Sie sind auf Seite 1von 11

Journal of Counseling Psychology © 2013 American Psychological Association

2013, Vol. 60, No. 2, 228 –238 0022-0167/13/$12.00 DOI: 10.1037/a0031545

Male Role Norms Inventory–Short Form (MRNI-SF):


Development, Confirmatory Factor Analytic Investigation of Structure,
and Measurement Invariance Across Gender
Ronald F. Levant, Rosalie J. Hall, and Thomas J. Rankin
The University of Akron

The current study reports the development from the Male Role Norms Inventory–Revised (MRNI-R;
Levant, Rankin, Williams, Hasan, & Smalley, 2010) of the 21-item MRNI-Short Form (MRNI-SF).
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

Confirmatory factor analysis of MRNI-SF responses from a sample of 1,017 undergraduate participants
This document is copyrighted by the American Psychological Association or one of its allied publishers.

(549 men, 468 women) indicated that the best fitting “bifactor” model incorporated the hypothesized
7-factor structure while explicitly modeling an additional, general traditional masculinity ideology factor.
Specifically, each item-level indicator loaded on 2 factors: a general traditional masculinity ideology
factor and a specific factor corresponding to 1 of the 7 hypothesized traditional masculinity ideology
norms. The bifactor model was assessed for measurement invariance across gender groups, with findings
of full configural invariance and partial metric invariance, such that factor loadings were equivalent
across the gender groups for the 7 specific factors but not for the general traditional masculinity ideology
factor. Theoretical explanations for this latter result include the potential that men’s sense of self or
identity may be engaged when responding to questions asking to what extent they agree or disagree with
normative statements about their behavior, a possibility that could be investigated in future research by
examining the associations of the general and specific factors with measures of masculine identity.
Additional exploratory invariance analyses demonstrated latent mean differences between men and
women on 4 of the 8 factors, and equivocal results for invariance of item intercepts, item uniquenesses,
and factor variances-covariances.

Keywords: Male Role Norms Inventory (MRNI-R), Male Role Norms Inventory–Short Form (MRNI-SF),
traditional masculinity ideology, measurement invariance, masculine identity

Levant et al. (1992) developed the Male Role Norms Inventory hypothesized dimensionality of the MRNI-R was provided by an
(MRNI), a 57-item instrument with seven subscales that measure exploratory factor analysis (EFA) described in a second study of
the norms of traditional masculinity ideology and one subscale that the MRNI-R (Levant, Rankin, Williams, Hasan, & Smalley, 2010).
measures nontraditional male norms. The MRNI has been used However, Levant et al. (2010) found some items were not associ-
in over 40 empirical studies, including multicultural investigations ated as expected with their theorized subscales, necessitating re-
in the United States (examining masculinity ideologies in African thinking of the meaning and names of some subscales. Specifi-
American, Latino(a), Asian American, and European American cally, the results suggested a slightly revised seven-factor
communities), cross-national studies (in Russia, China, and Paki- structure, based on 39 items: Avoidance of Femininity, Negativity
stan), and studies in which the relationships between masculinity toward Sexual Minorities, Self-reliance through Mechanical Skills
ideologies and other constructs have been examined (see also (formerly Self-reliance), Toughness (formerly Aggression), Dom-
Levant & Richmond, 2007, for a review of the MRNI studies inance, Importance of Sex (formerly Non-relational Attitudes to-
reported over the 15-year period from 1992 to 2007). ward Sexuality), and Restrictive Emotionality. The 2010 study
In 2007, a revision of the MRNI made several significant
also supported the internal consistency of the MRNI-R total score
changes, resulting in an instrument called the Male Role Norms
and the seven-factor analytically derived subscale scores, and
Inventory–Revised (MRNI-R; Levant et al., 2007). Support for the
provided evidence of validity.
The current study reports the development of a Short Form of
the MRNI-R, called the MRNI-SF. Short forms of masculinity
This article was published Online First February 18, 2013. instruments, such as the Conformity to Masculine Norms Inven-
Ronald F. Levant, Rosalie J. Hall, and Thomas J. Rankin, Department of tory (Parent & Moradi, 2009) and the Gender Role Conflict Scale
Psychology, The University of Akron. (Wester, Vogel, O’Neil, & Danforth, 2012) have appeared re-
We gratefully acknowledge the assistance of Samantha Gregus, Britney cently, indicating an interest in instruments with good psychomet-
Kurtz, Zac Brubach, Amanda Beck, Noelle Connell, Gregory Harrison,
ric properties that can be completed quickly. Given the centrality
Karolina Kristof, Allison Miller, Baron Rogers, and Eric Simon, all student
research assistants at the University of Akron.
of traditional masculinity ideology to the Gender Role Strain
Correspondence concerning this article should be addressed to Ronald F. Paradigm (Levant, 2011), it seems important to have a short
Levant, Department of Psychology, The University of Akron, Akron, OH multidimensional scale with good psychometric properties that
44325-4301. E-mail: Levant@uakron.edu measures this construct. The initial step in our development of the

228
MALE ROLE NORMS INVENTORY–SHORT FORM 229

MRNI-SF was the selection of three of the highest loading item- endorsement of traditional masculinity ideology may reflect dif-
level indicators for each latent variable, based on the Levant et al. ferences in gender empowerment between these countries. Women
(2010) EFA of the MRNI-R. Care was taken in selecting the items in countries with lower empowerment may feel that they have no
to ensure that their content minimized redundancy while reflecting choice but to subscribe to a traditional view of masculine norms,
the desired construct. Next, a new sample was collected in the which presupposes men’s dominance over women. According to
current study to assess— using confirmatory factor analysis (CFA) the Gender Empowerment Measure published by the United Na-
techniques—whether this smaller subset of items comprising the tions Development Programme (2007– 08), the United States is
MRNI-SF retained the hypothesized seven-factor structure and ranked 12 out of 177 nations in gender empowerment, whereas
desired pattern of factor loadings in a sample of college men and China, Russia, and Pakistan are ranked 81, 67, and 136, respec-
women. tively. The relatively high ranking on gender empowerment in the
In addition to the more specific subscale scores that corre- United States is the result of five decades of feminist influence on
sponded to the dimensions identified by the previous factor anal- culture and legislation. U.S. women’s low endorsement of tradi-
ysis, measures that were predecessors of the MRNI-SF (i.e., the tional masculinity ideology may thus reflect their rejection of male
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

MRNI and the MRNI-R) were used to create a total score intended dominance (Levant, 1996, 1997).
This document is copyrighted by the American Psychological Association or one of its allied publishers.

to broadly represent traditional masculinity ideology in general. These previous results regarding mean differences across gen-
Ideally, this more general latent construct also could be repre- ders in the endorsement of traditional masculinity ideology are
sented in our CFA model of the newly created MRNI-SF, along limited by the fact that they assumed, but did not assess, that the
with the seven more specific dimensions. Two potential alternative MRNI instrument operated in the same manner for men and
model specifications accommodating a broad general factor were women, that is, was equivalent or invariant across the genders.
considered: (a) a hierarchical model in which a general traditional Measurement invariance indicates whether scores from a measure
masculinity ideology latent factor was specified at the higher level, that operationalizes a construct have the same meaning over dif-
thus implying that it causes each of the seven more specific lower ferent populations, for example, those defined by gender (Meade
level latent factors, and (b) a “bifactor” model in which responses & Lautenschlager, 2004; Meredith, 1993). When comparing mean
to each measured indicator (i.e., item-level response) were mod- differences, invariance over populations must be present in order
eled as caused by both a general traditional masculinity ideology to rule out the possibility of “construct bias,” which implies that a
latent factor and a specific latent factor corresponding to one of the test measures something different in one group (e.g., men) than in
subscales. This latter model is similar to those used in modeling another (e.g., women).
the structure of measures of intelligence, in which specific subtest It is indeed possible that traditional masculinity ideology means
scores are modeled as being caused both by a general intelligence something different to men than it does to women. Thus, the
factor, g, and specific factors for particular abilities such as verbal second goal of the current study was to assess the measurement
reasoning. Thus, in the current study, the fit of the MRNI-SF data invariance of the MRNI-SF across the two gender groups. The
was assessed with respect to a seven-factor model without an assessment of measurement invariance is a multistep process, with
additional broad factor, as well as the two alternative models. various levels (e.g., full vs. partial) and forms of invariance po-
tentially present. We used a series of CFA models to assess the
Do Men and Women Construe Traditional following hypothesized forms of invariance (Horn & McArdle,
1992; Meredith, 1993): (a) configural invariance, a weak form of
Masculinity Ideology Similarly?
factorial invariance, which specifies that the same factor dimen-
One topic that has been only partially addressed in prior re- sionality and pattern of fixed and free factor loadings holds across
search is how both genders construe masculinity ideology. As groups and (b) (at least partial) metric invariance, which is a
Whorley and Addis (2006) stated: “when we study masculinity stronger form of factorial invariance in that the unstandardized
only in men we can easily fall prey to an implicit essentialism by factor loadings are proposed to be equal across groups. Results
failing to distinguish . . .[sex and gender]; thoroughly studying from the metric invariance model were also used to address the
masculinity means understanding how it operates in the lives of issue of whether men and women had equal factor means. Stronger
both men and women” (p. 656). Differential endorsements of forms of invariance were also explored using CFA models, al-
masculinity ideology by men and women have been investigated. though we did not hypothesize them, and view them as less critical
For example, Levant and Richmond (2007) summarized four stud- for our purposes, namely, (a) scalar invariance, which concerns
ies using the MRNI total traditional score with U.S. participants invariance of the regression intercepts; (b) invariance of unique-
from African American, Latino(a), and European American com- nesses, which are the residuals of the indicators; and (c) invari-
munities. They found that men tended to endorse traditional mas- ances of factor variances/covariances. In general, invariance across
culinity ideology to a greater extent than did women and that, gender for a given parameter was supported when the relative fit of
within the United States, the effect size of gender was larger than a more parsimonious invariance model was not significantly
that of race/ethnicity. poorer than that of a less constrained model, which did not force
However, a set of cross-national studies using the MRNI total gender equivalence on that parameter.
traditional score found that both men and women from China, As noted above, we first assessed the relative fits of the seven-
Russia, and Pakistan tended to endorse traditional masculinity factor, hierarchical, and bifactor models and then conducted the
ideology to a greater extent than did U.S. men and women, that the invariance analyses with the best fitting model of these three. We
differences between men and women in these countries were much should note that there was a certain initial conceptual appeal to the
smaller than for their U.S. counterparts, and that nationality had a bifactor model, particularly when one considers invariance. In
larger effect size than gender. These nation-level differences in light of the studies (limited though they may be) that revealed
230 LEVANT, HALL, AND RANKIN

gender differences in the endorsement of traditional masculinity and were subsequently e-mailed the link to the online survey site.
ideology using the MRNI total traditional scale in U.S. samples, The first page of the site reviewed the informed consent informa-
we did not expect to find invariance on the general traditional tion. Participants who consented clicked “yes” and were taken to
masculinity latent factor, whereas we did expect to find it on the the survey. Measures were presented in the following order: De-
specific latent factors. That is, men and women are likely to have mographic Form (sex/gender, race/ethnicity, age, relationship sta-
very similar understandings of the specific norms for the male role, tus, sexual orientation, family/household income, and socioeco-
as these are expected to be pervasive within a culture—thus nomic status), MRNI-R. Upon completion of the survey,
suggesting that we would find measurement invariance on the participants were provided with an educational briefing on the
specific latent factors. However, we also speculate that when men study rationale and hypotheses.
respond to a measure of traditional masculinity ideology, they may
do so with a different mind-set than do women, potentially result-
Measure
ing in some degree of measurement invariance across genders.
Compared with women, when men respond to questions asking to The MRNI–Short Form (MRNI-SF). The MRNI-SF is a
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

what extent they agree or disagree with statements about how men short measure of traditional masculinity ideology developed in the
This document is copyrighted by the American Psychological Association or one of its allied publishers.

should think, feel, and behave, their sense of self or identity may current study from the item pool of the longer MRNI-R; thus, we
be more likely to become engaged. These questions may provoke first briefly describe this parent instrument. The MRNI-R has
consideration not only of the cultural norms for men in general, but seven-factor analytically derived subscales: Avoidance of Femi-
also the thoughts about themselves, and what they personally are ninity, Negativity toward Sexual Minorities, Self-reliance through
likely to think, feel, or do. Mechanical Skills, Toughness, Dominance, Importance of Sex,
and Restrictive Emotionality. As reported by Levant et al. (2010),
Method alpha coefficients for the MRNI-R ranged from .75 to .92 for
subscale scores, and .96 for the total scale score. Analyses of
men’s responses provided evidence for convergent validity based
Participants on the significant correlation of the MRNI-R with the Male Role
Participants were recruited from psychology, computer science, Attitudes Scale (Pleck, Sonenstein, & Ku, 1994), another measure
and physics courses (the latter two disciplines were selected to of traditional masculinity ideology. Evidence for discriminant va-
increase the number of male participants) from a large, public, lidity was found through the nonsignificant correlation with the
midwestern university in three collections of data during the Masculinity Scale of the Personal Attributes Questionnaire
2009 –2011 academic years. Eleven hundred eighty participants (Spence & Helmreich, 1978), which fundamentally differs from
began the study, and 1,017 (549 men, 468 women) completed it, the MRNI-R because it measures masculine personality traits as
for a completion rate of 86.2%. We estimate that approximately contrasted with normative expectations for masculine behavior.
3,000 students were solicited, so that about 39% of those solicited Evidence was also found for concurrent validity through the sig-
agreed to participate. Most participants identified as White/Euro- nificant correlations with three scales that measure related con-
pean American (82.9%), although 9.1% identified as Black/Afri- structs: the Conformity to Masculine Norms Inventory (Mahalik et
can American, and 6.7% identified as either Latino(a)/Hispanic, al., 2003), the Gender Role Conflict Scale (O’Neil, Helms, Gable,
Asian/Asian American, American Indian, Pacific Islander/Inuit, David, & Wrightsman, 1986), and the Normative Male Alexithy-
Middle Eastern, or bi/multiracial. The remainder checked “other” mia Scale (Levant et al., 2006).
or did not respond. Ages ranged from 18 to 59 years, with a mean The MRNI-SF was developed by selecting three of the highest
of 21.18 years (SD ⫽ 5.00). The median age was 20, and the modal loading items from each subscale of the MRNI-R as reported in
age was 19. Most participants (94.8%) identified as heterosexual, Levant et al. (2010). Items were chosen to capture the specific
but 2.1% identified as gay/lesbian, 1.8% identified as bisexual, and construct and to avoid redundancy. The resulting set of 21 items
1.4% did not respond to this question. Most of the participants (see Table 2 for item wordings) was expected to show a similar
indicated that they were in relationships, either dating one person seven-factor dimensionality to the MRNI-R. Responses to items
exclusively (37%) or married/partnered/engaged (13.4%); yet, are made on a 7-point Likert-type scale (1 ⫽ strongly disagree,
17.6% were engaged in casual nonexclusive dating, 29.9% were 7 ⫽ strongly agree). Higher scores indicate higher levels of
not currently dating anyone, and 1.8% were divorced or separated. endorsement of traditional masculinity ideology. In the current
The median family/household income was $40,001–$60,000. In study, alphas for subscale scores ranged from .79 to .90 for men,
terms of self-identified socioeconomic status, most (54.2%) iden- and from .75 to .88 for women; alphas for the total scale were .92
tified as middle class, 20.7% as lower middle class, 6.7% as lower for men and .94 for women.
class, 15.8% as upper middle class, and .9% as upper class.
Analytic Approach
Procedure
In the current study, CFA was implemented with the Mplus
The study was approved by the university Institutional Review Version 6 (Muthén & Muthén, 1998 –2008) structural equation
Board. Undergraduate students were solicited and offered extra modeling (SEM) software. Maximum likelihood robust (MLR)
credit for their participation in the study, which involved filling out estimation was used to accommodate any nonnormality in the data.
a web-based survey using a survey utility licensed by the univer- Analyses were first performed to confirm the seven-factor mea-
sity. Students who wished to participate provided their e-mail surement model for the total sample, then to assess whether either
address to the research assistant who had visited their classroom, of two proposed alternative measurement structures (i.e., the hier-
MALE ROLE NORMS INVENTORY–SHORT FORM 231

archical and bifactor models) explicitly incorporating a traditional data were only mildly nonnormally distributed, with values of
masculinity factor provided a better fit to the data. The best fitting skew ranging from ⫺.77 to 1.27 and values of kurtosis ranging
model of these three was then used as a basis for specifying from ⫺1.10 to .90. Hence (as noted above), we used the MLR
multigroup models testing measurement invariance across men estimation procedure for all analyses and reported adjusted chi-
and women. square statistics and corrected chi-square difference tests.
The overall fit of the SEM models was assessed with the
chi-square goodness-of-fit statistic. However, this statistic may be
CFA of Total Responses to the MRNI-SF
overly sensitive to minor and theoretically uninteresting sources of
model misfit, especially when sample sizes are large, as in the The first CFA model that was tested (Model SG1) was a
current study. Thus, we also used a set of alternative fit indices that single-group model, fit to a data set that combined responses of the
are typically consulted to determine whether a model demonstrates male and female participants. In this model, participant responses
adequate fit (Kahn, 2006). These indices and the criteria used to to the 21-item MRNI-SF were used as indicators of seven hypoth-
assess their values (see Kline, 2011) were the (a) comparative fit esized factors. Each item-level indicator loaded on only one factor,
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

index (CFI), where values of ⱖ .90 indicate reasonable fit; (b) for a total of three indicators per factor. The purpose of this model
This document is copyrighted by the American Psychological Association or one of its allied publishers.

root-mean-square error of approximation (RMSEA), where good was to determine whether the seven-factor dimensionality of the
model fit is suggested by values of .05 or lower, and values MRNI-R was maintained in this new, short form of the measure,
between .05 and .08 suggest reasonable fit; and (c) standardized and whether the items loaded on factors in a manner consistent
root-mean-square residual (SRMR), where values of less than .10 with Levant et al.’s (2010) earlier MRNI-R results in another data
are considered good. As Kline (2011) noted, these criteria are rules set (see Figure 1 and Table 2 to determine which specific items
of thumb, and the indices should be regarded as providing descrip- were specified to load on a given factor).
tive information about model fit, the value of which “. . . increases The resulting chi-square goodness-of-fit statistic for the seven-
when you report values of indices that as a set assess model fit factor model was statistically significant, ␹2(168) ⫽ 680.060, p ⬍
from different perspectives . . .” (p. 205). .0001, indicating that the null hypothesis of perfect fit should be
Relative fits of relevant single- or multigroup SEM models were rejected. However, the remaining indices were all within the
compared with each other using (a) chi-square difference tests guidelines described earlier and suggested that the model fit ade-
(adjusted to take into consideration the use of the MLR estimator; quately, CFI ⫽ .949, RMSEA ⫽ .055, 90% CI [.050, .059],
see Satorra & Bentler, 1999, and instructions at the Mplus website SRMR ⫽ .041. Considering these criteria, a seven-factor solution
by accessing http://www.statmodel.com/chidiff.shtml), (b) change for the 21-item MRNI-SF appeared quite plausible. The fit statis-
in CFI, and (c) a comparison of the values of the Bayesian tics for this and all subsequently tested single-group models are
information criterion (BIC)—a parsimony-adjusted predictive in- displayed in Table 1.
dex of fit. If the fit of a model is not worsened by imposing Next, we assessed the two alternative structures (i.e., hierarchi-
equality constraints on the factor loadings or a similar parameter cal and bifactor), both of which explicitly added a traditional
such as the intercept, mean or variance, then the more constrained masculinity ideology factor to the initial model’s seven-factor
model is preferred to the less constrained model because it is more structure. The hierarchical factor model (SG2A) had two levels of
parsimonious. For example, a more parsimonious model would be latent constructs. The lower level consisted of the initial seven
one in which a single set of factor loadings suffices for both factors, and the higher level consisted of a single traditional
genders, rather than a separate set of loadings for each gender. The masculinity ideology latent construct whose indicators were the
chi-square difference test allows an assessment of whether the seven lower level factors. This model is more parsimonious than
decrement in fit associated with a more constrained model is the initial seven-factor model, because it replaces the set of 21 free
statistically significant. If it is not statistically significant, the more covariances among the seven factors with the set of seven higher
parsimonious model would be preferred. However, the chi-square order factor loadings. The hierarchical model implies that all
difference test may be too sensitive to misfit when sample sizes possible bivariate associations among the set of seven lower order
are large. Thus, two alternative indices were also consulted: the factors can be explained through their common association with
change in CFI (Cheung & Rensvold, 2001) and a comparison of the single higher order factor (which represents the commonality
the BIC values for the two models. Support for a more parsimo- across different traditional masculine norms, and may be inter-
nious equality-constrained model would consist of a change in CFI preted as a broad traditional masculinity ideology construct).
of less than .01 and a smaller BIC value for the constrained model. Overall, the hierarchical SG2A model had a borderline fit,
␹2(182) ⫽ 1043.874, p ⬍ .0001, CFI ⫽ .915, RMSEA ⫽ .068,
90% CI [.064, .072], SRMR ⫽ .067. However, the factor loadings
Results
of the seven lower level factors on the single, higher order tradi-
The data were screened before conducting statistical analyses to tional masculinity ideology factor were all strong and statistically
eliminate random responding and data entry errors. A low level of significant, ranging in standardized values from .67 to .88, indi-
missing data was observed (zero to 18 missing responses per item, cating that the indicators shared substantial common variance that
with no evident patterns of nonresponse), and the average number spanned the seven traditional masculinity domains.
of missing responses per participant was .035. Because the rec- The hierarchical SG2A model is nested in the seven-factor SG1
ommended practice of full information maximum likelihood esti- model, and thus the relative fits of these two models can be
mation procedures was used (Schlomer, Bauman, & Card, 2010), compared by determining the statistical significance of the differ-
no cases were deleted and no missing values were imputed; rather, ence in chi-square values for the two models. The hierarchical
all available responses for each item were used in the analysis. The SG2A model places more constraints on the relationships among
232 LEVANT, HALL, AND RANKIN
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
This document is copyrighted by the American Psychological Association or one of its allied publishers.

Figure 1. Preferred bifactor structure showing seven specific factors and one general factor. Factor indicators
show relevant Male Role Norms Inventory–Revised item numbers. RE ⫽ Restrictive Emotionality; SR ⫽
Self-reliance through Mechanical Skills; NT ⫽ Negativity toward Sexual Minorities; AF ⫽ Avoidance of
Femininity; IS ⫽ Importance of Sex; Do ⫽ Dominance; T ⫽ Toughness; TM ⫽ traditional masculinity ideology.
The set of seven specific factors are allowed to freely covary; the general factor is fixed to be orthogonal to the
seven specific factors.

the seven lower order factors than does the seven-factor SG1 of .010. Finally, the BIC value was smaller (more desirable) for
model. Because of this, the hierarchical model would be expected the seven-factor SG1 model. These results suggest not only that
to show a poorer fit to the data; hence, if the restrictions lead to a the SG2A model does not fit well in an absolute sense but also
significantly poorer fit for the hierarchical model, the seven-factor that it fits significantly more poorly than does the seven-factor
model is preferred. However, if the observed decrement in the fit SG1 model. On the basis of modification indices, we also
of the hierarchical model is not statistically significant, then it estimated an alternative hierarchical Model SG2B identical to
would be preferred on the basis of parsimony. SG2A, except it freed three pairs of covariances between first-
The chi-square difference between the SG2A and SG1 models order factor disturbances; however, this revised model still also
was statistically significant, ⌬␹2 ⫽ 343.752, ⌬df ⫽ 14, p ⬍ .0001, fit significantly more poorly than Model SG1 (p ⬍ .001; see
and in addition, the ⌬CFI was .035, a value greater than the cutoff Table 1 for details).

Table 1
Model Fit Statistics and Comparisons of Nested Single-Group Models

Single-group model ␹2(df) CFI RMSEA estimate and 90% CI SRMR BIC

SG1: Seven-factor 680.060 (168) .949 .055 [.050, .059] .041 70294.163
SG2A: Hierarchical 1043.874 (182) .915 .068 [.064, .072] .067 70639.671
SG2B: Hierarchical ⫹
(covary factor disturb) 822.410 (179) .936 .059 [.055, .064] .055 70393.036
SG3: Bifactor 478.725 (147) .967 .047 [.042, .052] .030 70165.125

⌬␹2(df) p ⌬CFI Conclusion


Model comparison:
SG2A vs. SG1 343.752 (14) ⬍.001 .035 Prefer SG1
SG2B vs. SG1 135.444 (11) ⬍.001 .014 Prefer SG1
SG1 vs. SG3 162.284 (21) ⬍.001 .017 Prefer SG3
Note. In the model comparison section, the more parsimonious model is listed first. The conclusion is based on a joint consideration of ⌬␹2, ⌬CFI, and
a comparison of BIC values. CFI ⫽ comparative fit index; RMSEA ⫽ root-mean-square error of approximation; CI ⫽ confidence interval; SRMR ⫽
standardized root-mean-square residual; BIC ⫽ Bayesian information criterion.
MALE ROLE NORMS INVENTORY–SHORT FORM 233

Table 2
Factor Loadings and Intercorrelations From Single-Group Confirmatory Factor Analysis of the MRNI-SF Using the Bifactor Model

General and specific factors


Scale and item TM F1 F2 F3 F4 F5 F6 F7

F1: Restrictive Emotionality (RE)


38. A man should never admit when others hurt his feelings. .51 .66
41. Men should be detached in emotionally charged situations. .43 .63
53. Men should not be too quick to tell others that they care about them. .52 .53
F2: Self-Reliance through Mechanical Skills (SR)
13. Men should have home improvement skills. .48 .79
14. Men should be able to fix most things around the house. .53 .74
36. A man should know how to repair his car if it should break down. .42 .48
F3: Negativity toward Sexual Minorities (NT)
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

1. Homosexuals should never marry. .51 .59


This document is copyrighted by the American Psychological Association or one of its allied publishers.

8. All homosexual bars should be closed down. .56 .69


25. Homosexuals should never kiss in public. .68 .47
F4: Avoidance of Femininity (AF)
7. Men should watch football games instead of soap operas. .70 .35
15. A man should prefer watching action movies to reading romantic novels. .67 .17
19. Boys should prefer to play with trucks rather than dolls. .83 .68
F5: Importance of Sex (IS)
16. Men should always like to have sex. .58 .49
20. A man should not turn down sex. .51 .55
43. A man should always be ready for sex. .56 .63
Dominance (Do)
2. The President of the U.S. should always be a man. .52 .60
3. Men should be the leader in any group. .44 .85
21. A man should always be the boss. .56 .58
Toughness (T)
42. It is important for a man to take risks, even if he might get hurt. .44 .42
45. When the going gets tough, men should get tough. .59 .49
48. I think a young man should try to be physically tough, even if he’s not big. .61 .46
Factor intercorrelations
F1: Restrictive Emotionality — ⫺.06 .24ⴱ .18 .39ⴱ .49ⴱ .34ⴱ
F2: Self-Reliance through Mechanical Skills — ⫺.15 .34 .17 ⫺.05 .40ⴱ
F3: Negativity toward Sexual Minorities — .02 ⫺.19 .45ⴱ ⫺.18
F4: Avoidance of Femininity — .10 .09 .20
F5: Importance of Sex — .27ⴱ .48ⴱ
F6: Dominance — .13
F7: Toughness —
Note. Item numbers refer to the number of the item in the Male Role Norms Inventory–Revised. Standardized factor loadings are reported. All factor
loadings are statistically significant at p ⬍ .05. All factor intercorrelations marked with an asterisk are statistically significant at p ⬍ .05. MRNI-SF ⫽ Male
Role Norms Inventory–Short Form; TM ⫽ general traditional masculinity ideology factor.

Next, a bifactor model (SG3) was investigated. In this model, seven-factor Model SG1. Or conversely, because the SG1 model is
each of the item-level indicators was specified as being caused by more parsimonious than the SG3 model, for it to be preferred, the
(i.e., had factor loadings on) both a general traditional masculinity chi-square difference should not be statistically significant, any
ideology latent factor and a specific latent factor corresponding to drop in CFI should be less than .01, and it should have the lower
one of the factors in the original seven-factor model. The general BIC value. However, when these two models were compared with
traditional masculinity ideology factor was specified as orthogonal the chi-square difference test, the significant chi-square statistic
to the seven specific factors, but those seven factors were allowed favored the SG3 model, ⌬␹2 ⫽ 162.284, ⌬df ⫽ 21, p ⬍ .0001. In
to freely intercorrelate among themselves. The chi-square addition, the SG3 model had the higher (preferred) value of CFI,
goodness-of-fit statistic for the bifactor Model SG3 was statisti- with a ⌬CFI ⫽ .017. Finally, the smaller value of BIC for the
cally significant, ␹2(147) ⫽ 478.725, p ⬍ .0001, indicating that the bifactor model also supported preferring it. Thus, the bifactor SG3
null hypothesis of perfect fit should be rejected. But all remaining model fit better on all three criteria than did the seven-factor SG1
fit indices were well within the guidelines described earlier and model.
had values superior to those for the seven-factor structure Model Thus, of the set of single-group models tested and summarized
SG1, suggesting that in most aspects, the bifactor model fit the in Table 1, the bifactor model provided the superior fit to the data.
data very well, CFI ⫽ .967, RMSEA ⫽ .047, 90% CI [.042, .052], All factor loadings of this model were statistically significant for
SRMR ⫽ .030. the (a) General Traditional Masculinity Ideology factor, 21 items
The seven-factor Model SG1 is nested within the bifactor Model with loadings ranging from .42 to .83, and (b) seven specific
SG3. Thus, support for the bifactor model would consist of show- factors of three items each corresponding to each of the subscales
ing that it significantly improves fit over the more parsimonious found in Levant et al. (2010), with loadings ranging from .17 to .85
234 LEVANT, HALL, AND RANKIN

(all but two of the loadings were ⬎ .40). The factor intercorrela- factor loadings from this model in the male and female groups
tions among the seven specific factors ranged from .02 to .49 (all revealed an interesting pattern. Namely, in the male group, all
correlations of specific factors with the general factor were fixed to factor loadings were statistically significant (p ⬍ .05), but in the
zero as is typical for this type of model); none of these correlations female group, although all 21 loadings on the seven factors rep-
was so high as to suggest redundancy in the factors. These results resenting specific traditional masculinity ideology domains were
are reported in greater detail in Table 2. statistically significant, 13 of the 21 standardized loadings on the
general factor representing traditional masculinity ideology overall
Assessment of Measurement Invariance of the were not statistically significant. This pattern of results is interest-
MRNI-SF Factor Loadings Across Gender ing from a conceptual perspective, which we return to below. In
sum, however, we conclude that configural invariance was clearly
The second goal of this study was to use multigroup CFA of the
demonstrated in this data set for the seven specific factors and
MRNI-SF responses to assess the hypothesized configural and
provisionally for the general traditional masculinity factor (assum-
metric invariance across gender. Then, the preferred metric invari-
ing the potential for significant factor loadings for women in future
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

ance model was used to determine whether the latent factor means
models) and note that establishing configural invariance is a pre-
This document is copyrighted by the American Psychological Association or one of its allied publishers.

were equal for men and women (we had expected that they would
condition for continuing on to tests of stronger factorial invariance
not be). Other forms of invariance were then investigated in an
(Horn & McArdle, 1992; Vandenberg & Lance, 2000).
exploratory fashion. To meet these objectives, a series of nested
Model MG2A constrained parallel factor loadings (i.e., for the
models was estimated, treating the male and female respondents as
separate subsamples in simultaneous estimations. All of these same item on the same factor) to be equal for men and women, and
models used as their fundamental structure the bifactor model, thus allowed for a test of full metric invariance. Except for the
which was found to be best fitting in the previous single-group statistically significant chi-square statistic, overall the fit of Model
analyses. The chi-square change statistic provided a test of statis- MG2A was acceptable, ␹2(341) ⫽ 749.862, p ⬍ .0001, CFI ⫽
tical significance for model comparisons, and alternative indices .955, RMSEA ⫽ .049, 90% CI [.044, .053], SRMR ⫽ .044.
were also consulted. See Table 3 for the relevant multigroup Furthermore, given that all factor loadings in this model were
results. equality constrained across the two groups, Model MG2A allowed
Hypothesized forms of invariance: Configural and metric, a comparison of each of the factor means by gender, thus providing
with accompanying tests of invariance of factor means. a test of the invariance of latent factor means. The means for the
Model MG1 imposed no equality constraints across gender groups traditional masculinity ideology general factor, and for four of the
on any model parameters, and thus allowed for a very basic test of seven specific factors (all except Negativity toward Sexual Minor-
configural invariance (i.e., that the factor dimensionality and as- ities, Avoidance of Femininity, and Toughness), were significantly
signment of items to factors was correct). As would be necessary different for men versus women, with p values ranging from .050
to conclude that configural invariance holds, this least parsimoni- to ⬍ .001. For all factor means except Self-Reliance through
ous model in the series showed a good fit to the data, ␹2(307) ⫽ Mechanical Skills, the difference had a positive sign, meaning that
706.840, p ⬍ .0001, CFI ⫽ .956, RMSEA ⫽ .051, 90% CI [.046, men had higher values of these latent means than did women. The
.056], SRMR ⫽ .037. However, an inspection of the estimated two largest latent factor mean differences (unstandardized) were

Table 3
Model Fit Statistics and Comparisons of Nested Multiple-Group Models

Invariance model ␹2(df) CFI RMSEA estimate and 90% CI SRMR BIC

MG1: Configural 706.840 (307) .956 .051 [.046, .056] .037 69969.452
MG2A: Full metric 749.862 (341) .955 .049 [.044, .053] .044 69837.166
MG2B: Partial metric 718.399 (321) .956 .049 [.045, .054] .040 69928.411
MG3: Scalar (free intercepts)a 710.978 (308) .955 .051 [.046, .056] .039 69988.969
MG4: Uniquenesses 819.736 (342) .947 .052 [.048, .057] .045 69940.240
MG5: Variancesa 769.989 (328) .951 .051 [.047, .056] .041 69949.940
MG6: Variances ⫹ Covariances 807.160 (349) .949 .051 [.046, .055] .052 69865.430

⌬␹2(df) p ⌬CFI Conclusion


Model comparison:
MG2A vs. MG1 56.402 (34) .009 .001 Equivocal
MG2B vs. MG2A 33.913 (20) .027 .001 Equivocal
MG2B vs. MG1 22.807 (14) .063 .000 Prefer MG2B
MG2B vs. MG3 16.186 (13) .239 .001 Equivocal
MG4 vs. MG2B 79.150 (21) ⬍.001 .008 Equivocal
MG5 vs. MG2B 37.500 (7) ⬍.001 .005 Equivocal
MG6 vs. MG2B 81.501 (28) ⬍.001 .015 Equivocal
Note. In the model comparison section, the more parsimonious model is listed first. The conclusion is based on a joint consideration of ⌬␹2, ⌬CFI, and
a comparison of BIC values. CFI ⫽ comparative fit index; RMSEA ⫽ root-mean-square error of approximation; SRMR ⫽ standardized root-mean-square
residual; BIC ⫽ Bayesian information criterion.
a
A nonpositive definite matrix was encountered in the estimation of this model.
MALE ROLE NORMS INVENTORY–SHORT FORM 235

for Restrictive Emotionality (.92, p ⬍ .001) and Dominance (1.01, traditional masculinity ideology concept appears to exist for
p ⬍ .001). women, but it may not be exactly the same construct for them as
When the more parsimonious metric invariance Model MG2A for men.
was compared with the configural invariance Model MG1 to see Exploratory forms of invariance: Scalar, uniquenesses, vari-
whether imposing equality constraints on all of the factor loadings ances, and covariances. At least partial metric invariance is a
significantly degraded model fit, our criteria suggested conflicting precondition for testing additional forms of invariance (Vanden-
conclusions. The chi-square difference test was statistically signif- berg & Lance, 2000). Thus, given that partial metric invariance
icant, ⌬␹2(34) ⫽ 56.402, p ⫽ .009. By this criterion, Model was established with our data, we estimated exploratory models
MG2A, specifying complete metric (factor loading) invariance, fit specifying scalar invariance, invariance of uniquenesses, and in-
significantly worse than did configural invariance Model MG1, variance of factor variances and covariances. Partial metric invari-
suggesting that full metric invariance was not obtained. However, ance Model MG2B served as the comparison model for signifi-
two other criteria—a ⌬CFI of .001 (less than the rule-of-thumb cance testing for all of these remaining forms of invariance.
criterion of .010) and a lower value of the BIC for Model MG2A— Because we did not have specific expectations for these forms of
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

suggest support for the more parsimonious Model MG2A over invariance— except that full invariance on these stricter forms was
This document is copyrighted by the American Psychological Association or one of its allied publishers.

Model MG1. These latter two criteria may be less adversely unlikely—we report those results in the next several paragraphs
influenced by large sample sizes (such as our N of 1,017) than is and in Table 3, but do not pursue them in great detail.
the chi-square difference test. Thus, the results of the test of this Model MG3 was estimated to investigate whether the measure
model for metric invariance were equivocal, and interpretation showed scalar invariance (i.e., invariance of the item intercepts)
depends on which comparative fit criterion is given the most across the gender groups. This model was identical to partial
consideration. metric invariance Model MG2B (which by default had equivalent
The pattern of conflicting results like that seen when comparing item intercepts across groups), except that it freed the item inter-
Model MG2A with Model MG1 could be a result of partial metric cepts to vary between the two groups and fixed the factor means to
invariance—that is, a situation in which many factor loadings are zero in order to achieve identification. Model MG3 was less
indeed equivalent across gender groups, but some are not. Thus, parsimonious than Model MG2B because of the freed intercepts,
we explored the potential for partial invariance in one further and thus was expected to fit better. However, if the test criteria for
metric invariance model, Model MG2B. As mentioned earlier, comparing the fits of these two models showed no substantial
theoretical arguments suggested to us that men and women might difference between the two models, we would conclude that scalar
not globally conceive of traditional masculinity ideology in the invariance held.
same manner, and we observed a pattern of nonsignificant factor Model MG3 fit reasonably well overall, ␹2(308) ⫽ 710.978, p ⬍
loadings for women on the general traditional masculinity ideol- .0001, CFI ⫽ .955, RMSEA ⫽ .051, 90% CI [.046, .056],
ogy factor in Model MG1. Taken together, these considerations SRMR ⫽ .039. When the fit of Model MG3 was compared with
suggested the potential for a modified model specifying partial Model MG2B, the fit indices provided equivocal information with
metric invariance across gender. respect to whether there was full scalar invariance. The chi-square
Model MG2B constrained parallel factor loadings to be equal difference test was not statistically significant, ⌬␹2(13) ⫽ 16.186,
for men and women for the seven specific factors, but imposed no p ⫽ .239, and ⌬CFI ⫽ .001, suggesting that scalar invariance held.
equality constraints on the general traditional masculinity ideology However, the BIC value was higher for the less parsimonious
factor. Overall, this model fit adequately, ␹2(321) ⫽ 718.399, Model MG3, which freed the item intercepts, suggesting the pos-
p ⬍ .0001, CFI ⫽ .956, RMSEA ⫽ .049, 90% CI [.045, .054], sibility that some intercepts were not invariant.
SRMR ⫽ .040. It also significantly improved model fit over the Model MG4 was estimated to investigate invariance of the
full metric invariance Model MG2A by the change in chi-square residuals of the item-level indicators, also known as uniquenesses.
criterion, ⌬␹2(20) ⫽ 33.913, p ⫽ .027. However, the alternative Although the overall fit of this model was not excellent, it was
criteria of ⌬CFI (which equaled .001) and BIC favored Model borderline acceptable, ␹2(342) ⫽ 819.736, p ⬍ .0001, CFI ⫽ .947,
MG2A. Importantly, though, in contrast to the full metric invari- RMSEA ⫽ .052, 90% CI [.048, .057], SRMR ⫽ .045. When
ance Model MG2A, partial metric invariance Model MG2B did compared with Model MG2B, the chi-square difference test was
not fit significantly worse than did the configural invariance Model statistically significant, ⌬␹2(21) ⫽ 79.150, p ⬍ .001, and the BIC
MG1 according to the chi-square change statistic, ⌬␹2(14) ⫽ value was higher for Model MG4, suggesting that invariance of
22.807, p ⫽ .063. Model MG2B also fit as well as Model MG1 on uniquenesses does not hold. However, the ⌬CFI ⫽ .008, suggest-
the basis of the ⌬CFI ⫽ .000, and was the preferred model of the ing that the fit of Model MG4 does not differ substantially from
two on the basis of its lower BIC. that of Model MG2B, raising the potential for partial uniqueness
In sum, the results support configural invariance as well as the invariance.
metric equivalence of factor loadings in the two gender groups for We also investigated the equivalence across genders of the
the seven specific factors. However, the results also suggest that factor variances and factor covariances for the seven specific
the general factor of traditional masculinity ideology may not factors only (these analyses should be performed only if metric
show metric invariance across gender. Interestingly, when factor invariance is demonstrated, so should not be done for the general
loadings for the seven specific factors are invariant but the load- factor of traditional masculinity ideology). The rows for Models
ings for the traditional masculinity ideology general factor are MG5 (variances) and MG6 (variances ⫹ covariances) in Table 3
freely estimated in the two groups, the loadings on the general report the results from tests of the models specifying invariance
factor, although not equal for men and women, become statisti- across these parameters. On the basis of the chi-square difference
cally significant for both genders—thus indicating that a general test, both models fit significantly more poorly than the relevant
236 LEVANT, HALL, AND RANKIN

comparison model MG2B, indicating that at least some variances MRNI-SF total and subscale scores except for the Self-reliance
differ by gender, ⌬␹2(7) ⫽ 37.500, p ⬍ .001, and that restrictions through Mechanical Skills subscale. Where significant differences
of both variances and covariances to be equivalent are not tenable, occurred, men endorsed traditional masculinity to a greater extent
⌬␹2(28) ⫽ 81.501, p ⬍ .001. However, again making these con- than did women.
clusions equivocal, we note that (a) the ⌬CFI is only .005 (not
meeting the criterion of .01) for the restricted variance Model Discussion
MG5, suggesting the potential that the variances are “close
enough” in value across genders, and (b) the BIC value for the Our results build on previous research on the MRNI (summa-
restricted variance ⫹ covariance Model MG6 is smaller than for rized in Levant & Richmond, 2007) and the MRNI-R (Levant et
the comparison MG2B model. Taken together, these results sug- al., 2010, 2007) to develop a short form of the instrument, namely,
gest that although invariance is not perfect across the set of factor the 21-item MRNI-SF. The MRNI-SF overcomes limitations of
variances and covariances, the values of these parameters are some of the most commonly used measures of masculinity ideol-
likely still relatively similar across genders. ogy, such as the Male Role Norms Scale (Thompson & Pleck,
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

1986) and the MRNI, by assessing multiple dimensions of mas-


This document is copyrighted by the American Psychological Association or one of its allied publishers.

Descriptive Statistics for Men and Women on culinity ideology that are confirmed by factor analysis and that
have evidence for reliability and validity. The MRNI-SF thus
MRNI-SF
provides a brief measure appropriate for assessing traditional mas-
Given the support for the new bifactor structure of the MRNI-SF culinity ideology—a construct central to the Gender Role Strain
in the total sample, and for partial metric invariance across gender Paradigm.
groups, we provide raw score-based subscale and total scale scores
for the MRNI-SF, for purposes of comparison with future studies Implications of Empirical Results for Quality of
in which this instrument is used. These scores were created by
Measurement and Conceptual Understanding
averaging the observed values for the relevant items. Means,
standard deviations, and alpha coefficients for these subscales are Counseling Psychologists have taken the lead in research on the
in Table 4, reported separately for men and women, along with the linkages between masculinity and mental health. Researchers who
results of significance tests comparing mean scores across men and are interested in a brief measure of traditional masculinity ideology
women. (Note that these mean comparisons are not identical to that nevertheless allows for the assessment of specific dimensions
comparisons made with the latent factor means.) of that ideology should consider the MRNI-SF. The analyses that
The alpha coefficients in Table 4 were evaluated using the we report provide empirical evidence for the hypothesized seven-
criteria developed by Ponterotto and Ruckdeschel (2007, p. 1003). factor dimensionality of the specific MRNI domains, and they
For the MRNI-SF subscales and total scale, those for men were strongly support the intended assignment of items to the specific
good (.80 –.84) to excellent (.85 and up), with the exception of factors. The findings of partial metric invariance, and exploratory
Toughness, which was moderate (.75–.79) at .79. Women’s alphas results suggesting that there may be partial invariance for other
were also good to excellent, with the exceptions of the moderate parameters (i.e., intercepts, uniquenesses, variances and covari-
alphas observed for the subscales of Toughness (.75), Importance ances), suggest that comparisons between genders on this instru-
of Sex (.76), and Restrictive Emotionality (.76). ment are tenable. This is true particularly if researchers are work-
On the MRNI-SF, both men and women on average scored ing in a latent variable context such as that involved in testing
toward the traditional end on two subscales, Self-reliance through SEM models, an option that many researchers on this topic may
Mechanical Skills and Toughness. In addition, men on average consider critical.
scored toward the traditional end on Avoidance of Femininity. In addition, as a direct result of this study, our understanding of
Significant mean differences by gender were found on all the construct of traditional masculinity ideology was expanded by

Table 4
MRNI-SF Raw Score Means, Standard Deviations, and Coefficient Alphas by Gender

Men (n ⫽ 549) Women (n ⫽ 468)


Measure M SD 〈 M SD ␣ F(1,1015)

Restrictive Emotionality 2.92 1.22 .83 1.87 0.78 .76 257.84


Self-Reliance through Mechanical Skills 4.52 1.36 .86 4.41 1.41 .84 1.64, ns
Negativity toward Sexual Minorities 3.18 1.57 .88 2.29 1.34 .84 90.98
Avoidance of Femininity 4.24 1.52 .90 3.43 1.50 .88 72.62
Importance of Sex 3.76 1.51 .83 2.85 1.25 .76 106.88
Toughness 4.68 1.23 .79 4.08 1.23 .75 60.51
Dominance 3.05 1.39 .87 1.78 0.94 .80 282.04
MRNI-SF Total Score 3.74 1.05 .92 2.91 0.86 .94 185.68
Note. MRNI-SF scale scores may range from 1 to 7. Higher values indicate greater endorsement of traditional masculinity ideology. All gender
comparisons were statistically significant at p ⬍ .01 or better, except for a nonsignificant finding for the Self-Reliance through Mechanical Skills subscale.
MRNI-SF ⫽ Male Role Norms Inventory–Short Form.
MALE ROLE NORMS INVENTORY–SHORT FORM 237

CFA results supporting a bifactor model. The bifactor model is sample characteristics (i.e., a relatively younger college student
interesting because it implies that item-level responses on the sample).
MRNI-SF reflect both a specific norm of traditional masculinity
ideology and a general factor reflecting an overarching traditional Implications for Practice With Consideration of
masculinity ideology concept. Interestingly, the pattern of factor Boundary Conditions and Limitations
loadings on the seven specific factors was invariant across gender
but varied somewhat across gender for the general factor, thus There are also implications of the measurement development
supporting the partial metric invariance of this measure. undertaken in this study for practicing Counseling Psychologists.
The mix of invariant and differential factor loadings for specific Given the association of its predecessor, the MRNI, with a range
versus general factors observed in this study was not unexpected. of problematic individual and relational variables (Levant & Rich-
Men and women are likely to have very similar concepts regarding mond, 2007), the MRNI-SF may be of use to counselors who wish
specific norms for the male role, as reflected in metric invariance to assess their clients’ endorsement of traditional masculinity
on the specific latent factors, because those norms are communi- ideology to see whether what may be largely implicit assumptions
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

cated very broadly within our culture to persons of both genders at are in fact a source of conflict or stress. There may be cases in
This document is copyrighted by the American Psychological Association or one of its allied publishers.

all ages. However, the lack of metric invariance for the general which discussing a client’s overall scores and/or responses to
traditional masculinity latent factor could indicate that men’s sense specific items may be of value in the counseling process, espe-
of self or identity may be engaged in a manner somewhat different cially where the treatment plan incorporates a client’s “Gender
than is experienced by women when responding to items asking to Role Journey” (O’Neil, 1996).
what extent they agree or disagree with statements about how men However, we wish to acknowledge some limitations of the
should think, feel, and behave. Future research should investigate current study that, at present, may also place some boundaries
this possibility by examining the associations of both the general around the generalizations of results. Participants in the study were
and specific factors with measures of masculine identity, such as all students from the same university, which draws primarily from
Wade and Gelso’s (1998) Male Reference Group Identity Scale. the surrounding midwestern geographic region. Most participants
Not surprisingly, we observed significant differences both on were young, White, European American, middle class, and het-
the latent factor means and on the raw score-based subscale means erosexual. Future investigations should aim to recruit a more
for the general factor and most of the seven specific factors. diverse sample in terms of demographic and geographic charac-
Interestingly, the tests of the latent factor means found significant teristics. Additionally, the self-report nature of the surveys intro-
latent factor mean differences on all except the Negativity toward duces the possibility of socially desirable responding and does not
Sexual Minorities, Avoidance of Femininity, and Toughness fac- allow for a multimethod evaluation of the data (e.g., including
tors, whereas the subscale means indicated significant differences interviewing), which is important in assessing construct validity.
on all subscales except the Self-Reliance through Mechanical We encourage future investigations with the MRNI-SF that con-
Skills score. Regardless, when significant differences were ob- tinue to address these issues.
served, they were almost all in the direction of higher values for
men, with the exception of the Self-Reliance through Mechanical References
Skills factor, which was endorsed more strongly by women. Both Cheung, G. W., & Rensvold, R. B. (2001). The effects of model parsimony
the presence of these significant differences and their direction and sampling error on the fit of structural equation models. Organiza-
help support the construct validity of the MRNI-SF. Future re- tional Research Methods, 4, 236 –264.
search should focus on further assessing the convergent, discrim- Horn, J. L., & McArdle, J. J. (1992). A practical and theoretical guide to
inant, and concurrent construct validity of the MRNI-SF, both for measurement invariance in aging research. Experimental Aging Re-
the general traditional masculinity ideology factor and the seven search, 18, 117–144. doi:10.1080/03610739208253916
specific factors. Kahn, J. H. (2006). Factor analysis in counseling psychology research,
In addition, one empirical finding suggests potential particular training, and practice: Principles, advances, and applications. The Coun-
relevance of the specific factors of Restrictive Emotionality and seling Psychologist, 34, 684 –718. doi:10.1177/0011000006286347
Kline, R. B. (2011). Principles and practice of structural equation mod-
Dominance for an understanding of traditional masculinity ideol-
eling (3rd ed.). New York, NY: Guilford Press.
ogy. The six items comprising these two factors were observed to Levant, R. F. (1996). The crisis of connection between men and women.
show the largest gender differences in factor means. Specifically, Journal of Men’s Studies, 5, 1–12.
men tended to endorse these items noticeably more strongly than Levant, R. F. (1997). The masculinity crisis. Journal of Men’s Studies, 5,
did women. This is similar to findings from a study on an adoles- 221–231.
cent version of the MRNI (Levant et al., 2012), which revealed that Levant, R. F. (2011). Research in the psychology of men and masculinity
there is a connection between Restrictive Emotionality and Dom- using the gender role strain paradigm as a framework. American Psy-
inance for boys in that being constricted serves the purpose of chologist, 66, 765–776. doi:10.1037/a0025034
being dominant; that is, it is hard to dominate someone if you are Levant, R. F., Good, G. E., Cook, S., O’Neil, J., Smalley, K. B., Owen,
emotionally vulnerable to them. Finally, we mention that the K. A., & Richmond, K. (2006). Validation of the Normative Male
Alexithymia Scale: Measurement of a gender-linked syndrome. Psychol-
exploratory investigations of invariance in the tests of item unique-
ogy of Men & Masculinity, 7, 212–224. doi:10.1037/1524-9220.7.4.212
nesses, factor variances, and factor covariances were not focal for Levant, R. F., Hirsch, L., Celentano, E., Cozza, T., Hill, S., MacEachern,
this particular study but may be of interest to future researchers as M., . . . Schnedeker, J. (1992). The male role: An investigation of norms
the MRNI-SF continues to be used and studied. It may be that the and stereotypes. Journal of Mental Health Counseling, 14, 325–337.
lack of clear-cut conclusions from these tests are characteristic of Levant, R. F., Rankin, T. J., Williams, C., Hasan, N. T., & Smalley, K. B.
the measure, but it may also be that they are a reflection of the (2010). Evaluation of the factor structure and construct validity of the
238 LEVANT, HALL, AND RANKIN

Male Role Norms Inventory-Revised (MRNI-R). Psychology of Men & Ponterotto, J. G., & Ruckdeschel, D. E. (2007). An overview of coefficient
Masculinity, 11, 25–37. doi:10.1037/a0017637 alpha and a reliability matrix for estimating adequacy of internal con-
Levant, R. F., & Richmond, K. (2007). A review of research on masculinity sistency coefficients with psychological research measures. Perceptual
ideologies using the Male Role Norms Inventory. Journal of Men’s and Motor Skills, 105, 997–1014.
Studies, 15, 130 –146. doi:10.3149/jms.1502.130 Satorra, A., & Bentler, P. M. (1999). A scaled difference chi-square test
Levant, R. F., Rogers, B. K., Cruickshank, B., Rankin, T. J., Kurtz, B. A., statistic for moment structure analysis. Psychometrika, 66, 507–514.
Rummell, C. M., . . . Colbow, A. (2012). Exploratory factor analysis and Schlomer, G. L., Bauman, S., & Card, N. A. (2010). Best practices for
construct validity of the Male Role Norms Inventory-Adolescent-revised missing data management in counseling psychology. Journal of Coun-
(MRNI-A-r). Psychology of Men & Masculinity, 13, 354 –366. doi: seling Psychology, 57, 1–10. doi:10.1037/a0018082
10.1037/a0029102
Spence, J. T., & Helmreich, R. L. (1978). Masculinity and femininity: Their
Levant, R. F., Smalley, K. B., Aupont, M., House, A., Richmond, K., &
psychological dimensions, correlates, and antecedents. Austin: Univer-
Noronha, D. (2007). Initial validation of the Male Role Norms
sity of Texas Press.
Inventory–Revised (MRNI-R). Journal of Men’s Studies, 15, 83–100.
Thompson, E. H., & Pleck, J. H. (1986). The structure of male norms.
doi:10.3149/jms.1501.83
American Behavioral Scientist, 29, 531–543. doi:10.1177/
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

Mahalik, J. R., Locke, B. D., Ludlow, L. H., Diemer, M. A., Scott, R. P.,
Gottfried, M., & Frietas, G. (2003). Development of the Conformity to 000276486029005003
This document is copyrighted by the American Psychological Association or one of its allied publishers.

Masculine Norms Inventory. Psychology of Men & Masculinity, 4, 3–25. United Nations Development Programme. (2007– 08). Human development
doi:10.1037/1524-9220.4.1.3 report, 2011. Retrieved from http://hdr.undp.org/en/media/
Meade, A. W., & Lautenschlager, G. J. (2004). A comparison of item HDR_20072008_GEM.pdf
response theory and confirmatory factor analytic methodologies for Vandenberg, R. J., & Lance, C. E. (2000). A review and synthesis of the
establishing measurement equivalence/invariance. Organizational Re- measurement invariance literature: Suggestions, practices, and recom-
search Methods, 7, 361–388. doi:10.1177/1094428104268027 mendations for organizational research. Organizational Research Meth-
Meredith, W. (1993). Measurement invariance, factor analysis, and facto- ods, 3, 4 –70. doi:10.1177/109442810031002
rial invariance. Psychometrika, 58, 525–543. doi:10.1007/BF02294825 Wade, J. C., & Gelso, C. J. (1998). Reference Group Identity Dependence
Muthén, L. K., & Muthén, B. O. (1998 –2008). Mplus user’s guide (5th Scale: A measure of male identity. The Counseling Psychologist, 26,
ed.). Los Angeles, CA: Author. 384 – 412. doi:10.1177/0011000098263002
O’Neil, J. M. (1996). The Gender Role Journey Workshop: Exploring Wester, S. R., Vogel, D. L., O’Neil, J. M., & Danforth, L. (2012).
sexism and gender role conflict in a coeducational setting. In M. P. Development and evaluation of the Gender Role Conflict Scale Short
Andronico (Ed.), Men in groups: Insights, interventions, psychoeduca- Form (GRCS-SF). Psychology of Men & Masculinity, 13, 199 –210.
tional work (pp. 193–213). Washington, DC: American Psychological doi:10.1037/a0025550
Association.
Whorley, M. R., & Addis, M. E. (2006). Ten years of psychological
O’Neil, J. M., Helms, B., Gable, R., David, L., & Wrightsman, L. (1986).
research on men and masculinity in the United States: Dominant meth-
Gender Role Conflict Scale: College men’s fear of femininity. Sex Roles,
odological trends. Sex Roles, 55, 649 – 658. doi:10.1007/s11199-006-
14, 335–350. doi:10.1007/BF00287583
9120-1
Parent, M. C., & Moradi, B. (2009). Confirmatory factor analysis of the
Conformity to Masculine Norms Inventory and development of the
Conformity to Masculine Norms Inventory-46. Psychology of Men &
Masculinity, 10, 175–189. doi:10.1037/a0015481
Pleck, J. H., Sonenstein, F. L., & Ku, L. C. (1994). Attitudes toward male Received June 8, 2012
roles: A discriminant validity analysis. Sex Roles, 30, 481–501. doi: Revision received November 29, 2012
10.1007/BF01420798 Accepted November 30, 2012 䡲

Das könnte Ihnen auch gefallen