsh, 1996a; Marsh,
Richards, Johnson, Roche & Tremayne et al, 1994), a well defined, replicable factor structure as
shown by confirmatory factor analysis (Marsh, 1996b; Marsh et al, 1994), a factor structure that is
invariant over gender as shown by multiple group CFA (Marsh et al, 1994), good test-retest stability
over both the short and longer term (Marsh, 1996b), convergent and discriminant validity as shown
by an MTMM study of responses to three Physical Self-concept instruments (Marsh et al, 1994).
2. Show that the short retains the content coverage of each factor.
The invariance of the factor structure based on the responses to the full 70 item PSDQ by the
normative archive group and the short 47 item PSDQ-S by the cross validation sample demonstrate
that the content coverage of the two instruments is reasonably invariant.
3. Show that each factor on the short form is adequately reliable.
Reliability coefficients of over .80 were obtained for all the scales on the PSDQ short version with
the exception of Global Esteem, which had a reliability coefficient of .77. However, Smith et al
(2000) consider reliability coefficients of over .70 to be adequate.
4. Demonstrate that the short form has adequate overlapping variance with the original Long Form
Based on Independent Administrations.
In this study, the CFA multi-group invariance approach was adopted. Two independent samples
one completing the PSDQ long version and the other the PSDQ short version, represents a more
practical alternative to Smiths recommendations that the long and short version of an instrument be
administered to the same sample. This study provides support for the invariance of the factor
structure across the normative archive and cross-validation samples, hence evidence that the factors
on the short version of the PSDQ are highly similar to those on the long version of the PSDQ.
5. Show that the short form retains the factor structure of the original form.
The results of the CFA multi-group invariance tests confirm this guideline has been met.
6. If appropriate, show that the factors on the short form preserve the content of subdomains or facets
of each factor in the long form.
As the PSDQ-S contains no subdomains, this guideline is not applicable to the present investigation.
7. Show that each factor has validity in an independent sample.
Consistent with the construct validity approach adopted for the PSDQ long version, this guideline
will be addressed as part of an ongoing research program involving the PSDQ-S. This research will
include between-network investigations to test the relationship between the PSDQ-S and external
criteria. This study represents a within-networks investigation and as such is primarily concerned
with the internal construct validity of the instrument. We would also be looking to investigate the
test-retest reliability of this instrument in multi-wave design studies.
8. If appropriate, show that classification rates remain high with the short form.
Neither the PSDQ long nor PSDQ short versions are used to classify subjects into classification
categories therefore this guideline is not applicable to the present investigation.
9. Show that the trade-off in savings of time and resources is acceptable in relation to potential loss of
validity.
Savings of time can only be confirmed in future administrations of the PSDQ-S. The psychometric
analyses conducted in this investigation show that mean reliabilities for the PSDQ long and short
versions are exactly the same - .88, therefore there has been no loss of reliability or validity between
the PSDQ and the PSDQ-S.
References
Carron, A. V., Hausenblaus, H. A. & Estabrooks, P. A. (2003). The Psychology of Physical Activity.McgrawHill Higher Education
Guay, F., Marsh, H. W. and Boivin, M.(2003).Academic self-concept and academic achievement:
Developmental perspectives on their causal ordering. Journal of Educational Psychology, 95(1): 124-136.
Irving, H. M., Adlaf, E. M., Allison, K. R., Paglia, A., Dwyer, J. M. & Goodman, J. (2003). Trends in
physical activity participation among Ontario adolescents, 1997-2001. Canadian Journal of Public Health,
94(4): 272-274.
Marsh, H. W. (2002). A multidimensional physical self-concept: A construct validity approach to theory,
measurement and research. Psychology, 9(4): 459-493.
Marsh, H. W. (1997). The measurement of physical self-concept: A construct validation approach. In K. Fox
(Ed.), The physical self: From motivation to well-being, (pp. 27-58), Champaign, IL: Human Kinetics
Marsh, H. W. (1996b). Construct validity of Physical Self-Description Questionnaire responses: Relations to
external criteria. Journal of Sport and Exercise Psychology, 18(2): 111-131.
Marsh, H. W. (1996a). Physical Self-Description Questionnaire: Stability and discriminant validity.
Research Quarterly for Exercise and Sport, 67: 249-264.
Marsh, H. W. (1993b). Physical fitness self-concept: Relations to field and technical indicators of physical
fitness for boys and girls aged 9-15. Journal of Sport and Exercise Psychology, 15: 184-206.
Marsh, H. W. (1993). Academic self-concept: Theory, measurement and research. Psychological
perspectives on the self, 4: 59-98.
Marsh, H. W. (1992). Self Description Questionnaire (SDQ) II: A theoretical and empirical basis for the
measurement of multiple dimensions of adolescent self-concept. Penrith, New South Wales, Australia:
University of Western Sydney, SELF Research Centre.
Marsh, H. W. (1990b). The structure of academic self-concept: The Marsh/Shavelson model. Journal of
Educational Psychology, 82: 623-636.
Marsh, H. W. (1990a). A multidimensional, hierarchical self-concept: Theoretical and empirical justification.