Multivariate Behavioral Research

ISSN: 0027-3171 (Print) 1532-7906 (Online) Journal homepage: https://www.tandfonline.com/loi/hmbr20

Masculinity, Femininity and Androgyny: Their

Relations With Multiple Dimensions of Self-
Concept

Herbert W. Marsh

To cite this article: Herbert W. Marsh (1987) Masculinity, Femininity and Androgyny: Their
Relations With Multiple Dimensions of Self-Concept, Multivariate Behavioral Research, 22:1,
91-118, DOI: 10.1207/s15327906mbr2201_5

To link to this article: https://doi.org/10.1207/s15327906mbr2201_5

Published online: 10 Jun 2010.

Submit your article to this journal

Article views: 53

View related articles

Citing articles: 1 View citing articles

Full Terms & Conditions of access and use can be found at

https://www.tandfonline.com/action/journalInformation?journalCode=hmbr20
 Multivariate Behavioral Research, 1987, 22, 931-118

Masculinity, Femininity and Androgyny:

Their Relations With Multiple Dimensions of
Self-concept

Herbert W. Marsh
Department of Education
The University of Sydney, Australia

Masculinity (M) and femininity (F) were related to multiple dimensions of self-concept
in responses from 962 high school students. Androgyny theory predicts that both M and
F will contribute msitively and uniquely to self-cor~cept,but previous research, typically
relying on undiffGrentiatd self-concept-measures, has found the unique contribution of
F to be nil. In contrast the resent investigation found that M and F each contributed
positively and uniquely to ;he prediction i f well differentiated facets of self-concept.
Consistent with a new model to explain MF/self-concept relations, the differentiated
additive model, the relative contribution of M and F varied substantially depending on
the area of self-concept;F contributed more positively to the self.concept facets for which
girls had higher self-concepts than boys, and in some areas the contribution of F was
more positive than the contribution of M. Contrary to predictions for interactive
androgyny models and the sex-typed model, M-by-F interactions were not significant,
and the effects of M, F and M-by-F did not depend1 on gender. The social desirability of
MF items, whether they were negatively or positively valued, was more highly
correlated with the self-concept responses than whether the items were M or F.

The Multidimensionality of Self-concept

Historically, self-concept research has emphasized a total, undif-
ferentiated self-concept that was typically inferred from an ill-defined
hodge-podge of items, and specific facets of self-concept were relegated
a minor role. However, systematic reviews of this self-concept research
(e.g., Bums, 1979; Shavelson, Hubbard & Stanton, 1976; Welles &
Marwell, 1976; Wylie, 1974, 1979) have emphasized the lack of
theoretical models, the poor quality af instruments, methodological
shortcomings, and a general lack of consistent findings. In an attempt
to remedy this situation Shavelson et al. (1976) reviewed tlieoretical
and empirical research, and posited a multifaceted model of self-
concept in which separate facets of self-concept were clearly differen-
tiated. More recently, theoretical and empirical research has provided
I would like to acknowledge the Australian Research Grants Scheme for their
financial support of this research, Jennifer Barnes and Margaret Myers for their
assistance in collecting the data and for comments on earlier versions of this paper, Ian
Smith for his assistance on earlier portions of this study, and John Ailtill for his
permission to use the Australian Sex Role Sqale.
Correspondence related to this study should be sent to Herbert W. Marsh, Depart-
ment of Education, The University of Sydney, NSW 2006, Australia.
JANUARY 1987 91
H. W. Marsh

strong support for the multidimensionality of self-concept (Byrne,

1984; Dusek & Flaherty, 1981; Fleming & Courtney, 1984; Harter,
1982; Marsh, Barnes & Hocevar, 1985; Marsh, Smith, Barnes &
Butler, 1983; Shavelson, et al., 1976; Shavelson & Marsh, 1986; Soares
& Soares, 1982). Perhaps the strongest support for the multidimen-
sionality of self-concept,and particularly the facets from the Shavelson
model, comes from research with the set of Self Description Question-
naire (SDQ) instruments, one of which is used in the present investi-
gation.
In a review of empirical support for the multidimensionality of
self-concept Marsh and Shavelson (1985; Shavelson & Marsh, 1986)
concluded that: (a) self-concepts in specific areas will be more posi-
tively correlated with external criteria than will broad measures of
general self-concept; (b) self-concepts in specific areas will be more
positively correlated to external criteria to which they are mast
logically and theoretically related than will other specific or general
facets; and (c) the relation between self-concept and other constructs
cannot be adequately understood if the multidimensionality of self-
concept is ignored. Support for this pattern of relations between
multiple dimensions of self-concept and a variety of external criteria
has provided further support for the multidimensionality of self-
concept and its construct validity (see Byrne, 1984, for an application
of this approach to testing construct validity between multiple dimen-
sions of self-concept and academic achievement).
Studies of the relation between self-concept and measures of
masculinity (M), femininity (F) and androgyny have typically relied on
ill-defined global measures of self-concept and have not incorporated
the recent emphasis on multidimensionality in self-concept.Hence, the
purpose of the present investigation is to test the generality of the
MarshIShavelson conclusions in a study of the relations between
multiple dimensions of self-concept and MF measures. While
androgyny researchers have not typically considered the multidimen-
sionality of self-concept (Whitely, 19831, they have examined socially
desirable activities in which women might be expected to excel (e.g.,
Bem, 1975, 1977; Cook, 1985; Helmreich, Spelrce & Holahan, 1979;
also see review by Taylor & Ball, 1982) on the assumption that F
scores will contribute substantially in these areas. Consistent with
this approach, sex differences in spqcific areas of self-concept should be
related to the pattern of correla$ions found between these specific
facets and MF scores. In particular, the c~ntributionof F is likely to be
larger in those specific areas of self-concept where females have higher
92 MULTIVARIATE BEHAVIORAL RESEARGH
 H. W. Marsh

self-concepts than males, while the contribution of M is likely to be

larger in those facets where males have higher self-concepts than
females. This hypothesis is more fully articulated along with the
discussion of other theoretical models of the MF/self-concept relation
in the discussion that follows.

The Androgyny Construct

Virtually all researchers prior to 1973 and many personality
inventories still in use today assume that: (a) IM and F are the
end-points of a single, bipolar dimension; (b) M and F are correlated
close to -1.0; and (c) more M (F) implies less F (M). More recently
Constantinople (1973), Bern (1974), Spence (1984), Heilbrun (1976),
and others have argued that it is logically possible to be both 18 and F,
and the existence of both in the same person has been labeled
androgyny. The two key assumptions to Bem's I974 theoretical de-
scription of androgyny are that M and F are distinguishable, orthog-
onal dimensions, and that individuals high on both are mentally
healthier and socially more effective.
BSRI and PAQ are the most widely used androgyny instruments,
but their reliance only on socially desirable attributes may constitute
an important weakness. For example, a response bias may affect the
"true" correlation between M and F, and correlations between MF and
self-esteem (Baumrind, 1982; Kelly, Claudill, Hathorn & O'Brien,
1977; Kelly & Worell, 1977; Marsh & Myers, 1986; Pedhauzer &
Tetenbaum, 1979). This potential problem is both likely to affect
correlations between M and F, and to affect relations between MF and
other constructs such as self-concept. In response to this potential
weakness, Spence, Helmreich and Holahan (1979) expanded PAQ
(EPAQ) to include socially undesirabb characteristics, and Antill,
Cunningham, Russell and Thompson (1981) develalped the Australian
Sex-Role Scale (ASRS) to measure M and F with positively valued
characteristics (Mi- & F+) and with negatively valued characteristics
(M- & F-1. The ASRS is used in the present investigation.
In support of androgyny theory, androgyny researchers have
typically found that MF correlations differ significantly from - 1.0.
However, Marsh & Myers (1986; Marsh, 1985b)found MF correlations
for different instruments varied from moderately positive to close to
-1.0 and showed how these differenceswere logically consistent with
the design and the selection of items for each instrument. Other
investigators have found that responses to the adjectives "masculine"
and "feminine" are substantially negatively correlated and form a
JANUARY 1987 93
H. W. Marsh

two-item bipolar factor (e.g., Pedhauzur & Tetenbaum, 1979). Hence,

empirical MF correlations apparently will not provide unequivocal
support for either the androgyny or the bipolar assumptions, and
support for androgyny theory requires the demonstration that both M
and F contribute uniquely and positively to the prediction of appropri-
ate criteria. The most frequently studied criterion has been measures
of self-concept, self-esteem or psychological well-being that are central
to androgyny theory, and tests of this assumption are the focus of the
present investigation.

Theoretical Models of MFISelf-concept Relations and Their Tests

For purposes of the present investigation five different models of
MFlself-concept relations are examined. Since the first three of these
models have been well described elsewhere (Antill & Cunningham,
1980; Hall & Taylor, 1985; Lubinski, Tellegen & Butcher, 1983;
Spence, 1984; Taylor & Hall, 1982; Whitely, 1983) they will be
summarized only briefly. More detailed descriptions are presented for
the interactive model that has been the source of recent confusion and
controversy, and for the differentiated additive model that was first
formulated as part of the present investigation. As described below
support for each of these models is interpreted from the results of
ANOVA or regression analyses that relate self-concept measures to
the main effects of M and F, the M-by-F interaction, and the inter&-
tion of these effects with gender.
Sex-Typed Model. This model posits that the acquisition of a
masculine identity by males and of a feminine identity by females
leads to higher self-concepts, and that, perhaps, support for such a
model should be strongest during early adolescent years when such
acquisition processes ape typically assumed to be most important (e.g.,
Kohlberg, 1966: Lamke, 1982). This model is consistent with, but does
not require the bipolarity of MF (Whitely, 1983). Support for this
model requires that the effects of M and F each interact with gender,
and for purposes of the present investigation this interaction is
interpreted to imply that: (a) for females F will be more positively
correlated with self-concept than will M, but for males M will be mqre
positively correlated with self-concept than will F; and (b) F will be
more positively correlated with self-concept for females than males,
and M will be more positively correlated with salf-concept for males
than females.
Additive Androgyny Model. As described by Spence and her
colleagues and apparently consistent with Bem's original formulatian,
94 MULTIVARIATE BEHAVIORAL RESEARCH
 H. W. Marsh

this model posits that M and F each contribute positively and uniquely
to the prediction of self-concept. This implies that the main effects of
both M and F will be statistically significant.
Masculinity Model. This model posits that self-concept, at least in
modern Western societies, is primarily determined by M rather than F,
and is sometimes called the "masculine supremacy effect" (Cook, 1985,
p. 96). This model appears to be prompted primarily by empirical
findings rather than theory, though it may be consistent with feminist
perspectives on the organization of society. This model is consistent
with but does not require the bipolarity of MF. Support for this model
implies a main effect of NZ but no main effect of F, or perhaps a main
effect of F in which F contributes negatively to self-concept after
controlling for the main effect of M.
Interactive Androgyny Model. This model proposes that androgyny
is more than the additive sum of M and F. In the ANOVA approach
this model posits an M-by-F interaction and in the typical multiple
regression approach it posits that the MF crosspraduct contributes
significantlyto the prediction of self-concept beyond the contribution of
M and F.In order to test this model Lulbinski et al. (1983) recommend
an ANOVA or regression approach that tests the main effects of M, F,
gender and each of the possible interactions; the main effects of M and
F test the additive androgyny model and the masculinity model, the
M-by-F interaction tests the interactive androgyny model, and inter-
actions between these effects and gender test the sex-typed mlodel. Hall
and Taylor (1985) distinguished between what they called a balance
interactive model that posits an M-by-F interaction without main
effects of M or F, and an emergent interactive model that posits
significant effects of M, F, and M-by-F. Heilbru111(1976) proposed an
earlier version of this emergent model in which he defined androgyny
to be the sum of M and F (as in the additive model) minus the absolute
difference of M and F (as in the balance notion), though the absolute
difference has undesirable characteristics (see Hall & Taylor, 1985)
and his definition confounds the additive and interactive effects.
Applications of the ANOVA approach typically use a 2 x 2 design
(Hall & Taylor, 1985; Lubinski, et al., 1983; Taylor & I-Iall, 1982;
Whitely, 1983)that throws away much systematic variance in its gross
classification of M and F into dichotomies, considers only the linear
effects of M and F, and aonsiders only the linear-by-linear component
of the M-by-F interaction. The typical regression approach provides a
stronger test of the main effects since M and F are not dichotomized,
but still tests only the linear components of the main effects and the
JANUARY 1987 95
H. W. Marsh

linear-by-linear component of the interaction effect. These limitations

are not inherent in either the ANOVA approach or the regression
approach, and more sophisticated designs are readily available (e.g.,
Cohen & Cohen, 1975; Pedhazur, 1982). Spence (1984) also notes that
there are many forms of an M-by-F interaction that may be theoreti-
cally meaningful but cannot be tested with the linear-by-linear com-
ponent, and she recommends 4 x 4 ANOVA design in which M and F
are divided into quartiles. This is advantageous in that it provides a
stronger test of the main effects than the 2 x 2 design, it provides tests
of more complex interactions than just the linear-by-linear component,
and the linear-by-linear component considered by most researchers is
just a special case of this more general model. This 4 x 4 design,
supplemented by subsequent regression analyses is to be used in the
present investigation.
DifferentiatedAdditive Androgyny Model. Like the additive model
this model posits that M and F each contribute positively to the
prediction of self-concept, however its critical prediction is that the
relative contribution of M and F will differ systematically according to
the specific facet of self-concept. Consistent with Marsh and
Shavelson's general proposal that was described earlier it is posited
that: (a) M will be significantly more strongly related (for both males
and females) to those facets of self-concept that are more logically
related to M; and (b) F will be significantly more strongly related (for
both males and females) to those facets of selfeconcept that are more
logically related to F. In order to eliminate the possible circularity in
this hypothesis (i.e., assuming that facets that are more positively
correlated to M or to F are more logically related to M or to F), facets
most logically related to M is translated to mean thosa facets of
self-concept for which males have higher scores, and facets most
logically related to F is translated to mean those facets of self-concept
for which females have higher scores than males.
A general test of this model is based on a repeated measures
ANOVA. In addition to the effects of M, F, gender, and their interac-
tions that have been considered previously, each different facet of
self-concept will correspond to one level of the repeated measures
variable that are called (self~concept)scales in the present investiga-
tion. For this analysis the model posits that the main effects of M and
F will be statistically significant, that theke will be a sigpificant
M-by-scale interaction (that the effect of M will vary with the facet of
self-concept), that there will be a significaat F-by-scale interaction
(that the effect of F will vary with the facet of self-concept),and that no
96 MULTIVARIATE BEHAVIORAL RESEARCH
 H.W. Marsh

other interactions will be of practical significance. If support for these

broad predictions is found, then further analyses are necessary to
determine if the form of the M-by-scale and F-by-scale interactions
corresponds to the model's predictions.
The differentiated additive model was specifi~callydeveloped for
purposes of the present investigation, but the model is consistent with:
(a) the pattern of results relating multiple facets of self-concept to
other external constructs as described earlier; (b) Whitely's s~~ggestion
that 'home dimensions of self-esteem may be more closely related to
sex role orientations than others" (1983, p. 774); (c) Cook's suggestion
that the "use of global measures of other constructs, especially self-
esteem, may obscure associations with femininity. Such measures may
be more fruitfully broken down into aspects differentially related to
masculinity and femininity" (1985, p. 95); (d) conclusions from Taylor
and Hall's 1982 meta-analysis (also see Cook, 1985) that Nt is more
consistently correlated to male-type dependent measures tlnan is F,
while F is somewhat more consistently correlated to female-type
dependent measures than is M; and (e) particularly Flaherty and
Dusek's conclusion that "the relation of masculinity and femininity to
self-concept depends on the aspect of self-concept that is being mea-
sured. If the self-concept measure reflects a traditional masculine
orientation, then androgynous and masculine subjects, regardless of
gender, will score high. If the self-concept measure reflects the tradi-
tionally female expressive role, the androgynous arid feminine subjects
will score high" (1980, p. 990).

Empirical Support For the Models

Empirical support (e.g., Antill & Cimningham, 1979; 1980), liter-
ature reviews and meta-analyses (e.g., Bassoff & Glass, 1982; Cook,
1985; Hall & Taylor, 1985; Spence, 1984; Taylor & Hall, 1982; Whitely,
1983) of MFlself-concept relations consistently provide the strongest
support for the masculinity model. In perhaps the most systematic
single study Antill and Cunningham (1979; 1980) related responses
from five different IMF instruments and two different esteem measures
and concluded that: "In every case masculinity showed significant
positive correlations with self-esteem in both sexes whereas the
correlations with femininity were generally nil or negative" (1979, p.
783). Meta-analyses by Taylor and Hall and by Whitely found.that
most of the variance in MFIself-conceptrelations could be explained by
Mywhile little contribution was made by the main effect of F or M-by-F
interactions, and that these findings were consistent across responses
JANUARY 1987 97
H. W. Marsh

by males and by females. Bassoff and Glass's meta-analysis of a wider

variety of mental health measures reached similar conclusions. In her
recent review Cook notes that: "The best documented and robust
association in the androgyny literature is that between masculinity
and paper and pencil tests of self-esteem. Femininity is more weakly
related, if at all" (1985, p. 94). Hence, androgyny researchers have
been unable to find much support for either the additive or interactive
models that are derived from and central to androgyny theory, and this
represents, perhaps, the most devastating appraisal of androgyny
theory.
However, most previous research suffers from two limitations that
will be examined here. First, most studies have used MF instruments
that assess only positively valued M and F characteristics, even
though these may be unduly influenced by social desirability, and MF
stereotypes include undesirable as well as desirable characteristics.
Second, most studies have used undifferentiated, global self-concept
measures, even though recent self-concept research and theory empha-
sizes the multidimensionality of self-concept. Hence, most previous
research has been unable to test the differentiated additive androgyny
model that is the focus of the present investigation even though the
model appears to be consistent with commonsense, suggestions by
many androgyny researchers, and some empirical findings from both
self-concept research and androgyny research.

The Present Investigation

The purpose of the present investigation is to examine the rela-
tions between 4 ASRS (M+, M-, F+, F-) scales and 11 facets of
self-concept measured by the SDQ I1 with respect to the theoretical
models and issues discussed earlier. In the first set of analyses a
repeated measures ANOVA is used to test predictions for each of the
models and to determine the extent to which the effects vary with the
facet of self-concept. In the second set of analyses multiple regression
is used to further explore the specific nature of MFIself-concept
relations found in the first set of analyses.
The major prediction of this study is that the differentiated
additive model will be supported in that: (a) the relative contribution
of M and F will vary according to the specific area of self-concept; (b)
the contribution of M will be relatively larger for thase facets for which
males have higher self-concepts; and (c) the contribution of F will be
relatively larger for those facets for which females have higher
self-concepts. Nevertheless, it is also predicted that the social desir-
98 MULTIVARIATE BEHAVIORAL RESEARCH
 H. W. Marsh

ability of MF items, independent of whether they are M or F, will be

substantially related to multidimensiona~self-concepts.
Method
Sample and Procedures

The sample consisted of 962 (49% female) high-school students

(grades 7-11) who attended one of two neighboring schools that served
the same predominantly middle class suburb of metropolitan Sydney,
Australia. The two self-report instruments, the SDQ I1 and the ASRS,
were administered to large groups of students in each of the bwo high
schools on consecutive days near the end of the academic year by three
researchers not connected with the schools. Each group consisted of all
the students in the same year at that school. For both self-report
instruments instructions were read aloud, several practice items were
administered, questions about the instructions were answered, and
then the items were read aloud at a fairly rapid pace (though students
had a copy of the instrument in front of them so that they could read
along if they chose to do so). The primary purpose of reading tlie items
aloud was to ensure that students spent a standard amount of"time on
each item of both instruments and finished within the 40 minutes
allowed for the task.

Instruments
ASRS. In constructing the ASRS, Antill, et al. (1981) began with
a pool of 591 items including the original pool of items used to develop
the BSRI, and all items from the PAQ, from Heilbrun's (1976)
measure, and from other MF instruments. Subjects rated each item as
a desired characteristic and as an expected characteristic for males and
for females, and these four responses were used to select items as
Spence, Helmreich, and Holahan (1979) selected items for PAQ (i.e.,
M i items were seen as desirable for males by males and females and
were seen as significantly more typical of males than females by males
and fem@le$).Form A and form B of the ASRS each consist of 50
personality-like characteristics and subjects respond to each item
according to how true it is as a self-description on a "Never or almost
never true" (1)to "Always or almost always true" (7)scale. Each form
contains 20 M items, 20 F items and 10 neutral items with half the
items within each category being positively valued (i.e., socially
desirable) and half negatively valued. For purposes of the present
investigation only the 40 MF items from form A are! used, and the M i ,
JANUARY 1987 99
H. W. Marsh

M-, F+, F- scores are each represented as the unweighted sum of

responses to ten items. Examples of items representing each scale are:
firm, confident, skilled in business (M+); noisy, aggressive, boastful
(M-1; patient, sensitive to the needs of others, responsible (F+); and
dependent, shy, weak (F-1. Marsh and Myers (in press) presented
psychometric properties for each of these four scales based on an
earlier study (also see Antill, et al., 1981; Russell & Antill, 1984) and
used confirmatory factor analysis to demonstrate that a four-factor
(M+, M-, F+, F-) solution provided an adequate fit to ASRS re-
sponses. Reliability estimates and correlations with gender are pre-
sented in Table 1for the present investigation.
SDQ II. The $DQ I1 is one of a set of three instruments designed
to measure self-concepts for preadolescents (SDQ), early-adolescents
(SDQ 111, and late-adolescents (SDQ 111). The SDQ instruments are
based upon the Shavelson model of self-concept (Shavelson, et al.,
1976; Shavelson & Marsh, in press) and the multiple dimensions of
self-concept proposed in that model. A detailed review of these instau-
ments is beyond the scope of the present investigation and the relevant
information is summarized only briefly here (see Marsh, 1985a; 1986;
Marsh, Barnes & Hocevar, 1985; Marsh & O'Niell, 1984; Marsh,
Parker & Barnes, 1985;Marsh, Relich & Smith, 1983;Marsh, Richards
& Barnes, 1986; Marsh, Smith & Barnes, 1983, 1984, 1985; Marsh,
Smith, Barnes & Butler, 1983; Shavelson & Marsh, 1986). Factor
analyses have identified the factors that each is designed to measure
and other research has shown that: (a) the reliability of each factor is
generally in the 0.80's and 0.90's while correlations among the factors
are modest (median r's generally .2 or less); (b) the self-concept factors
are substantially carrelated with self-concepts in matching areas as
inferred by teachers and significant others); (c) the self-con~eptsin
academic areas are substantially carrelated with academic achieve-
ment indicators while nonacademic self-concepts are not; and (dl
self-concept factors are systematically and logically related to a vari-
ety of other constructs such as gender, age, locus of control, self-
attributions for the causes of academic success and failure, and
interventions designed to affect specific facets of self-concept. Thege
findings support the construct validity of interpretations based upon
the SDQ instruments.
The SDQ I1 is designed to measure 11areas of self-concept defined
by responses to 122 items approximately half of which are negatively
worded, on a "1-False" to "6-True" response scale. In previous research
with the SDQ I1 (Marsh, Parker & Barnes, 1985), factor analysis
100 MULTIVARIATE BEHAVIORAL RESEARCH
 H. W. Marsh

clearly identified each of the 11 SDQ I1 factors, the reliability of each

scale was high (median alpha = 0.861, correlations among the 11
factors were modest (median r = .17), the scales were significantly and
logically related to gender, and school performances in math and
English classes were substantially correlated with Math and Verbal
self-concepts but relatively uncorrelated with self-concepts in
nonacademic areas. A brief description of the SDQ I1 scales and
examples of items are as follows:
General Self-a scale based on the Rosenberg (1965) self-esteem
scale. Example items are: "I can't do anything right", "Overall, I have
a lot to be proud of', and "Overall, most things I do turn out well".
Mathematics-student perceptions of their mathematical skills1
reasoning ability, and their enjoyment,linterest in mathematics. Ex-
ample items are "I get good marks in mathematics", "Mathematics is
one of my best subjects", and "I hate mathematics".
Verbal-student perceptions of their verbal skillslreasoning abil-
ity and their enjoymentfinterest in verbal activities. Example items
are: "I get good marks in English, "I hate reading", and "I do badly on
tests that need a lot of reading ability".
General School-student perceptions of their ability and
enjoyment/interest in school in general. Example items are: "I'm good
at most school subjects", "I'm too stupid at school to get into a
university", "I'mnot very interested in any school subjects".
Physical Abilitieestudent perceptions of thellr skills and interest
in sports and physical activities. Example items are: "I'm good at
things like sport, gym and dance", "I am lazy when it comes to sports
and hard physical exercise", and "I enjoy things like sports, gym and
dance".
Physical Appearance-student perceptions of their physical at-
tractiveness. Example items are: "I am good looking", "I have a nice
looking face", and "I hate the way I look".
Relations With Same Sex and Opposite Sex Peers-these two
scales measure student perceptions of their interactions with peers.
Within each scale, some items spacifically refer to same and opposite
sex like "I enjoy spending time with friends of the same sex" and
"When I'm alone with members of the opposite sex I feel shy & unsure
of myself". Other items refer to boys and girls and are scored according
to the sex of the respondent. Examples of these items are "Boys often
make fun of me" and "I do not get along well with girls7'.
Relations With Parents-student perceptions of their interactions
with their parents. Example items are: "My parents treat me fairly", "I
JANUARY 1987 101
H. W. Marsh

get along well with my parents" and "It is difficult for me to talk to my
parents".
Honesty-student perceptions of their honesty and trustworthi-
ness. Example items are: "I am honest", "Cheating on a test is OK if I
don't get caught", and "People can count on me to do the right thing".
Emotional Stability-student perceptions of their emotional sta-
bility. Example items are: "I am often depressed and down in the
dumps", "I am a nervous person", "I get upset easily".

Statistical Analyses
All statistical analyses in this study were conducted with the
commercially available SPSS program (Hull & Nie, 1981; Nie, Hull,
Jenkins, Steinbrenner & Bent, 1975). Except for the 66 students (7%)
who failed to complete one or the other of the two instruments, mean
responses were substituted for missing values for all completed instru-
ments. A pair-wise deletion for missing values was used in computing
correlations, but correlations based on a case-wise deletion for missing
values were virtually the same. Because of the very large sample size,
the emphasis will be placed on effect sizes and variance explained,
rather than on mere statistical significance.
Preliminary Analyses. Preliminary analyses were conducted to
examine psychometric properties of responses to both instruments. For
the SDQ I1 (see Marsh, Parker & Barnes, 1985, for a more complete
description of the analyses and similar findings for a different sample)
a factor analysis identified the 11SDQ I1 factors and was used to define
factor scores to represent the 11SDQ I1 facets; an item analysis showed
the factors to be reliable (alphas from 0.84 to 0.92; median alpha =
0.89); and correlations among the factors were modest (median r =
0.20). For the ASRS (see Table 1)coefficient alphas for the four ASRS
scales were 0.67 (MS), 0.78 (M-1, 0.75 (F+), and 0.67 (F-); alphas
were similar for responses by males and females; and ASRS scores
were significantly though modestly related to gender in the expected
direction.
During the academic year in which these data were collected the
two schools were in the process of switching from being single-sexed
schools to being coeducational schools; grades 7, 9 and 11 were
coeducational, while students in grades 8 and 10 attended single-sex
classes. In order to examine specific characteristics of the present
investigation, additional variables were defined to represent the school
(a dichotomous variable scored 1 or 2), student gender (1= male, 2 =
female), linear and nonlinear components of age (i.e., age, age squared,
102 MULTIVARIATE BEHAVIORAL RESEARCH
 H. W.Marsh
Table 1
Relations Aeon Masculinity/Femininity (MF) Scores and Back round
,
Variables For idle (m) Females ( f ) and the Total Sample (t?
_________-______-___-------_-------A--------------------------

MF Scores
-

Scores Sample !1) (2) (3) (41 (5) (6)

_-____________-____-------_-------------_---------------------
MF Scores
1 Masculine P o r i t ~ v e
(N+)

2 Masculine Negative
(M-)

4 Feminine Negative
iF-)

5 Masculine Total
rw+) -t <I?-1

b Feminine Total
(F+) + (F-)

Demograph1r Scores
Age/Grade Level

Gender i I=Boys,
Z=Girls)

Single Sex Class

!l=Coed classes, 2=
single se:: claiises)
a
School

gpte, Coefficient alpha estimates of reliability for the W F scores

appear in parentheses.
<
-+ Since
p -05
there are only two schools, school is a dichotomous
variable. The school scored " 1 " was formerly an ail-boys school; it
had only boys in grades 8 and 1 0 , but the other grades were
coeducational. The school scored "2" was formerly an a1 $-girls
school; it had only girls in grader 8 a.nd 10 but nther grades were
coeducational.

JANUARY 1987
H. W. Marsh

and age cubed) for which age was summarized according to the grade
level, and the type of class (1= single sex, 2 = coeducational). Each of
these variables was correlated with the four ASRS scores for the total
sample, and separately for males and females (except for gender). All
correlations (Table 1)other than those involving gender were trivial;
the largest of the 36 correlations was 0.16, few were statistically
significant, and most fell in the range of +0.05 to -0.05. As a further
test, these additional variables were included in multiple regressions
predicting each of the 11 self-concepts on the basis of the 4 ASRS
scales. In a few instances the inclusion of these additional variables
resulted in a significantly, albeit small, increase in multiple R, but the
size of the standardized beta weights for the 4 ASRS scores was nearly
unaffected. While characteristics particular to the present investiga-
tion dictate caution in generalizing the findings, these characteristics
apparently have little effect on the MF scores and their relations to
self-concept.
ANOVA Analyses. For purposes of just these analyses subjects
were divided into one of four F groups and one of four M groups. The
tests consisted of a series of 4 (levels of F) by 4 (levels of M) by 2
(gender) by 11 (self-concept scales-a repeated-measures or within-
subject variable) analyses conducted with the MANOVA procedure of
SPSS (Hull & Nie, 1981). Separate analyses were conducted for groups
based on positively valued MF items, on negatively valued MF items,
and on their total.
Correlation and Regression Analyses. Each of the four ASRS scales
and various unweighted combinations of the scales were correlated
with the SDQ II scales separately for responses by males, by females
and for the total sample. Then a set of multiple regression analyses
was used to predict each self-concept score from the four ASRS scores
or from a variety of scores derived from the original four scales. The
size and direction of first-order correlations, and of standardized beta
weights from the multiple regressions, were used to examine the
contributions of various MF scores to the prediction of the multiple
self-concepts.
Results
ANOVA Tests of Five Theoretical Models of MFISelf-concept
Relations

The sex-typed model posits that the effect of M and the effect of F
(and perhaps their interaction) will vary according to gender, and,
perhaps, that this effect should be largest in the early adolescent
104 MULTIVARIATE BEHAVIORAL RESEARCH
 H. W. Marsh

period considered in the present investigation. In particular, M should

contribute more positively to self-concept for boys, and F should
contribute more positively for girls. However, gender fails to lnteract
significantly with M, with F, or with M-by-F, for MF groups based on
positively valued MF items, negatively valued MF items, or their total.
Furthermore, this lack of support for the sex-typed model is consistent
across all 11 areas of self-concept as evidenced by the nonsignificant
M-by-gender-by-scale,F-by-gender-by-scale and M-bly-F-by-gender-by-
scale interactions. Given the statistical power of this test ;and the
consistency of the findings across the 11areas of self-concept,~thelack
of support for the sex-typed model is compelling.
The androgyny interactive models posit an M-by-F interaction,
though different forms of this model may imply different types of
interaction and not all interactions may be consistent with an~drogyny
theory. However, the M-by-F interaction fails t o reach statistical
significance (Table 2) for MF groups based on positively valued MF
items, negatively valued MF items, or tlheir total. Furthermore, this
lack of M-by-F interaction is consistent across all 11 areas of self-
concept as evidenced by the nonsignificant M-by-F-by-scale interac-
tion. Given the statistical power of this test and consistency of the
findings across the 11areas of self-concept, the lack of support for the
interaction model is also compelling.
The additive androgyny model is supported for MF groups based
on positively valued items in that both M and F contribute substan-
tially (Table 2) and the direction of the contribution is positive for all
areas of self-concept (as will be discussed later). However, the substan-
tial M-by-scale and F-by-scale interactions demonstrate, as predicted
by the differentiated additive model, that the size of the positive
contribution of M and F varies significantflywith the particular facet of
self-concept. These interactions and their relation to the specific
pattern of results posited in the differentiated additive model will be
the focus of subsequent analyses.
Tests of the additive and differentiated additive models are more
complicated for MF groupings based on the negatively valued MF
items and the sum of the positively and negatively valued items. For
the negatively valued items, the contribution of M across the areas of
self-concept is not statistically significant while the significant contri-
bution of F is negative (i.e., the selfhendorsement of negatively valued
F items is associated with lower seIf~concept). For the MF groups based
on the total of positively and negatively valued items, the contribution
of M is significant and positive (due primarily to the positive contri-
JANUARY 1987 105
H. W. Marsh

Table 2
ANDVA of Effects of Gender Masculinity (M) and Femininit !F) on Multi le
Facets of Self -concept ~ o r ' ~ o stivel
i y Valued and ~egatlveryValued MF reems
..........................................................................
Total
-------------- Posi ti ve
------------- Negative
-------------
Effect df MS F-Ratio MS F-Ratio MS F-Ratio
..........................................................................
Between 863 2.50 1.94 2.32
Femininity (Fj 3 5.39 2.2 91.60 47.1*x 79.95 5 4 . 5 ~ ~
Mbsculinity (M) 3 25.67 10.3xx 115.20 59.3** 3.88 1.7
M x F 9 0.77 C1.Z 0.83 1:1.4 2.11 0.9
Gender (GI 1 0.97 0.4 9.33 4.8 0. 49 Ci. 2
G x F
G x M
G x M x F 9 4.18 1.7 1.63 0-8 2.15 1.0
Total 894 2.58 2.62 2.58
Within 8630 0.61 0 . 62 Sl.61
Scale (S! 10 O.DV 0.2 0.07 0.1 it.07 0.1
M x S 7-
-4 6.18 ?<I.l x * 4.10 6.6** 6.12 10. l x x

G x S 10 19.20 T1.3xx 18.62 30.::5+* 22-18 36. &*x

M x F x G x S sO 0.43 0.7 0.81 1.51 0.5 0.8

Total 8950 0.67 3. 67 13- 67

Ngfe, Subjects were divzded into four groups on the basis of their M and F
scores and a 4 in) z 4 :F> x 2 iGender) x 12 <Self-concept scales -- a
repeated measure variabiej was performed on the multidimensional self-
concept responses with the MONOVG procedure from the SFSS iHull % Nie,
1981). Separate analyses were performed on t? and F derived qrom positively
valued items, negatively valued items, and their total.
x p < .o:; ** p 4 .001

bution of M+ items) but the contribution of F is not significant (i.e., the

positive contribution of F+ and the negative contribution of F- cancel
out). However, the substantial M-by-scale and F-by-scale interactions
in both of these analyses again demonstrate that the effects of both M
and F vary significantly with the particular facet of self-concept as
predicted by the differentiated additive model, and the pattern of these
effects also will be considered in the next section.
The masculinity model posits that the main effect of M on
self-concept will be positive while the eEect of F will be nonsignificant
or negative. For the positively valued MF items there is no support for
this model in that contributions of M and F are both positive, though
support is somewhat stronger for the negatively valued MF items for
which the contribution of M items was relatively more positive, or at
least less negative, than the contribution of F. However, the relative
106 MULTIVARIATE BEHAVIORAL RESEARCH
 H.W. Marsh

contribution of M and of F varied substantially with the particular

facet of self-concept and this does not seem to be consistent with the
masculinity model. Hence, support for the masculilnity model is weak.
In summary, these ANOVA results provide no support at all for
the androgyny interactive or sex-typed models. Support for masculin-
ity is weak in that F, particularly F+, contributes substantially to
self-concept,and the relative contribution of M and of F varies with the
area of self-concept. Support for the additive and particularly the
differentiated additive models is strong.

MF correlations with Multiple Dimensiolns of Self-concept

Correlations and beta weights relating the four ASRS scores to the
11 SDQ I1 scores are presented in Table 3.l Consistent with the
ANOVA results these findings show that when averaged across all
areas of self-concept (the mean of coefficcients in Table 3) correlations
are: (a) modestly positive for the M+ and F+ scores; (b) smaller and
negative for the F- score; (c) and close to zero for the M- score. Also
consistent with the ANOVA results and with the differentiated addi-
tive androgyny model, the pattern of MF/self-concept correlations
differs substantially depending on the area of self-concept.
F+ and M+ both contribute positively to self-concept. In three
areas-Honesty/Trustworthiness, Parent Relations, and Same Sex
Relations-the self-concept scores are more positively correlated to F+
than to M+, and two of these are areas of self-concept in which girls
have substantially higher self-concepts than do boys (see Table 4).
Similarly, boys have substantially higher self-concepts than girls in
Ph.ysica1 Appearance, Phypical Ability, and Mathematics (Table 4),
and these self-concept scores are more positively (correlatedwith M+
than with F+.
M- and particularly F- tend to contribute negatively to self-
concept (Table 3). However, M- contributed positively to Opposite Sex
Relations and particularly to Physical Appearance, and Physical
Appearance was the area of self-concept in which the sex {difference
favoring boys was largest. F- did not contribute positively to any
areas of self-concept,but Honesty was the area where the contribution
lCorrelations and standardized beta weights are only presented for analyses of the
total groqp in Tables 3 , 5 , 6 and 7 even though analyses were also performeld separately
for responses by males and by females. These additional analyses were not considered
further because the result8 of the ANOVA (Table 2) indicate that the effects of M, F, and
M-by-F do not vary according for gender, and thus provide no justification for the further
analyses. As implied by these results, corlclusions based on the coefficients for the total
group are generally consistent with those for males and females consideredl separately.
The inclusion of these findings would have made these tables unduly complicated.
JANUARY 1987 107
 H. W. Marsh

Table T
Relatiovr P m u n g Marc:~li~ity/Feminin;ty: P ? F I Scorer and Wultlple Self-coqrepts
- -.. .- -- - -- .- ----.
~- -~

t?F Scorer

Self-concepts
------------ - -----------
n+
r beta -
M-
beta
F+
r %eta
c-
r
~
beta
---.
~
nu1 t
R

Mathematics .2t1.x
Verbal .Ti*
General School . 3 3
Physical Abilities .1;*
Physical Appearance .54*
Opposlte Ser Peers .;ax
Same Sex Peers .21*
Parent Relations .OYx
Honesty ~ ..
?:7
E m o t ~ o n a l Stabll; ty .27*
General Sel* .46*
Mean Coeff. .2b4
Mean of Squared
See+ f . ~ ::,ST

!jptsl P Series of I ? multiple r e g ~ x i o v sw a c conductr< to r e ~ a t E the four

E-,
MF scores (!-I+- F+ F F-) ts the !1 ieii-rzncept scores. The simple
correlations a n d standardized beta welqhtc resultxng *roc these analyses
are presented I? this table. iKesu?ts fsr males iema:ez s;mllar
t o each othei a n d t n thore showv here.:
* -35

of F- was least negative (not significantly different from zero) and this
was the area of self-concept most favoring girls. The most negative
contribution of F- is for the Emotional scale, but boys and girls did not
differ significantly on this facet. An inspection of the content of the F-
items (e.g., anxious, nervous, worrying) suggests that this effect may
be idiosyncratic to the ASRS, but the F+ scale is also less positively
correlated with the Emotional scale than with other areas of self-
concept, and Spence, et al. (1979) reported F+ and F- to be signifi-
cantly related to neuroticism.
The correlations in Table 3 tend to support the differentiated
additive model but the large number of coefficients and the influence of
social desirability on the ASRS scores complicates the interpretations
so that a more objective index is needed. The size and direction of the
correlations between each self-concept and bipolar MF scores based on
positively valued items (M+ - F+), on negatively valued items (M- -
F-), and on their total (MTOT - FTOT) provides such an index. The
M scores are weighted +1 and the F scores are weighted -1 in the
computation of these bipolar scores. Thus, if the correlation between
any bipolar MF score and a self-concept score is positive, then M scores
108 MULTIVARIATE BEHAVIORAL RESEARCH
 H. W. Marsh

contribute more positively (or at least less negatively)than do F scores;

if the correlation is negative, then the positive contribution of the F
scores is larger; if the correlation does not differ significantly from zero,
then the relative positive contribution of M and F scores is about the
same. These three bipolar MF scores were correlated with the self-
concept scores (see Table 4). For each of the three bipolar MF scores, 6
of 11 correlations were significantly positiv~indicatinga larger
positive contribution for M than for F, 2 or 3 were significantly
negative-indicating a larger positive contribution for F than M, and
2 or 3 did not differ significantly from zero. The consistency of this
pattern of results for the positively and negatively valued MF items
offers support for the generality of the conclusions.
The differentiated additive model posits that F will contribute
more positively to those areas of self-concept in which females have
higher self-concepts and that M will contribute more positively in
those areas in which males have higher self-concepts. The set of 11
correlations between the bipolar MF scores and multiple self-concepts
provides a quantitative index of the relative positive contribution of M
and F to each area of self-concept, and similarly correlations between
gender (1= male, 2 = female; see Table 4) and each area of self-concept
provide a quantitative index of gender differences in each area of
self-concept. The relation between the two sets of correlations, corre-
lations between the bipolar MF scores and the selEconcept scores, and
correlations between gender and the self-concept scores, provide a test
of the differentiated additive model. This correlation is -31, -.61 and
-.71 (df = 9, all ps < .05) for the positive, negative and total MF
scores. Thus, the areas of self-concept most favoring girls (11.e.~ those
where correlations with gender are most positive) are the ones in
which the positive contribution of F is larger than M (i.e., col~elations
with bipolar MF are most negative). These findings provide strong
quantitative substantiation for conelusions based on an inspection of
correlations between self-concepts described above and offer strong
support for the differentiated additive model. The consistency of the
results across positively and negatively valued MF items offers support
for the generality of the conclusions.

A Priori & Empirical Combinations of MF Scores and Their Relation

to Self-Concepts
The purpose of the analyses described in this section is to deter-
mine how well various unweighted combinations of the four ASRS
JANUARY 1987 109
H. W. Marsh
Table 4
Re1 at ions Amonq Pi polar . Mascul i nitynemi ni ni ty (MF) Scores,
Gender, and Mu tldlmenslonal Self-concepts
...........................................................
B~polarMF Scores For:
................................
Positive % Gender
Po51tlve Negatl ve Negatlve ( l=male

Self-concepts MF xtems MF ~ t e m s MF ~ t e m s 2=f emale)

............................................................
Math -05 "04 -05 -.12*+
Verbal -03 -05 .04 .67*
School .0B** .07* .08** .Q7*
Physical .It+* .in* .14** -. 14**
Appearance .24+* .30** .Sl*x -.S4**
Opposite Sex .14x+ .27** .24** - -135
Same Sex -.lox* .0 9 X x .01 .29**
Parent -.12x* -.1zxx -.15x+ .OCI
Honesty -.31+u - .32** -.56r* .25*+
Emot~onal .It+* .Z5+* .24+* -.u5
General .08+ .I?** .11++ -.05
Gender -.37+r -.19+* -.30+* 1-00

Note: For each of the Bipolar MF scores (defined to b e M minus F).

posltive correlations wlth self-concept scores mean that M
contributed more pasltlvely khan F. whlle negatlve correlat~onsmean
that F contributed more positively than M. The relative size of the
poslt~vecontrlhutlon of M and F was related to gender differences
In the set of 11 self-concepts by correlating the values of the 11
correlations (exclud~ngthe gender correlation) in each of the flrst
three columns wrth the corresponding 11 correlations $ n the fourth
column. The results, -.81 iposltlve %terns). -.A1 (negative WF
items) and -.71 (total) were all statlstlcally siqniflcant ip i

.Q5). When Spearman's rho, a nonparametrlc correlation coefflclent

was used. the corresponding correlations were -.75. -.55, and -.75
(all signlflcant at p < -05). These subsequent analyses demonstrate
that arees of self-concept in which F contributed relatively more
positively than W arc the greas In whlch females have higher self-
concepts than males, and the similar~tyIn the patterns 1s both
strong and consistent across responses to posltlvely and negatively
valued MF items.
+ p r -05; ** p r -01

scores (i.e., each score is given a weight of + 1or - 1)are able to account
for variance in the multiple self-concepts. Across the 11 areas of
self-concept 18.1% of the variance in self-concept scores is explained by
the four empirically weighted ASRS scores (Table 3). While no a priori
linear weighting of the four scores can do any better than this
optimum, one that approaches it would be strongly supported. Aver-
aged across the 11self-concept scores, Mtot (M+ + M-) and Ftot (F+
-t F-) explain 3.8% and 1.6% of the variance respectively, while the

110 MULTIVARIATE BEHAVIORAL RESEARCH

 H. W. Marsh

positive total (Pos = M-t + F+) and the negative total (Neg = M- +
F-) explain 8.6% and 3.7%of the variance (Table 5). The unweighted
bipol.ar MF (sum of all M items minus the sum of all F items) explains
only 3.6% (Table 4) of the variance in self-concept responses, the
unweighted bipolar positive-negative scale (PN; sum of socially desir-
able items minus the sum of socially undesirable items) scale explains
11.9%, and the sum of these two bipolar scales explains 10.1%.
Heilbm's (1976) androgyny score that combines the additive and
balance notions of androgyny was also computed, but it explains only
2.1% of the variance. These findings demonstrate that the social
desirability of the items-independent of whether they are M[ or F-
explains almost two-thirds of the predictable variance.

Table 5
Relations Rmong A Prior1 Combinat~onsof MF Scores and Self-concepts

Unweighted MF Scores
........................................................
e f g h
b c d Ri olar Bi olar Hi olar Gen
....................
Self-concepts Wtot
a
Ftot Pas .-------
Neg MF PR ME % PN Andra
Mathematics .Il* -05 .20* -.ox
Verbal .14* .OF?* .54* -.li*
General School Ibx . 0 .72+ -.I>?*

Physlcal .16* -.03 .27* -.IT*

Appearance T2x . -.09 .22* .oz.
Opposite S e x .26* - 4 .:t:w - .!>a+
Same Ser Il* . .Irlx .?n - .!IF?
Parent -. IT* .o8* .I?* -.24*
Honesty - .77* .XI* -24s - - --:.*
.2.

Emotional -04 -.27* .72* -.at*

General Self .24* .II* .49* -.12*
Mean Coeff. -106 -925 -781 -. 145
Meam 04 Squared
Coe*f . -018 .C37

NoteL Each of the set of 9 a prlorl unweighted combinat~onsof the +our MF

scores were correlated with each a+ the ;I self-cpnccpt scores.
x p - -05;
a -- ntot = (w+, t it?-,
b -- Ftot = + <F-'
(F+)
c -- Fos = (M+\ + (F+l
d -- Neg = ( Y - ) + (F-!
e -- Bipolar MF = Vtet - =tot
f -- Bipolar PN = Pos - Neg
g -- Bipolar MF 9 FN = Bipelar M F - Bipolar F N
h -- Gen Andro = (Ntot * Ftotj - : Wtot - Ftot '

JANUARY 1987
H. W. Marsh

Empirically Weighted A Priori Combinations of MF Scores and Their

Relation to Self-concepts
In the last section, Mtot, Ftot, Pos, and Neg were correlated with
self-concepts, as were various a priori, unweighted combinations of
these scores. While some of these scores were able to explain substan-
tial portions of the self-concept variance, each performed substantially
poorer than the 18.1%optimum. If the two components that comprise
each of these scores are empirically weighted, their empirically de-
rived combination must perform a t least as well as the corresponding
unweighted total but still no better than the 18.1%optimum. Thus, for
example, the variance explained by the empirically weighted combi-
nation of the Pos and Neg scores must fall between the 11.9%obtained
with the unweighted combination and the 18.1%optimum. Whether or
not it is still reasonable to characterize the empirical combination as
bipolar will depend on the size, and particularly the sign of the derived
weights, and their consistency across the different self-concept scores.
The empirically weighted combination of Mtot and Ftot (Table 6)
does modestly better than its bipolar counterpart (5.5% vs. 3.6%).
However, the interpretation of the empirically defined variable as
bipolar is dubious. Beta weights for Mtot and Ftot (Table 7) are
sometimes positive and sometimes negative; the two beta weights for
the same self-concept score sometimes have the same sign and some-
times the opposite sign. Hence, the score represented by this empirical
combination of Mtot and Ftot is not a bipolar score.
The empirically weighted combination of Pos and Neg (Table 6)
performs marginally better than its unweighted bipolar counterpart
(13.8%vs. 11.9%).An inspection of the beta weights (Table 7) provides
clear support for the bipolarity of the empirically defined variable. In
every instance the beta weight for Pos is statistically significant and
positive (mean beta = 0.309), while the beta weight for Neg is negative
and usually statistically significant (mean beta = -0.193). While the
empirically derived weights suggest that the Pos component is some-
what more important than the Neg component, it seems reasonable to
characterize its effect on self-concept as bipolar, and the unweighted
bipolar score accounts for most of the variance in self-concepts that is
explicable by its empirically weighted counterpart (11.9113.8 = 86%).
The empirically weighted score representing the bipolar MF and
the bipolar PN performs substantially better than its unweighted
counterpart (15.6%vs. 10.1%).A major problem with the unweightad
sum of the two bipolar scores was the assumption that a positive
weight should be assigned to the bipolar MF score (i.e., that M
112 MULTIVARIATE BEHAVIORAL RESEARCH
 H.W. Marsh

Table 5
M u l t i p l e Correlations R e l a t l n g Em 1 r ; c a l l y Weighted C o m b ~ n a t l o n sOf
MF S c o r e s t o Mu1 t i p l e S e l f - c o n c e p f s
__-____~__&__i_____-------------------------------------------A----

Mu1 t i p l e R e g r e s s i o n s Based On:

a
B i p o l a r B i p o l a r Sum of Bipolar
MFR! PN. MF Zt PN, Sum
Mtot Po5 B i p o l a r Mtot, Dif f MF R Dif f
Self-Concepts % Ftot & Neg PN Ftnt of MF of MF
................................................................... ?d

Mathematics .ll*
Verbal .16*
General School .17*
P h y s i c a l A b i l i t y .17*
Appearance .34*
O p p o s i t e Sex .27*
Same Sex .l4*
P a r e n t R e l a t i o n s .15*
Honksty .36* .44* .54* .54* .05+ .45*
Emotional .28* .50* .56* .58* .05* .45*
General S e l f -25s .53* .49* .54x .26* .S3*
Mean Coef f . -219 -358 -381 -405 1 .:57
Wean of Squared
Coef f . -055 ,138 .15h -175 -025 -135

Npf%, A series of m u l t l p l e r e g r e s s i o n s w e r e c o n d u c t e d i n w h ~ c h
v a r i o u s combinations of t w o o r t h r e e MF s c o r e s , t h o s e l i s t e d a t t h e
top. w e r e u s e d t o p r e d i c t e a c h of t h e 1 1 SDQ 1 1 s c a l e s . The
m u l t i p l e U s resulting f r o m e a c h of t h e s e m u l t i p l e regressions are
l x s t e d I n this t a b l e . a n d b e t a w e i g h t s r e s u l t z n g from s o m e mf t h e s e
a n a l y s e s a p p e a r i n T a b l e 7.

: p < -05
Sum MF is t h e sum of Mtot and F t o t , and Dif MF is t h e a b s o l u t e
d i f f e r e n c e b e t w e e n Mtot a n d F t o t . T h e s e are t h e t w o c o m p o n e n t s u s e d
t o d e f l n e t h e g e n e r a l i z e d a n d r o g y n y s c o r e (see T a b l e 5 f o r
d e f i n l t l o n of o t h e r s c o r e s ) .

necessarily contributes more positively to self-concept than does F). As

observed with the zero-order correlations between the bipolar MF and
self-concepts, the sign of the empirically derived weights is sometimes
negative. The viability of the bipolarity of MF assumed in this score
was further examined in an additional set of multiple regressions in
which Mtot, Ftot and bipolar PN were used to predict each self-concept;
weights for Mtot and Ftot are estimated independently, rather than
assuming a bipolarity. The results of this new analysis represent an
improvement over the two empirically weighted bipolar scores (17.5%
JANUARY 1987 113
H. W. Marsh

Table 7
Beta Weights For Three Multiple Re ressions Helatin Empirically
Wei hted Combinations of ~asculini?y/~emininity(MF? Scores to Multiple
Self-concepts As Described In Table 5
........................................................................
Beta Weights For Mu1 tiple Regressions Based On:
--L-i----------__---------*----------------------------

tot % Ftot
------------- Pos % Neg
------------- Bipolar PN, Wtot % Ftot
......................
Bipolar
Self-concepts Mtot Ftot Pos Neg PN Mtot Ftot
-------------_----_----------------------2---------------------&--------

Mathematics .10* .04 .21* -.Ct7+ .is+ .11+ -02

Verbal .13+ .O8* .37+ -.I&+ .35* .15+ -05
General School .15* .Cm .34+ -. 14+ .;2* .17% .02
Physical Ability .I&+ -.04 .29+ -.17x .32* .IS* -.07*
Appearance .33x -.13+ .22+ -.01 -19s .34+ -.
10r
Opposlte Sex .26+ -.07* .52* -.12x .30x .79+ -.10x
Same Sex .lC)* .I0x .32* -.14r .31+ -12 .07r
Parent Re1 at i ons -.13x .09+ -271 --
28+ .31* -.ll+ .07*
Honesty - .29+ .ZSx -301 -.38s .41+ -.27x .21+
Stability .09* -.27s .29+ -.45+ .51+ .12* -.32X
General Self .24* .09* .51* -.20+ .48* .2b+ -04
Mean Coef f . -104 -017 -309 -.I93 5 -7
.124 -.010
Wean of Squared
Coef f . .OJ9 .01B -102 -053 -122 -043 -017

&gfe, See note in Table 6 .

+ p < .05

vs. 15.6%;Table 6), and this result is sufficiently close to the 18.1%
optimum to indicate that it is able to account for most of the variance
in self-concepts that is related to responses to the MF items.
Heilbrun's generalized androgyny score is also composed of the
unweighted sum of two components: the sum of Mtot and Ftot and the
absolute difference between Mtot and Ftot. Multiple regression was
used to empirically estimate the weights for these two components
(Table 6), but the empirically defined score did little better than its
unweighted a priori counterpart (2.5% vs. 2.1%). Even when the
bipolar PN score was included in the multiple regression, the three
components (13.5%)did little better than the bipolar PN by itself
(11.9%).These results again fail to support the ability of Heilbrun's
androgyny score to explain relations between MF and self-concept
responses.

Summary and Implications

The purpose of this study was to examine MFIself-concept rela-

tions with respect to five theoretical models. Across all areas of
114 MULTIVARIATE BEHAVIORAL RESEARCH
 H.VV. Marsh

self-concept there was no support for the interactive or sex-typed

models, little support for the masculinity model, good support for the
additive androgyny, and particularly strong support for the diifferen-
tiated additive androgyny model. The lack of support for the interac-
tive and sex-typed models is particularly persuasive because of the
statistical power due to the extremely large sample size and the use of
the 4 x 4 design, because of the consistency of the lack of effect, across
the 11 facets of self-concept, and because the early adolescent period is
the period when the sex-typed effects are supposed to be strongest. In
support of the additive model, both M and F, particularly M+ and F+,
contributed positively to the prediction of self-concept. In support of
the differentiated additive model the relative contribution of N1 and F
varied substantially and predictably with the specific area of self-
concept. The contribution of F was more positive than M in the areas
of self-conceptwhere females had higher self-concepts,and the positive
contribution of M was greater in those areas where males had higher
self-concepts.
Across all MF scores there was weak support for the masculinity
model in that M tended to contribute more positively to self-concept
than F, even though the relative contribution of M and F varied
predictably with the particular facet of self-concept. However, most
previous research has considered just the positively valued MF items,
and for these the contribution of M+ and :F+ were both positive. The
bipolarity of MF failed to receive support from either of two tests of the
assumption in that Mtot and Ftot were nearly uncorrelated. and F
scores contributed uniquely and positively to self-concepts as well as M
scores.
The social desirability of MF items, independent of whethier they
were M or F, was an important determinant of the MFIself-concept
relations. This finding should not be interpreted as invalidity in the
responses to either instrument; social desirability was one basis for
selecting ASRS items, while self-endorsing socially desirable items
infers a positive self-concept. Indeed, if such a logical pattern did not
exist, then the validity of the instruments would be suspect. However,
this finding has important implications that, perhaps, have n.ot been
fully recognized. In particular, the apparent size of the MFIself-concept
relation will vary substantially depending on the social desirability of
the MF items-particularly if M and F are inferred from only socially
desirable items-and if M and F item13are not balanced in terms of
social desirability this imbalance may distort the apparent contribu-
tion of M and F to external constrwd;~.The EPAQ and ASRS reipresent
JANUARY 1987 115
H. W. Marsh

an improvement over the BSRI and PAQ in that the influence of social
desirability may be controlled by the inclusion of socially undesirable
characteristics. Nevertheless, these new instruments still comprise
items that maximize, rather than minimize, the influence of this
extraneous variable, and this represents a dubious test construction
practice (see Marsh & Myers, 1986, for further discussion).
The findings of the present investigation make an important
contribution to the study of MFIself-concept relations that are central
to androgyny theory because:
1. Previous research has found little or no positive contribution of
F after taking into account M, whereas F was shown to contribute
substantially in the present study;
2. Previous research has typically ignored the multidimensional-
ity of self-concept, whereas the present investigation demonstrated
that the MFIself-concept relation varies substantially with the specific
facet of self-concept;
3. Previous research has not considered positively and negatively
valued MF items, whereas the present study demonstrates that the
social desirability of MF items contributes substantially to MFIself-
concept relations.
4. The present investigation proposed and demonstrated strong
empirical support for a new model of the MFlself-concept relation, the
differentiated additive andragyny model.
The results of this study also contribute further support to the
construct validity of responses to the SDQ instruments and extend the
generality af the MarshlShavelson conclusions about the expected
pattern of relations between multiple dimensions of self-concept and
external constructs.
References
Antill, J . K., & Cunningham, J . D. (1979). Self-esteem as a function of masculinity i n
both sexes, Journal of Consulting and Clinical Psychology, 47, 783-785.
Antill, J . K., & Cunningham, J . D. (1980). The relationship of masculinity, femininity,
and androgyny to self-esteem. Australian Journal of Psychology, 32, 195-207.
Antill, J . K., Cunningham, J . D., Russell, G., & Thompson, N . L. (1981). An Australian
Sex-Role Scale. Australian Journal of Psychology, 33, 169-183.
Bassoff,E. S., & Glass, G. V . (1982). The relation between sex roles and mental health:
A meta-analysis of twenty-six studies. Counseling Psychologist, 10, 105-110.
Baurnrind, D. (1982). Are androgynous individuals more effectivepersons and parents.
Child Development, 53, 44-75.
Bern, S. L. (1974). The measurement of psychological androgyny. Journal of Consulting
and Clinical Psychology, 42, 155-162.
Bem, S. L. (1975). Sex ro-le adaptability: One consequence of psychological androgyny.
Journal of Personality and Social Psychology, 31, 634-643.
Bern, S. L. (1977). On the utility of alternative procedures for assessing psychological
androgyny. Journal of Consulting and Clinical Psychology, 45, 196-205.

116 MULTIVARIATE BEHAVIORAL RESEARCH

 H. W. Marsh

Burns, R. B. (1979). The self-concept: Theory, measurement, development and behaviour.

London: Longman.
Byme, B. (1984). The generallacademic self-concept nomological network: A review of
construct validation research. Review of Educational Research, 54, 427-4561.
Cohen, J., & Cohen, P. (1975). Applied multiple regressionlcorrelation analysis for the
behavioral sciences. Hillsdale NJ: Lawrence Erlbaum.
Cook, E. P. (1985). Psychological androgyny. New Pork: Pergamon.
Constantinople, A. (1973). Masculinity-feminity: An exception to a famous dictum?
Psychologzcal Bulletin, 80, 389407.
Dusek, J . B., & Flaherty, J. F. (1981). The development of self-concept d u r ~ ~ nthe
g
adolescent years. Monographs of the Society for Research in Child Deuelopnzent, 46
(4, Serial No. 191).
Flaherty, J. F., & Dusek, J. B. (1980). An investngation of the relationship between
psychological androgyny and components of self-concept. Journal of Personality and
Social Psychology, 38, 984-992.
Fleming, J. S., & Courtney, B. E. (1984). The dimensionality of self-esteem 11: Hierar-
chical facet model for revised measurement scales. Jolernal ofPersonality and Social
Psychology, 46, 404-421.
Hall, J. A., & Taylor, M. C. (1985). Psychological androgyny and the masculinity x
femininity interaction. Journal of Personality and Social Psychology, 49, 429435.
Harter, S. (1982). The Perceived Competence Scale For Children. Child Developnwnt, 53,
87-97.
Heilbrun, A. B, (1976). Measurement of masculine and feminine sex-role identities as
independent dimensions. Journal of Consulting and Clirzical Psychology, 44,
183-190.
Helmreich, R. L., Spence, J . T., & Holahan, C. K. (1979). Psychological androgyny and
sex role flexibility: A test of two hypotheses. Journal of Personality and Social
Psychology, 37, 1631-1644.
Hull, C. H., & Nie, N. H. (1981). SPSS Update 7-9. New York: Mcgraw-Hill.
Kelly, J., Caudill, S., Hathorn, S., & O'Brien, C. (1977). Socially undesira'ble sex-
correlated characteristics: Implications for androgyny and adjustment. Journal of
Cionsulting and Clinical Psychology, 45, 1186-1187.
Kelly, J., & Worrell, J. L. (1977). New formulations of sex roles and androgyny: A
critical review. Journal of Consulting and Clinical Psychology, 45, 1101-1115.
Kohlberg, L. (19661. A cognitive-developmental analysis of children's sex-role ~:oncepts
and attitudes. In E. E. Maccoby (ed.), The development of sex differences. Stanford,
CA: Stanford University Press.
Lamke, L. B. (1982). The impact of sex-role orientation on self-esteem in early adoles-
cence. Child Development, 53, 1530-1535.
Lubinski, D., Tellegen, A., & Butcher, J . W . (1983). Masculinity, feminin~ty,and
androgyny viewed and assessed as distinct concepts. Journal of Personality and
Social Psychology, 44, 428-439.
Marsh, H. W. (1985a). Age and sex effects in multiple dimensions of preadolescent
self-concept. Australian Journal of Psychology, 36, 197-204.
Marsh, H. W. (1985b). The structure of masc~linitylfemininity:An application of
confirmatory factor analysis Co higher-order factor structures and factorial
invariance. Multivariate Behavioral Research, 20, 427-449.
Marsh, H. W. (1986). Verbal and math self-concepts: An internavexternal frame of
reference model. American EducationaE Research Journal, 23, 129-149.
Marsh, H . W., Barnes, J., & Rocevar, D. (1985). Self-other agreement on multidimen-
sional self-concept ratings: Factor anal sis & multitrait-multimethod analysis.
Journal of Personality and Socral Psychofogy, 49,1560-1377.
Marsh, H. W., & Myers, M. R. (1986). Masculinity, femininity, and androgyny: A
methodological and theoretical critique. Sex Roles, 14, 397430.
Marsh, H. W., & Q'Niell, R. (1984). Self Description Questionnaire I11 (SDQ 111): The
construct validity of multidimensional self-concept ratings by late-adolescents.
Journal of Educational Measurement, 21, 153-174.
Marsh, H. W., Parker, J., & Barnes, J. (1985). Multidimensional adolescent self-
concepts: Their relationship to age, sex and academic measures. American Educa-
tional Research Journal, 22, 422-444.
JANUARY 1987 117
H. W. Marsh

Marsh, H . W., Relich, J . D., & Smith, I. D. (1983).Self-concept:The construct validity of

interpretations based upon the SDQ. Journal ofPersonality and Social Psychology,
45,173-187.
Marsh, H. W . , Richards, G., & Barnes, J. (1986). Multidimensional self-concepts: The
effect of participation i n an Outward Bound program. Journal of Personality and
Social Psychology, 50, 195-204.
Marsh, H. W . , & Shavelson, R. J. (1985). Self-concept: Its multifaceted, hierarchical
structure. Educational Psychologist, 20, 107-123.
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 Jourrial, 20,
333-357.
Marsh, H. W., Smith, I . D., & Barnes, J . (1984).Multidimensional self-concepts:Rela-
tionships to inferred self-concepts and academic achievement. Australian Journal of
Psychology, 36, 367386.
Marsh, H. W., Smith, I. D., Barnes, J., & Butler, S. (1983). Self-concept: Reliability,
dimensionality, validity, and the measurement of change. Journal of Educational
Psychology, 75, 772-790.
Nie, N. H,, Hull, C . H., Jenkins, J . G. Steinbrenner, K., & Bent, D. H. (1975).Statistical
Package for the Social Sciences. New York: McGraw-Hill.
Pedhauzer, E. J. (1982).M u l ~ p l eregression in behavioral research (2nd ed.). New York:
Holt, Rinehart & Winston.
Pedhauzer, E. J., & Tetenbaum, T . J. (1979).The Bern Sex-Role Inventory: A theoretical
and methodological critique. Journal of Personality and Social Psychology, 37,
- .. - - -
996-1016-

Rosenbarg, M. (1965). Society and the adolescent child. F'rinceton, NJ: Princeton
University Press.
Russell, G., & Antill, J. (1984).An Australian Sex-Role Scale: Additional psychometric
data and correlations with self-esteem. Australian Psychologist, 19, 13-18.
Shavelson, R,, Hubner. J . J., & Stanton, G. C. (1976). Self-concept: VaIidation of
construct interpretations. Review of Educational Research, 46,407-441.
Shavelson, R. J., & Marsh, H, W . (1986).On the structure of self-concept (pp. 305330).
In R. Schwarzer (Ed.),Anxiety and cognitions. Hillsdale, NJ: Lawrence Erlbaum.
Soares, L. M.,& Soares, A. T . (1982, July). Convergence and discriminatipn in academic
self-concepts. Paper presented at the 20th Congress of the International Association
of Applied Psychology, Edinburgh, Scotland.
Spence, J. T. (1984). Masculinity, femininity, and gender-related traits: A conceptual
analysis and critique of current research (p. 1-97). In B. A. Maher & W . B. Maher
(Eds.),Progress in Experimental Personality Research Wol. 13). New York: Aca-
demic Press.
Spence, J. T., Helmreich, R. L., & Holahan, C. K. (1979).Negative and positive compo-
nents of masculinity and femininity and relations to self-reports of neurotic and
actihg out behaviors. Journal of Personality and Social Psychology, 37, 1673-1682.
Taylor, M. C., & Hall, J . A. (1982),Psychological androgyny: A review and refdrinula-
tion of bheories, methods and conclusions. Psychological Bulletin, 92, 347-366.
Welles, L. E.,& Marwell, G. (1976),Self-esteem: Its conceptualizdtwn and measurement.
Beverly Hills, CA: Sage Publications.
Whiteley, B. E. (1983). Sex role orientation and self-esteem: A critical meta-analytic
review. Journal of Personality and Social Psychology, 44, 765-778.
Wylie, R. C . (1974). The self-concept (Rev. ed., Vol. 1) Lincoln: University of Nebraska
Press.
Wylie, R. C. (1979).The self-concept (Vol. 2) Lincoln: University of Nebraska Press.

MULTIVARIATE BEHAVIORAL RESEARCH