Professional Documents
Culture Documents
REFERENCES
Linked references are available on JSTOR for this article:
http://www.jstor.org/stable/1434540?seq=1&cid=pdf-reference#references_tab_contents
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at http://www.jstor.org/page/
info/about/policies/terms.jsp
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content
in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship.
For more information about JSTOR, please contact support@jstor.org.
Wiley and National Council on Measurement in Education are collaborating with JSTOR to digitize, preserve and extend
access to Journal of Educational Measurement.
http://www.jstor.org
This content downloaded from 134.176.129.147 on Thu, 26 Nov 2015 23:58:59 UTC
All use subject to JSTOR Terms and Conditions
JOURNAL OF EDUCATIONAL MEASUREMENT
VOLUME 21, NO. 2, SUMMER 1984, pp. 153-174
The authors would like to acknowledge Dr. Fred Orr for his assistance in collecting the data, and Jennifer
Barnes for her assistance in processing the data and her helpful reading of earlier drafts of this manuscript.
153
This content downloaded from 134.176.129.147 on Thu, 26 Nov 2015 23:58:59 UTC
All use subject to JSTOR Terms and Conditions
154 MARSH AND O'NEILL
model, one that incorporatesaspects from most theoretical positions. For purposesof this
study, the most important tenets of the Shavelson model are that self-concept is
multidimensional,that the facets of self-concept become more distinct with age, and that
the facets are hierarchically arranged. The Shavelson model, the proposeddimensions of
self-concept and their structure, are heuristic and plausible, but they were not empirically
demonstrated.
Soares and Soares (1977, 1982) also emphasized the need for within network research,
and described a theoretical model that is similar to the Shavelson model. They, however,
have argued that the low correlations among different facets of self-concept in their
research suggest a model consisting of nearly independent factors of self-concept, not the
strong hierarchical structure proposed by Shavelson. Other researchers (e.g., Cooper-
smith, 1967; Marx & Winne, 1978) have taken the opposite perspective, and argued that
self-concept is so heavily dominated by a general factor that separate facets cannot be
distinguished.
This content downloaded from 134.176.129.147 on Thu, 26 Nov 2015 23:58:59 UTC
All use subject to JSTOR Terms and Conditions
CONSTRUCT VALIDITY OF MULTIDIMENSIONAL SELF-CONCEPT RATINGS 155
This content downloaded from 134.176.129.147 on Thu, 26 Nov 2015 23:58:59 UTC
All use subject to JSTOR Terms and Conditions
156 MARSH AND O'NEILL
constructed and administeredto a new sample of students. The addition of items for these
2 areas and the selection of the most effective of the items for the other 11 areas resulted
in a questionnaire containing 136 items and measuring 13 self-concept facets.
STUDY 1-METHOD
This content downloaded from 134.176.129.147 on Thu, 26 Nov 2015 23:58:59 UTC
All use subject to JSTOR Terms and Conditions
CONSTRUCT VALIDITY OF MULTIDIMENSIONAL SELF-CONCEPT RATINGS 157
from "l-Very Inaccurate" to "9-Very Accurate." The response scales used in the
self-concept and summary description tasks were intentionally varied (i.e., an 8-point
true-false scale as compared to a 9-point accurate-inaccurate scale) in order to disrupt
response biases and make the tasks more distinct.
School Climate Questionnaire. The School Climate Questionnaire (see Flynn, 1975)
was the focus of a part of the study not reported here, but a numberof items were selected
for examination on an a priori basis as being potentially relevant to student self-concept.
These items relate to home environment, academic satisfaction, and academic expecta-
tions. These items were included to determine whether they formed a logical pattern of
relationships with the SDQ III factors.
School Certificate Scores. Near the end of the 10th grade, all students in the state of
New South Wales complete standardized tests in which both their math and their
English/language skills are evaluated. On the basis of test results and school performance
each student is given a separate certificate score for each subject in the range 1 to 5, where
1 represents the top 10%,2 the next 20%, 3 the next 40%, 4 the next 20%, and 5 the last
10%of the 10th grade population in this state. While school performance is a basis on
which students are ranked within each school, the number of students assigned each
certificate score is fixed by performanceon the state-wide tests. This is done in an attempt
to make the scores in different schools comparable. In this study, most of the students had
received scores of 1 or 2 in both mathematics (mean = 2.27, SD = 0.85) and language
(mean = 2.03, SD = 0.83), although there were some 3s and a few 4s and 5s. The skewed
distribution for this sample is due to the large number of students who leave school at the
end of the 10th grade. The language and mathematics scores in this study correlated 0.59
with each other.
Statistical Analysis
Responses to negatively worded items were reversedso that for all items a responseof 8
represented a positive rating of self-concept. A value of 5 (the median of the distribution
of obtained responsesto all items) was then substituted for the few missing responses(less
than 1/4Of 1%)to any SDQ III item. A preliminaryitem analysis of responsesto each scale
indicated that no item in any scale was a negative discriminator;132 of the 136 items had
"corrected item-total correlations" (see Hull & Nie, 1981, pp. 248-267) of at least 0.3.
The alpha coefficients from this analysis appear in Table 2.
For the factor analyses performed in this study, the 10 (or 12) items from each scale
were divided into 5 (or 6) item-pairs, such that the first 2 items in the scale were assigned
to the first pair, the next 2 items to the second, and so forth. Thus, the factor analyses
here, and in Study 2, were performed on responses to 68 item-pairs. This procedure is
preferable to factor analysis responsesto 136 items for the following reasons: (a) the ratio
of the number of subjects to the number of variables (which was still less than desirable in
both studies) is increased; (b) each variable should be more reliable and should have a
smaller unique component; (c) the factor loadings should be less affected by the
idiosyncratic wording of individual items; (d) the cost of the factor analyses will be
substantially reduced; and (e) it becomes possible to use the method of factor analyses in
cases when the number of items exceeds either the limitations of commercially available
factor analysis procedures (e.g., 100 variables for SPSS) or the memory capacity of the
available computers, and one-half the number of items does not. The major disadvantages
of this procedure are that information about individual items is lost, items must be
reasonably homogeneous with respect to the dimension that they were designed to
This content downloaded from 134.176.129.147 on Thu, 26 Nov 2015 23:58:59 UTC
All use subject to JSTOR Terms and Conditions
158 MARSH AND O'NEILL
measure, so that it is important to perform a preliminary item analysis such as the one
described above before this procedureis used, and the parameterestimates and the factor
scores resulting from the factor analysis are likely to vary somewhat, depending upon the
particular pairing of items that is used. However, the factor scores (Nie et al., 1975) that
were generated as part of the factor analysis and used in subsequent analyses were in this
study highly correlated with those based upon another, random pairing of items (median
r = 0.991 for the 13 SDQ III scales). Thus, the particularpairing of items apparently had
little impact in this application.
In both studies, conventional and confirmatory factor analyses were performed. The
former was done using the SPSS program (Nie et al., 1975), with iterated communality
estimates, a Kaiser normalization, and an oblique rotation with delta set to -2. The
confirmatory factor analyses (CFA) were performed with the LISREL V program
(Joreskog & Sorbom, 1981). In the CFA model employed in both studies, it was
hypothesized that: (a) the responsesto the SDQ could be explained by 13 factors, (b) each
item-pair would have a non-zero loading on the factor it was designed to measure and zero
loadings on all other factors, (c) the 13 factors would be correlated, and (d) the
error/uniqueness terms for the measured variables would be uncorrelated. The hypoth-
esized pattern of factor loadings for this model was similar to that expected to result from
a conventional factor analysis, except that in the CFA the factor loadings for each
measured variable were set to zero on all the factors except the one it was designed to
measure. The fit of this CFA model to the data was assessed by the magnitude of
parameter estimates relative to their standard errors of estimate, the ratio of the
chi-square to the degrees of freedom for the analysis, and the root mean square of the
residual differences between the observed and reproduced correlation matrices (see
Joreskog & Sorbom, 1981; Marsh & Hocevar, 1983, in press; Maruyama & McGarvey,
1980). As yet, there are no universally accepted guidelines about how to evaluate
goodness-of-fit, but the most widely used indicator is the chi-square/df ratio. Even here
there is disagreement;some researchersargue that chi-square/df ratios as high as 5 may
indicate a good fit (e.g., Wolfle, 1981), while others claim that a ratio of less than 2 is
required(e.g., Alwin & Jackson, 1981). Nearly all agree, however, that ratios of less than
2 indicate a good fit.
STUDY 1-RESULTS
Factor Analysis of the SDQ III Responses
The conventional factor analysis (see Table 1) identified the 13 SDQ III factors with
remarkable clarity.' Target loadings (i.e., the loadings on the factor that a variable was
'For the conventional factor analysis in Study 1, the 13th eigenvalue was 1.25 and the 14th was 1.03,
indicating that the 13-factor solution is reasonable. Similar factor analyses were also conducted with 10, 11, 12,
14, 15, and 16 rotated factors. For those solutions with fewer than 13 factors, at least one of the SDQ III factors
was not adequately represented in that the variables designed to measure it did not load substantially on any
factor and the error/uniqueness components (i.e., 1 minus the communality) for these variables were higher
than in the 13-factor solution. For solutions with more than 13 factors, there was always at least one empirical
factor that was not adequately defined in that few (or no) variables had substantial loadings on the factor, and
an inspection of those variables that did load highest (i.e., loadings in the 0.2s and 0.3s) did not lead to a
reasonable interpretationof the factor. Also, for solutions with more than 13 factors the median communality of
the measured variables did not differ appreciablyfrom that of the 13-factor solution. For the seven solutions that
were considered (i.e., solutions with 10 to 16 rotated factors), the median communalities were 0.59, 0.60, 0.64,
0.66 (for the 13-factor solution), 0.66, 0.66, and 0.67.
This content downloaded from 134.176.129.147 on Thu, 26 Nov 2015 23:58:59 UTC
All use subject to JSTOR Terms and Conditions
CONSTRUCT VALIDITY OF MULTIDIMENSIONAL SELF-CONCEPT RATINGS 159
Table 1
This content downloaded from 134.176.129.147 on Thu, 26 Nov 2015 23:58:59 UTC
All use subject to JSTOR Terms and Conditions
160 MARSH AND O'NEILL
Table 1 (Continued)
Factor Pattern Correlations
Mathematics 100
Verbal -03 100
Academic 21 26 100
Problem Solving 11 28 22 100
Physical Abilities 05 -03 -06 08 100
Physical Appearance 00 08 08 09 06 100
Same Sex Peers 01 14 06 10 17 09 100
Opposite Sex Peers -09 07 -07 15 13 16 29 100
Relations With Parents 01 06 17 03 05 08 04 02 100
Religion/Spirituality -01 01 16 08 04 02 03 -09 13 100
Honesty/Reliability 00 12 18 07 03 00 01 -01 20 07 100
Emotional Stability 03 07 -06 03 09 07 17 16 07 -05 09 100
General-self -03 20 15 21 06 25 28 23 15 13 07 25 100
Note: All Coefficients are presented without decimal points.
designed to measure) are consistently high (90% are 0.50 or higher, and none is less than
0.35), while nontarget loadings are low (98% are 0.20 or lower, and none is greater than
0.33). Correlations among the 13 oblique factors are small (mean r = 0.08) and
demonstrate that the factors are quite distinct. None of the correlations is greater than
0.37, and, surprisingly, not even the General-Self factor is substantially correlated with
other factors. These findingssuggest that no strong hierarchicalstructureunderlies the 13
SDQ III factors, but instead that there are 13 relatively distinct facets of self-
concept.
The results of the CFA indicate that the 13-factor solution fits the data well. The 68
factor loadings are each large (7 to 20 times the size of their standard errors) and
statistically significant. The ratio of the chi-square to the degrees of freedom (1.68) and
the root mean square residual (0.06) are also small, indicating that the data is well
described by this restrictive factor model.2
Correlations Between SDQ III Factors and Summary Descriptions
After completing the SDQ III, students judged the accuracy as self-descriptionsof 12
summary descriptions designed to characterize 12 of the SDQ III factors (all but the
General-Self factor). Correlations between the 12 self-concept scores (based upon factor
scores derivedfrom the analysis summarizedin Table 1) and the 12 summarydescriptions
(see Table 2) are presented in the form of a multitrait-multimethod matrix. The
application of the Campbell and Fiske (1959) guidelines (see Marsh & Hocevar, 1983)
reveals:
1. Convergencecoefficients (the underlinedvalues in the diagonal of the square matrix
in Table 2) are statistically significant and substantial (mean r = 0.60). This
demonstrates good agreement between the self-concept scores and the summary
descriptions.
2. For every one of the 264 comparisons that can be made between a convergent
2Null models correspondingto the CFA in Study 1, and the two CFAs in Study 2 had chi-square values of
15,332, 10,613, and 11,195, respectively, with 2,278 degrees of freedom. In Study 2, an additional model was
tested where all the parameter estimates (i.e., factor loadings, factor correlations, and error/uniquenesses) for
the self-ratings were required to be the same as the correspondingparameter estimates for the inferred-ratings
(i.e., complete invariance);it resulted in a chi-square/df ratio of 1.89. However, the computer analysis ran out of
time after 60 minutes and cost considerations precluded further analyses. Nevertheless, the partial solution was
reasonable, and the final solution would probably have been similar with a somewhat smaller chi-square value.
Thus, while any interpretationmust be made cautiously, it appears that the parameterestimates in solutions for
the self-ratings and the other-ratings are quite similar and nearly invariant.
This content downloaded from 134.176.129.147 on Thu, 26 Nov 2015 23:58:59 UTC
All use subject to JSTOR Terms and Conditions
Table 2
Self-concepts Summar
Self-concepts Al Bi C1 D1 El Fl G1 Hl Il Jl K1 L1 A2 B2 C2 D2 E
Al Math (93)
Bl Verb -64 (86)
Cl Acad 24 32 (90)
D1 Prob 14 37 27 (77)
El Phys 05 -04 -07 09 (95)
Fl Appr 00 10 09 11 06 (88)
G1 Ssex 00 16 07 13 21 12 (81)
H1 Osex -10 08 -08 19 15 18 34 (93)
Il Prnt 01 07 20 03 05 09 06 -03 (87)
J1 Sprt -01 01 18 10 04 -02 03 -09 15 (89)
K1 Hnst 00 15 24 09 04 00 00 -08 27 11 (75)
L1 Emot 03 10 -06 04 10 09 22 18 09 -07 -12 (88)
SummaryDescriptions
A2 Math 76 05 24 16 04 04 04 -07 02 -01 -06 00 --
B2 Verb 06 47 25 46 03 19 25 12 05 08 01 06 23 --
C2 Acad 29 39 51 32 -02 14 08 00 10 -03 04 -01 46 42 --
D2 Prob 07 29 21 57 01 00 10 21 04 11 -01 02 16 44 33 --
E2 Phys 06 -05 -07 07 79 09 22 12 07 00 02 11 10 07 03 04 -
F2 Appr 04 08 07 09 03 65 21 28 05 -01 -09 07 10 21 15 07 1
G2 Ssex -01 15 17 10 05 08 46 14 08 07 -05 06 02 14 06 16 1
H2 Osex -06 08 -07 24 19 19 37 76 -05 -12 -10 18 -05 17 04 23 1
12 Prnt 02 06 19 08 04 17 10 01 73 17 17 09 05 10 09 09 0
J2 Sprt 14 00 27 15 00 03 -05 -10 17 73 01 06 11 17 13 15 -0
K2 Hnst 19 09 18 17 09 08 12 -01 17 13 33 02 26 23 22 15 1
L2 Emot 05 06 05 18 11 09 24 13 20 22 06 47 11 24 12 27 1
Note: Correlations (presented without decimal points) greater than .12 are statistically s
in parentheses are coefficient alpha estimates of reliability. The underlined values are
This content downloaded from 134.176.129.147 on Thu, 26 Nov 2015 23:58:59 UTC
All use subject to JSTOR Terms and Conditions
162 MARSH AND O'NEILL
coefficient and another coefficient in the same row or the same column of the square
block, the convergent coefficient is higher. The average of the off-diagonal coefficients in
this block is 0.09.
3. For every one of the 264 comparisons between a convergence coefficient and one of
the 11 other coefficients in the same row or the same column of the two triangular blocks,
the convergent coefficient is higher. The average correlation is 0.08 among the
self-concept factors and 0.15 among the summary descriptions.
Not surprisingly, the convergence between self-concept factors and corresponding
summary descriptions is substantial. These findings also demonstrate that responses to
the SDQ III measure distinct and separate components of self-concept.
This content downloaded from 134.176.129.147 on Thu, 26 Nov 2015 23:58:59 UTC
All use subject to JSTOR Terms and Conditions
Table 3
Coefficients of Correlation Among Self-concept Scales and Other Crite
Self-Concept Scale
*
p.(.01
Note: The correlation coefficients are presented without decimal points.
This content downloaded from 134.176.129.147 on Thu, 26 Nov 2015 23:58:59 UTC
All use subject to JSTOR Terms and Conditions
164 MARSH AND O'NEILL
This content downloaded from 134.176.129.147 on Thu, 26 Nov 2015 23:58:59 UTC
All use subject to JSTOR Terms and Conditions
CONSTRUCT VALIDITY OF MULTIDIMENSIONAL SELF-CONCEPT RATINGS 165
analysis of the inferred-ratings. In both factor analyses, the highest factor loading for
each of the 68 item-pairs nearly always occurred for the factor that it was designed to
measure (target loadings), and other loadings (nontarget loadings) were smaller. The size
of the target loadings was large (median = 0.67 for self-ratings, 0.72 for inferred-ratings)
while the size of the nontarget loadings was quite small (median = 0.02 for both sets).
Correlations among the oblique factors were also small, varying between -0.19 and 0.32
(median = 0.07) for self-ratings and varying between -0.14 and 0.33 (median = 0.08)
for ratings by significant others. (The results of these factor analyses are described in
more detail by Marsh, Barnes, & Hocevar, 1983.) These findings replicate the findings in
Study 1, and show in addition that the factor structure underlying the inferred-ratingsis
similar to that underlying the self-ratings.
The results of the confirmatoryfactor analyses indicate that the 13-factor solution fits
the data for both self-ratings and inferred-ratings acceptably well. For self-ratings and
inferred-ratings, the target factor loadings are substantial, while the chi-square/degrees
of freedom ratios (1.78 and 1.59, respectively) and the root mean square residuals (0.07
for both analyses) are small.
DISCUSSION
The first purpose of both studies was to use factor analysis to demonstrate that
responses to the SDQ III define a set of distinct self-concept dimensions. The results of
separate conventional factor analyses of the self-ratings collected in Study 1, and of both
the self-ratings and the inferred-ratings collected in Study 2, clearly identified the 13
SDQ III factors. Confirmatoryfactor analyses demonstrated that the 13-factor solution
This content downloaded from 134.176.129.147 on Thu, 26 Nov 2015 23:58:59 UTC
All use subject to JSTOR Terms and Conditions
Table 4
Other's ratings
01 Math 77 -05 03 03 -08 -06 05 -03 07 05 -09 -06 05 (95)
02 Verb -07 51 31 12 -19 18 -01 06 22 04 -10 08 -06 00 (86)
03 Acad 23 15 31 09 -00 07 -04 -13 -02 14 -06 -11 00 22 32 (88)
04 Prob 08 20 17 52 04 12 -03 -03 -07 -08 -01 18 01 13 37 19 (82)
05 Phys 04 -17 -10 -04 78 09 -01 -11 06 08 -01 18 08 04 00 15 15 (97).
06 Appr 04 -03 -06 10 06 50 05 11 15 07 06 08 19 02 14 -05 17 20 (89)
07 Ssex 01 00 -12 -11 03 06 45 12 22 09 11 27 31 01 11 -01 02 09 23 (90)
08 Osex 01 -02 20 07 12 00 51 08 00 -03 30 20 00 02 -03 19 16 34 37 (9
09 Prnt 07 00 -07 -10 02 -08 i-
17 08 76 14 01 15 06 11 14 10 -01 15 08 32 0
010 Relg 05 -08 00 06 06 -13 04 -08 02 79 09 -09 01 05 -06 11 03 10 04 -02 -0
011 Hnst -07 -05 05 -01 04 00 -04 -02 10 25 44 08 -02 -01 17 24 15 12 15 15 0
012 Emot 02 18 -11 03 11 -01 12 23 24 05 00 62 12 -04 12 01 19 15 19 32 4
013 Genl 07 -08 -00 15 12 15 21 17 15 09 04 22 41 02 13 11 22 16 41 29 3
Note: Correlations (presented without decimal points) greater than .19 are statistically significant (p ( .05). The valu
alpha estimates of reliability. The values in the diagonal of the heterotrait-heteromethod square, the underlined values,
This content downloaded from 134.176.129.147 on Thu, 26 Nov 2015 23:58:59 UTC
All use subject to JSTOR Terms and Conditions
CONSTRUCT VALIDITY OF MULTIDIMENSIONAL SELF-CONCEPT RATINGS 167
was able to fit the data from all three sets of responses acceptably well. Also, for each set
of responses, the correlations among the factors were quite small; this argues that no
strong hierarchical structure underlies the SDQ III factors. The results from Study 2 both
replicate the results from Study 1, and generalize them to ratings by significant
others.
The second purposeof both studies was to demonstrate that the SDQ III factors form a
logically consistent pattern of relationships with external validity criteria. In Study 1, the
most important of these were achievement scores in mathematics and language. These
achievement scores are substantially correlated with self-concepts in the same area, less
correlated with other academic self-concepts, and nearly uncorrelated with nonacademic
self-concepts. Consistent with other research (e.g., Marsh, Smith, Barnes, & Butler,
1983), Verbal and Math self-concepts are nearly uncorrelated with each other even
though the corresponding achievement scores are substantially correlated. The correla-
tions between the SDQ III factors and the other external criteria in Study 1 also form a
pattern of high and low coefficients that is intuitively logical.
In Study 2, the results of the MTMM analysis demonstrate remarkably good
correlations between self-concepts as rated by the subjects themselves and self-concepts
as inferred by the significant others. For example, the self-other agreement in Study 2 was
nearly as strong as agreement between self-ratings and summary descriptions in Study 1.
The inferred-ratings in Study 2 provide a good external validity criterion for each of the
13 SDQ dimensions, hence the results provide evidence of the validity of interpretations
based upon each of the scales. The demonstration of divergent validity again shows that
each of the scales is distinct.
The MTMM results also contradict the Shrauger-Schoeneman conclusion that
self-perceptions and ratings-by-others show little correspondence. Significant self-other
agreement has been found in other research with school children (e.g., the series of
MTMM studies by Marsh and the Soares & Soares 1982 study described earlier).
However, supportfor convergence and divergence in this study is better than in those, and
in other MTMM studies, perhaps for the following reasons: (a) the subjects in this study
were older, (b) both subjects and significant others respondedto the same well-developed
instrument, (c) the self-other agreement shown here was for specific characteristics
rather than the broad, ambiguous characteristics employed in some other studies, and (d)
the significant others in this study knew the subjects they were rating better than did the
external observers in most other research.
The results of these two studies have important implications for further theoretical
work in self-concept research. A few researchers contend that self-concept is unidimen-
sional, while most argue that it is dominated by a general factor and that there is a strong
hierarchical structure, with specific facets of self-concept at the base and a general factor
at the apex of the structure. However, for the present study, neither the unidimensional
nor the strong hierarchical structures were found. Hence, while any significant
correlation among any of the self-concept facets can be used to argue for a hierarchical
structure, the correlations between self-rating factors in both studies and even the
correlations among other-rating factors are so small that the usefulness of such an
interpretation is dubious.
The weakness of the hypothesis that the structure of self-concept is hierarchical is also
demonstrated by correlations involving the General-Self factor. This factor was based
upon the Rosenberg scale, the most widely employed measure of general or overall
This content downloaded from 134.176.129.147 on Thu, 26 Nov 2015 23:58:59 UTC
All use subject to JSTOR Terms and Conditions
168 MARSH AND O'NEILL
self-concept. While the item analyses and factor analyses show that this factor is
well-defined, correlations between it and the other factors were small. This finding argues
that general self-concept has at best a modest relationship to self-concept in specific
areas.
The correlations among the self-concept factors reported here are much smaller than
those typically reported,perhaps for two reasons. First, the subjects in the present studies
were older than those in most other research, and areas of self-concept become more
distinct with age (see Shavelson et al., 1976; Marsh, Barnes, Cairns, & Tidman, in press).
Second, the design and the psychometric propertiesof the SDQ III appear to be superior
to those of the self-concept instruments used in most other research; the SDQ III may be
better able than other instruments to differentiate facets of self-concept.
Two of the most commonly used criteria for validating measures of self-concept are
measures of academic performance (particularly in research with school-aged subjects)
and the observations of an external observer. The literature reviews cited earlier
(Hansford & Hattie, 1982; Shrauger & Schoeneman, 1979) have suggested that both of
these external criteria have weak relationships with self-concept. In contrast to these
earlier findings, the results from both Studies 1 and 2 provide strong support for the
validity of self-concept and the SDQ III as a measure of self-concept. We suspect that the
clear definition of the multiple dimensions of self-concept employed in developing the
SDQ III is the primary reason for its superior performance. Other research has often
relied upon an overall or general measure of self-concept that is the sum of an ill-defined
set of items, or one that resembles the General-Self scale of the SDQ III. However, the
General-Self scale was uncorrelated with? academic achievement in Study 1, and
self-other agreement was lowest for this scale in Study 2. We contend that the relationship
between self-concept and other constructs cannot be adequately understood if the
multidimensionality of self-concept is ignored.
Another importantcontributionof these studies is the introductionof a new instrument
for the measurement of multiple dimensions of self-concept. The SDQ III has a
well-developed factor structure, and measures dimensions that are reliable, valid, and
based upon a strong theoretical model. Thus, the psychometric propertiesof the SDQ III
are stronger than those of other instruments. Reviewers in this field have commented that
one of the most important weaknesses of self-concept research is the poor quality of the
instruments used to measure it. It is hoped that the SDQ III will help remedy this
situation.
This content downloaded from 134.176.129.147 on Thu, 26 Nov 2015 23:58:59 UTC
All use subject to JSTOR Terms and Conditions
CONSTRUCT VALIDITY OF MULTIDIMENSIONAL SELF-CONCEPT RATINGS 169
8. I hardly ever saw things the same way as my parents when I was growing up.
10. I am never able to think up answers to problems that haven't already been figured
out.
12. I have few friends of the same sex that I can really count on.
18. I find it difficult to meet members of the opposite sex whom I like.
21. I would like to bring up children of my own (if I have any) like my parents
raised me.
23. I am good at combining ideas in ways that others have not tried.
24. I am ugly.
26. I am awkward and poorly coordinated at most sports and physical activities.
27. I have generally done better in mathematics courses than other courses.
38. I don't get along very well with other members of the same sex.
39. I have good endurance and stamina in sports and physical activities.
This content downloaded from 134.176.129.147 on Thu, 26 Nov 2015 23:58:59 UTC
All use subject to JSTOR Terms and Conditions
170 MARSH AND O'NEILL
41. Spiritual/religious beliefs make my life better and make me a happier person.
44. Most of my friends are more comfortable with members of the opposite sex than
I am.
47. My parents have usually been unhappy or disappointed with what I do and have done.
50. There are lots of things about the way I look that I would like to change.
58. I do not do well on tests that require a lot of verbal reasoning ability.
63. My body weight is about right (neither too fat nor too skinny).
This content downloaded from 134.176.129.147 on Thu, 26 Nov 2015 23:58:59 UTC
All use subject to JSTOR Terms and Conditions
CONSTRUCT VALIDITY OF MULTIDIMENSIONAL SELF-CONCEPT RATINGS 171
78. I'm not very good at any activities that require physical ability and coordination.
84. I often have to read things several times before I understand them.
96. I have had lots of feelings of inadequacy about relating to members of the
opposite sex.
This content downloaded from 134.176.129.147 on Thu, 26 Nov 2015 23:58:59 UTC
All use subject to JSTOR Terms and Conditions
172 MARSH AND O'NEILL
106. I am basically an atheist, and believe that there is no being higher than man.
108. I would feel OK about cheating on a test as long as I did not get caught.
110. In school I had more trouble learning to read than most other students.
116. Most people have more friends of the same sex than I do.
119. I believe that there will be some form of continuation of my spirit or soul
after my death.
122. I never seem to have much in common with members of the opposite sex.
133. Spiritual/religious beliefs have little to do with the type of person I want
to be.
This content downloaded from 134.176.129.147 on Thu, 26 Nov 2015 23:58:59 UTC
All use subject to JSTOR Terms and Conditions
CONSTRUCT VALIDITY OF MULTIDIMENSIONAL SELF-CONCEPT RATINGS 173
REFERENCES
ALWIN, D. F., & JACKSON, D. J. (1981). Applications of simultaneous factor analysis to issues
of factorial invariance. In D. D. Jackson & E. F. Borgotta (Eds.), Factor analysis and
measurement in sociological research: A multidimensional perspective. Beverly Hills, CA:
Sage.
BACHMAN, J. G. (1970). Youth in transition: Vol. 2. The impact of family background and
intelligence on tenth-grade boys. Ann Arbor, MI: Institute for Social Research, Survey Research
Center.
BURNS, R. B. (1979). The self-concept: Theory, measurement, development and behavior.
London: Longman.
CAMPBELL, D. T., & FISKE, D. W. (1959). Convergent and discriminant validation by
multitrait-multimethod matrix. Psychological Bulletin, 56, 81-105.
CONGALTON, A. (1963). Occupational status in Australia. Sydney: University of New South
Wales Press.
COOPERSMITH, S. A. (1967). The antecedents of self-esteem. San Francisco: W. H.
Freeman.
FLYNN, M. (1975). Some Catholic schools in action. Sydney: Catholic Education Office.
HANSFORD, B. C., & HATTIE, J. A. (1982). The relationship between self and achievement/
performance measures. Review of Educational Research, 52, 123-142.
HULL, C. H., & NIE, N. H. (1981). SPSS Update 7-9. New York: McGraw-Hill.
JORESKOG, K. G., & SORBOM, D. (1981). LISREL V. Analysis of linear structural relations
by the method of maximum likelihood. Chicago: International Educational Services.
MARSH, H. W., BARNES, J., CAIRNS, L., & TIDMAN, M. (in press). The Self Description
Questionnaire (SDQ): Age effects in the structure and level of self-concept for preadolescent
children. Journal of Educational Psychology.
MARSH, H. W., BARNES, J., & HOCEVAR, D. (1983). Self-other agreement on multidimen-
sional self-concept ratings: Factor analysis and multitrait-multimethod analysis. Manuscript
submitted for publication.
MARSH, H. W., & HOCEVAR, D. (1983). Confirmatory factor analysis of multitrait-
multimethod matrices. Journal of Educational Measurement, 20, 23 1-248.
MARSH, H. W., & HOCEVAR, D. (in press). The factorial invariance of students' evaluations of
college teaching. American Educational Research Journal.
MARSH, H. W., & PARKER, J. W. (in press). Determinants of self-concept: Is it better to be a
relatively large fish in a small pond even if you don't learn to swim as well? Journal of
Personality and Social Psychology.
MARSH, H. W., PARKER, J. W., & SMITH, I. D. (1983). Preadolescent self-concept: Its
relation to self-concept as inferred by teachers and to academic ability. British Journal of
Educational Psychology, 53, 60-78.
MARSH, H. W., RELICH, J. D., & SMITH, I. D. (1983). Self-concept: The construct validity of
interpretations based upon the SDQ. Journal of Personality and Social Psychology, 45,
173-187.
MARSH, H. W., SMITH, I. D., & BARNES, J. (1983). Multitrait-multimethod analyses of the
Self Description Questionnaire: Student-teacher agreement on multidimensional ratings of
student self-concept. American Educational Research Journal, 20, 333-357.
MARSH, H. W., SMITH, I. D., BARNES, J., & BUTLER, S. (1983). Self-concept: Reliability,
dimensionality, validity, and the measurement of change. Journal of Educational Psycholo-
gy.
MARUYAMA, G., & MCGARVEY, W. (1980). Evaluating causal models: An application of
maximum likelihood analysis of structural equations. Psychological Bulletin, 87, 502-512.
This content downloaded from 134.176.129.147 on Thu, 26 Nov 2015 23:58:59 UTC
All use subject to JSTOR Terms and Conditions
174 MARSH AND O'NEILL
AUTHORS
HERBERT W. MARSH, Senior Lecturer, Department of Education, University of Sydney,
Sydney, NSW 2006, Australia. Degrees: BA, Indiana University; MA, PhD, University of
California, Los Angeles. Specializations: Educational measurement and evaluation, self-
concept.
ROSALIE O'NEILL, Postgraduate Student, Macquarie University, North Ryde, NSW, Austra-
lia.
This content downloaded from 134.176.129.147 on Thu, 26 Nov 2015 23:58:59 UTC
All use subject to JSTOR Terms and Conditions