ABSTRACT
There exists a scale, developed in Mexico, aimed to measure attitude towards sexuality. Good metric properties have been reported for this scale, but there have been no studies aimed at corroborating those observations. This study aims to verify the validity, as well as the invariance between two samples, of the threefactor model originally proposed for the scale, to estimate its internal consistency reliability, to describe its distribution, and to verify its concurrent validity in relation to the attitude towards homosexuality. Two incidental samples of 402 psychology students and 198 medical students were recruited. The Attitude towards Sexuality Scale (ASS20) was applied to both samples. Likewise, the attitude towards homosexuality scale was also applied to the psychology students, whereas the factor of subtle rejection towards gay men, taken from the Mexican adaptation of Herek's Attitude towards Lesbians and Gay Men Scale, was applied to medical students. Singlefactor, correlatedfactor, hierarchical, and bifactor models were tested. Multigroup confirmatory factor analysis revealed that the correlatedfactor model had the best data fit as well as convergent and discriminant validity properties in both samples, although it was not invariant. The internal consistency of the scale was good. The total scores in ASS20 followed a normal distribution and their average showed a liberal attitude with no difference between both samples. The correlation of ASS20 with the two scales of attitude towards homosexuality was medium. It is concluded that AAS20 shows internal consistency reliability, structural validity, and concurrent validity in relation to attitude towards homosexuality.
Key words: Attitude, sexuality, homosexuality, psychometrics, students.
Abbreviation:
ASS20, Twentyitem Attitude towards Sexuality Scale; AHS10, Tenitem Attitude towards Homosexuality Scale; ATGS, Subtle rejection Attitude Towards Gay men subscale from the Mexican adaptation of Attitudes Towards Lesbians and Gay men (ATLG) scale; MAS, ASS20 rejection of Masturbation and Sex factor; SSM, AAS20 sexual shyness, shame, and modesty factor; VCP, AAS20 appraisement of Virginity and Condemnation of Pornography factor.
The attitude towards sexuality
Sexuality is an aspect of human behavior towards which society traditionally exercises strong control due to its implications for human bonding, progeny, inheritance, and motivation. Although some societies are more liberal and others more repressive of human sexuality, none of them stop exercising control over sexuality or stop dictating the canon of morality, normality, and abnormality regarding sexual behavior (Clark, 2019).
In Latin America, repressive discourses of sexuality that consider pornography and masturbation as something "sinful", homosexuality as something "against the law of God", and promiscuity and infidelity in women as something "immoral and deserving disgrace” have prevailed from the Colony to our times; it is only in the most cosmopolitan cities these discourses have lost strength although have not disappeared (Sanabria, 2019). On the other hand, social networks in which adolescents are constantly immersed (such as Twitter, Instagram, and YouTube) do not usually echo these repressive values. Likewise, civic education and secular sex education in public schools also do not support traditional moralizing discourses. On the contrary, homophobia is fought and gender equality is defended. Nevertheless, some of those people who provide this education, as well as many families to which adolescents belong and the churches they attend, defend moralizing repressive discourses. Thus, we live in a society in change and with contradictions (Imhoff et al., 2020; Moral, 2010; PinosAbad et al., 2017).
The students of the degrees of psychology and medicine, among which the present validation study of a measurement instrument is carried out, are evolutionarily late adolescents or are entering their adulthood and are economically dependent on their parents. They are immersed in these contradictory discourses, but they are being formed to contribute to the scientific and secular discourse about human sexual behavior. What is the attitude of these young people? What effect does this attitude have on various areas of your present behavior and on their professional work? These are relevant questions and their answers require reliable and valid measuring instruments.
If the objective to be assessed is the attitude towards sexuality, we first need to define such attitude. Attitude towards sexuality can be defined as a predisposition to express opinions, feel and act in the face of sexual objects (pornography), situations (nudity), different people (homosexuals, transvestites, transsexuals), social norms or customs (fidelity, virginity), and sexual behaviors (masturbation, oral sex, anal sex) (Moral and Ortega, 2008).
Measurement of attitude towards sexuality
There are several scales used to measure attitude towards sexuality. Some examples are the Sexual Conservatism Scale by Burt (1980) and the Brief Sexual Attitude Scale by Hendrick et al. (2006) developed in the United States. In Spanish language countries, there are some scales, such as the scale created by LimaSerrano et al. (2013) in Spain and the scales developed by Honold (2005) and Moral and Ortega (2008) in Mexico.
The 20item Attitude towards Sexuality Scale (ASS20) by Moral and Ortega (2008) is a multifactorial scale with good properties developed in the context of Mexican psychology students. The items of the ASS20 were prepared from a qualitative study with a focus group. From 26 initially designed items, 20 were selected based on their properties of discriminability, internal consistency, and factorial configuration. In a sample of 395 students, the overall internal consistency was good (Cronbach’s α = 0.84), the scores on the scale followed a normal distribution (KolmogorovSminrov Z = 1.10, Lillefors’ p = 0.17) and its average (M = 2.37; 95% CI: 2.32, 2.42; range 15) reflected a liberal attitude with no difference between both sexes. The number of factors was determined by the Cattell’s criterion and was three. When extracting three factors by Principal Axis Factoring, 31.6% of the total variance was explained. After rotating the factor matrix by the Varimax method, a first factor with seven indicators was obtained and was named appraisement of Virginity and Condemnation of Pornography (VCP) due to its content (items 2, 4, 6, 8, 11, 15, and 19). Its internal consistency reliability was acceptable (α = 0.75) and its scores adjusted to normal distribution (KolmogorovSmirnov Z = 0.94, Lillefors’ p = 0.34). The second factor with six indicators (items 1, 7, 9, 13, 14, and 17) was named rejection of Masturbation and Sex (MAS), its internal consistency reliability was also acceptable (Cronbach’s α = 0.72), its distribution was positively skewed, and did not follow a normal distribution (KolmogorovSmirnov Z = 2.12, Lillefors’ p < 0.01). The third factor with seven indicators (items 3, 5, 10, 12, 16, 18, and 20) was named sexual Shyness, Shame and Modesty (SSM), had a questionable internal consistency reliability (Cronbach’s α = 0.67), its distribution was also skewed to the right, and did not follow a normal distribution (Z_{KS} = 1.70, p = 0.01). By confirmatory factor analysis, estimating the parameters through the Maximum Likelihood method, the model of three correlated factors presented adequate fit indices: the relative chisquare (χ^{2}/df = 2.61) was lower than 3, the Standardized Root Mean Square Residual (SRMR = 0.06) and the Root Mean Square Error of Approximation (RMSEA = 0.07) were lower than 0.08, and the Population Gamma Index (PGI = 0.93), Adjusted Population Gamma Index (APGI = 0.91), and Jöreskog’s Goodness of Fit Indices (GFI = 0.93 and AGFI = 0.91) were higher than 0.90. The correlations between the three factors were positive and with an association strength that ranged from medium (r = 0.45 for the correlations of both the first and the second factor with the third factor) to high (r = 0.54 for the correlation between the first two factors), and therefore there was discriminant validity between the three factors (Moral and Ortega, 2008).
Honold (2005)’s 25item Attitude towards Sexuality Scale (ASS25), in its original study carried out with a sample composed of 150 women and 150 men, all of them psychology students, showed an acceptable overall internal consistency reliability (Cronbach’s α = 0.72), and a structure of eight components that explained 57.4% of the total variance. The number of components to retain was determined using the Kaiser’s criterion (eigenvalues greater than 1). The distribution of the total scores on the scale showed positive asymmetry and reflected a very conservative attitude with no difference between both sexes (Honold, 2005). TrejoPérez and DíazLoving (2013) eliminated one item and obtained an acceptable overall internal consistency reliability (Cronbach’s α = 0.74). They determined the number of components by the Cattell’s criterion and it was four. These four factors explained 72.9% of the total variance. After an orthogonal rotation by the Varimax method, the first component reached an acceptable internal consistency reliability (Cronbach’s α = 0.75) and was named “sexism and stereotypes”. The second component showed a questionable internal consistency reliability (Cronbach’s α = 0.67) and was named “morality”. The third component had an unacceptable internal consistency reliability (Cronbach’s α = 0.49) and was named “restricted sexuality”. The fourth presented a fair internal consistency (Cronbach’s α = 0.54) and was named “myths and taboos”. In Mexico, MoralOrtega ASS20 compared Honold’s ASS25 had better psychometric properties.
Problem statement
The AAS20 has been applied in other studies (Moral and Ortega, 2009; Moral, 2010), but its threefactor structure and other metric properties have not been subsequently verified. In the present study, the three factors are specified not only through the correlatedfactor model, as in previous studies, but also through the hierarchical model and the bifactor model. In the correlatedfactor model, each item is directly determined by a factor and a measurement error, and the factors are linearly interrelated. In the hierarchical model, each item is determined indirectly by a general higherorder factor and directly by a hierarchical lowerorder factor and a measurement error. In the bifactor model, each item is directly determined by a general factor, a specific factor, and a measurement error. It should be noted that the use of a general score is theoretically justified in the last two models, whereas its justification is merely practical or utilitarian in the correlatedfactor model (DomínguezLara and Rodríguez, 2017). In the three aforementioned models, it is required to check the convergent validity of each factor and the discriminant validity between the factors (Gignac and Kretzschmar, 2017). There exists convergent validity in a factor if the proportion of covariance between the items explained by the factor is higher than that explained by chance, that is, by nonattributable factors. There exists discriminant validity between two factors if a substantive proportion of variance shared between the items is exclusively attributable to the factor to which they belong and not to the other factor (Henseler et al., 2015). If the variances attributable to the specific factors are trivial or the proportions of variance shared between the factors are unitary, then a single factor underlies (Gignac and Kretzschmar, 2017); consequently, the onefactor model must be tested as an alternative hypothesis to multifactor model.
The objectives of this study are to test the validity of the threefactor model proposed in the original study carried out by Moral and Ortega (2008) using two samples, one composed of psychology students (as in the original study) and another one composed of medical students; to estimate the internal consistency reliability (variance proportion of the test scores measured without error); to describe the distributions of the scores on the scale and its factors; and finally, to verify the convergent validity in relation to the attitude towards male homosexuality, using two different measurement instruments.
The bifactor model is expected to have the best fit and good properties of convergent and discriminant validity, especially compared to the onefactor model, although it may not be strictly invariant between both types of students. Likewise, good reliability is expected in the scale and acceptable in its three factors, normal distribution in the scale and the factor of appraisement of virginity and condemnation of pornography, as well as a distribution with positive skewness in the other two factors (Moral and Ortega, 2009). Finally, the expectation is that there is a positive correlation with a medium strength of association between the attitude of rejection towards sexuality and the attitude of rejection towards gay men (Moral and Valle, 2014), since the attitude of rejection towards sexuality is usually consistent with prejudices towards sexual minorities, whose sexuality is disqualified as unnatural and promiscuous (Kite et al., 2020).
Research design
The present nonexperimental study is instrumental in its objectives and its design is crosssectional.
Participants
The inclusion criteria were: being a student in the first semesters of the psychology and medicine career and age 18 to 26 years. The exclusion criterion was to express refusal to participate and the elimination criterion was to leave two o more questions unanswered in the two scales that were applied. Nonprobability convenience sampling was used. The printed questionnaire was applied in the classrooms by the article authors. All participants invited to the study gave their consent and no case was required to be eliminated due to incomplete data. A sample composed of 402 psychology students and 198 medical students was obtained.
Table 1 describes and statistically compares the sociodemographic and sexual behavior variables of both samples. There was a difference in the sex ratio. In the sample of psychology students there were three women per one man while in the sample of medical students there was one woman per one man. The effect size of career/university on sex ratio was small (Cramer’s V = 0.25). Psychology students (M = 19.59; 95% CI: 19.44, 19.74) were significantly younger (MD = 0.23; 95% CI: 0.45, 0.01) than medical students (M = 19.82; 95% CI: 19.66, 19.99), although the effect size of career/university on age was trivial (Hedges' g = 0.16; 95% CI: 0.01, 0.33). The mean age in both samples corresponded to the late adolescence stage. There were also differences in the distribution of marital status. All medical students were single whereas 2.2% of psychology students reported other marital status (married, cohabiting, and separated). The effect size of career/university on marital status was trivial (Cramer’s V = 0.09). In subjective socioeconomic status, there was a clear difference, and this was higher in medical students. This is because the medical students came from a private university and the psychology students from a public one. The size effect of the career/university on the subjective socioeconomic status was large (Rosenthal’s r = 0.61). The distribution of religious affiliation was homogeneous between both samples. Overall, 78% of students reported being Roman Catholics, 12.3% having no religion (agnostics or atheists), 6.7% being nonCatholic Christians and 3% belonging to a nonChristian religion. There was no difference in the median of attendance at religious services between the two samples and it corresponded to “infrequent”. The percentage of people with an active sexual life was higher among psychology students (57.5%) than among medical students (48%). This difference was maintained when controlling for age (MantelHaenszel’s χ^{2}[1] = 5.18, p = 0.023; fulfilling the assumption of homogeneity of the odds ratio among the five age groups by the BreslowDay test: χ^{2}[5] = 7.54, p = 0.184). However, the effect size of sample/university on being sexually active was trivial (Cramer’s V = 0.09). The frequency of homosexual behaviors was statistically not different between both samples. Overall, 95.3% reported not having had homosexual behaviors compared to 4.7% who did (Table 1).
Instruments of measurement
The questionnaire is made up of the informed consent form, the sociodemographic data sheet, and two attitude scales, one towards sexuality and the other towards homosexuality. Twentyitem Attitude towards Sexuality Scale (ASS20) (Moral and Ortega, 2008). Is composed of 20 items with a Likerttype response scale with five ordered categories of agreement level: SD = Strongly Disagree, D = Disagree, nAnD = neither Agree nor Disagree, A = Agree, and SA = Strongly Agree (Annex). Higher score reflects an attitude of greater rejection. The items 1, 3, 5, 6, 7, 9, 11, 13, 15, 17, and 19 are positivelykeyed items and are scored in this way: TD = 1, SD = 2, nAnD = 3, SA = 4, and TA = 5. The items 2, 4, 8, 10, 12, 14, 16, 18, and 20 are negativelykeyed and are scored in this way: TD = 5, BD = 4, nAnD = 3, BA = 2, and TA = 1. The ASS20 total score is obtained as an average score on the 20 items. First, the scores on the 20 items are summed and then this sum is divided by 20. The range of scores on the ASS20 varies on a continuum from 1 to 5. By dividing this continuum into five intervals with the same amplitude (amplitude = [maximum value in the item  minimum value in the item] / number of values in the item = [51]/5 = 0.8), the score on the scale can be interpreted in an absolute sense from the response label of the items. A score from 1 to 1.79 reflects an attitude of decided acceptance, from 1.80 to 2.59 an attitude of acceptance, from 2.6 to 3.39 an ambiguous or ambivalent attitude, from 3.4 to 4.19 an attitude of rejection, and from 4.20 to 5 an attitude of decided rejection. The ASS20 is made up of three factors: appraisement of virginity and condemnation of pornography (VCP) (items 2, 4, 6, 8, 11, 15, and 19), rejection of masturbation and sex (MAS) (items 1, 7, 9, 13, 14, and 17) and sexual shyness, shame and modesty (SSM) (items 3, 5, 10, 12, 16, 18, and 20). Its metric properties were presented in the Introduction section.
Tenitem Attitude towards Homosexuality Scale (AHS10) (Moral and Ortega, 2008): It comprised 10 items with a Likerttype response scale with five ordered categories. Higher score reflects higher rejection of male homosexuality. The items 1, 3, 5, 7, and 9 are positivelykeyed items and are scored from 1 to 5. The items 2, 4, 6, 8, and 10 are negativelykeyed items and are scored from 5 to 1. The AHS10 score is obtained as an average score on the 10 items. First, the scores on the 10 items are summed and then this sum is divided by 10. The range of AHS10 varies on a continuum from 1 to 5 and can be interpreted in the same way as the ASS20. The overall internal consistency reliability was good (Cronbach’s α from 0.84 to 0.87), its structure was one factor by Kaiser’s criterion, the singlefactor model showed an acceptable fit to the data by Maximum Likelihood, and the scores on AHS10 followed a normal distribution (Moral and Ortega, 2008; Moral and MartínezSulvarán, 2011). The AHS10 was administrated only among psychology students and its reliability in this sample was excellent (ordinal α = 0.90).
Subscale of Subtle rejection Attitude Towards Gay men (ATGS) (Moral and Valle, 2011) from the Attitudes Towards Lesbians and Gay men (ATLG) Scale (Herek, 1984): It comprised five items with a Likerttype response scale with five ordered categories that are scored from 1 to 9. It is derived from the threefactor structure reported by Moral and Valle (2011) for the ATLG scale. Higher scores show greater rejection. The contents included are: qualification of male homosexuality as a natural sexual orientation and sex between men as a natural behavior, marriage between two men, the adoption of children by homosexual couples, and homosexuality in a male child. It is composed of three positively keyed items (G1, G5, and G7), which are scored: 1 = totally agree, 3 = agree, 5 = neither agree nor disagree, 7 = disagree, and 9 = totally disagree, as well as two negativelykeyed items (G8 and G9), which are scored from 9 = totally agree to 1 = totally disagree. The ATGS score is obtained as an average score in the 5 items. First, the scores on the five items are summed and then this sum is divided by 5. The range of scores on the ATGS varies on a continuum from 1 to 9. By dividing this continuum into five intervals with the same amplitude (a = [91]/5 = 1.6), the score on the scale can be interpreted in an absolute sense from the response label of the items. A score from 1 to 2.59 reflects an attitude of decided acceptance, from 2.60 to 4.19 an attitude of acceptance, from 4.20 to 5.79 an ambiguous attitude, from 5.80 to 7.39 an attitude of rejection, and from 7.40 to 9 an attitude of decided rejection. Its overall internal consistency reliability was adequate (Cronbach’s α = 0.78). The distribution of scores on ATGS was mesokurtic (Z_{K} = 0.18), but showed slight positive asymmetry (Z_{Sk} = 4.77); therefore, it did not fit a model of normal distribution (Moral and Valle, 2014). The ATGS was administrated only among medical students and its reliability in this sample was good (ordinal α = 0.86).
Procedure
The permission and approval of the academic authorities of the two faculties in which the data were collected was obtained. The informed consent of the students was requested for their participation in the research. This appeared on the first page of the questionnaire. No personally identifiable information was requested to guarantee the anonymity of the responses. The name and email address of those responsible for the research were provided to request information in relation to any concern raised by this study. In this way, the ethical research standards of the American Psychological Association (2017) were fulfilled.
Data analysis
For the first objective of testing the validity and invariance of four hypothetical models across the two types of students, multigroup confirmatory factor analysis was used. Models nested in constraints (equality of parameters between the two samples) were defined. The discrepancy function was optimized through the Unweighted Least Squares method. Moment matrices (arithmetic mean, standard deviation, and polychoric correlation) were used as input data. The nested model with constraints on intercepts could not be tested, because we opted for this method, which is suitable for ordinal variables, such as items with a Likerttype scale. The standard deviation, 95% confidence interval, and the significance test for each parameter were calculated through BiasCorrected Percentiles with the extraction of 2,000 bootstrap samples. The parameters between the two samples were compared by the Z test, using the bootstrap standard error of each parameter. The fit of the models to the data was assessed using eight indices: χ^{2}/df = relative chisquare, GFI = GoodnessofFit Index, AGFI = Adjusted GoodnessofFit Index, NFI = Normed Fit Index, CFI = Comparative Fit Index, RFI = Relative Fit Index, SRMR = Standardized Root Mean Square Residual, and RMSEA = Root Mean Square Error of Approximation. It was stipulated that values for χ^{2}/df ≤ 2, GFI, NFI, CFI and RFI ≥ 0.95, AGFI ≥ 0.90, and SRMR and RMSEA ≤ 0.05 reflect a close fit. On the other hand, values for χ^{2}/df ≤ 3, GFI, NFI, CFI and RFI ≥ 0.90, AGFI ≥ 0.85, SRMR < 0.10 and RMSEA < 0.08 reflect an acceptable fit. The equivalence in goodness of fit between the nested models was evaluated by the quotient between the difference in chisquare statistics and the difference in their degrees of freedom (Δχ^{2}/Δdf ≤ 3 for acceptable value and ≤ 2 for a close value)’ the difference in the Akaike Information Criterion (ΔAIC < 7 for acceptable value and < 2 for a close value) and the difference in the GFI, NFI, CFI, RFI, RMSEA and SRMR statistics (ΔGFI, ΔNFI, ΔCFI, ΔRFI, ΔRMSEA, and ΔSRMR ≤ 0.01) (Byrne, 2016).
In the singlefactor and correlatedfactor models, the convergent validity of each factor was established through three criteria: omega coefficient or compound reliability (ω) ≥ 0.70, average of the standardized measurement weights M_{λ} > 0.50, and Mean Variance Extracted (AVE) ≥ 0.25 for 7 or 12 indicators and 0.28 for 6 indicators (Moral, 2019).
In the correlatedfactor model, the discriminant validity between factors was established through the heterotraitmonotrait ratio of correlations (HTMT). The HTMT is calculated as the ratio between the arithmetic mean of the n*m nonredundant correlations of the items crossed between two factors and the geometric mean of the means of the [n*(n1)]/2 or [m*(m1)]/2 nonredundant correlations between the n or m items of each factor. It was stipulated that an HTMT value ≤ 0.85 reflects discriminant validity or at least ≤ 0.90 (Henseler et al., 2015).
In the hierarchical model, two submodels are distinguished. On one hand, there is the higherorder submodel that corresponds to the direct effect of the general higherorder factor (GF) on the three hierarchical lowerorder factors (HF). On the other hand, there is the lowerorder submodel that corresponds to both the indirect effect of the general higherorder factor and the direct effect of the hierarchical lowerorder factor on each item of a content domain. In the higherorder submodel, only convergent validity is assessed and it was done as in the singlefactor model (ω ≥ 0.70, M_{λ} > 0.50 and AVE ≥ 0.44 for three indicators). In the lowerorder submodel, as in the bifactor model, both convergent and divergent validity are checked for items in each content domain. The discriminant validity or contribution of each (hierarchical and highorder or specific and general) factor was assessed thorough six indices: the Average Variance Explained by the hierarchical lowerorder factor (AVE_HF) and general higherorder factor (AVE_GF) or by the specific factor (AVE_SF) and general factor (AVE_GF), the Common Variance Explained by the hierarchical lowerorder factor (ECV_HF) and general higherorder factor (ECV_GF) or by the specific factor (ECV_SF) and general factor (ECV_GF), as well as McDonald’s hierarchical omega related to the hierarchical lowerorder factor (ωh_HF) and general higherorder factor (ωh_GF) or to the specific factor (ωh_SF) and general factor (ωh_GF). Values between 0.30 and 0.70 for ωh_ and ECV_ indices reflect a significant and balanced contribution; values below 0.30 indicate a poor contribution, and values above 0.70 an excessive contribution (Brunner et al., 2012; DomínguezLara and Rodríguez, 2017). Considering the minimum AVE_ values (≥ 0.44 for three indicators, ≥ 0.28 for six indicators and ≥ 0.25 for seven indicators or more) and a contribution of at least 30%, the minimum AVE values for the hierarchical lowerorder factor (AVE_HF), specific factor (AVE_SF) or higherorder or general factor (AVE_GF) should be 0.14, 0.09, and 0.08, respectively. The maximum value would be 0.70 for a maximum contribution of 70% when explaining 100% of the variance (Moral, 2019). Convergent validity was established from the total effect or sum of direct effect of lowerorder hierarchical factor and indirect effect of general higherorder factor (hierarchical model) or the sum of direct effects of specific and general factors (bifactor model). As previously, three criteria were used: AVE_ total ≥ 0.25 for seven or twelve indicators and 0.28 for six indicators, M_{λ_total} > 0.50 and ωh_total ≥ 0.70.
For the second objective, the internal consistency reliability of each factor was calculated through the ordinal alpha coefficient (ordinal α), which is obtained by calculating the standardized alpha coefficient from the polychoric correlation matrix. It was stipulated that ordinal α values between 0.70 and 0.79 reflect an acceptable internal consistency, between 0.80 and 0.89 good, and ≥ 0.90 excellent (Viladrich et al., 2017). Interval estimations with a 95% confidence level were calculated using the formula of Feldt et al. (1987) and coefficients between the two samples were compared through Feldt’s (1969) test.
For the third objective of describing the distributions, fitting the scores to a normal distribution was tested using the D’Agostino K^{2} test (D’Agostino et al., 1990). This test is based on the transformation to normality of the coefficients of skewness (D’ Agostino, 1970) and kurtosis (Anscombe and Glynn, 1983). This statistic (D’Agostino and Pearson, 1973) was also calculated using the FisherPearson standardized moment coefficients of skewness and kurtosis (Fisher, 1930). Following the suggestion made by D’Agostino et al. (1990), it was complemented with the normal quantilequantile plot, which was quantitatively assessed through the correlation between the theoretical and empirical quantiles. A 95% confidence interval for the correlation that includes 1 reflects a good fit to normality.
As the model with constraints on intercepts could not be tested, the means between the two samples were compared through Hotelling’s T^{2} test, assuming homogeneity of covariances between the two samples. This assumption was checked using Box’s M test. The assumption of multivariate normality was tested by the multivariate version of the JarqueBera test (Koizumi et al., 2009). Following the recommendations of the study on the power of multivariate normality statistics (Joenssen and Vogel, 2014), multivariate asymmetry was calculated by the KankainenTaskinenOja (2007) U statistic, which follows a chisquare distribution with both degrees of freedom as variables are included in its calculation (k). The standardized value of this statistic (Z_{U}) was also calculated, using the WilsonHilferty (1931) transformation of chisquared variables to normality. Multivariate kurtosis was calculated using the Mardia’s statistic b_{2}, which follows a standard normal distribution by subtracting its mathematical expectation or mean, μ_{b2} = k*(k + 2), and dividing this difference by its standard error: σ_{b2} = √[(8*k*(k + 2))/n], where n is the sample size (Mardia, 1970). On the other hand, the means among the factors within each sample were compared using repeated measures analysis of variance. Pairwise comparisons were made by the pairedsample ttest with the Bonferroni’s correction for significance level.
For the fourth objective of concurrent validity, the correlations were calculated by Pearson’s productmoment coefficient (r), their significance (H_{0}: ρ = 0) was tested by Fisher’s Z test, and Fisher’s transformation was also used to calculate 95% confidence intervals, following SPSS convention. Bivariate normality was verified by tests based on asymmetry (KankainenTaskinenOja Utest), kurtosis (Mardia’s Ztest), and multivariate version of the JarqueBera test, using two previous multivariate statistics. In all cases, there was a good approximation to bivariate normality. Values of â”‚râ”‚ < 0.10 were interpreted as a trivial association strength, between 0.10 and 0.29 weak, between 0.30 and 0.49 medium, between 0.50 and 0.69 strong, between 0.70 and 0.89 very strong, and ≥ 0.90 perfect. The same thresholds were also used to interpret the size of the effect of the factors on their indicators estimated through the standardized measurement weights (Byrne, 2016). The significance of the difference in Pearson’s correlation coefficients between the two samples were tested using Fisher’s Z test and within each sample using Steiger’s Z test. The significance level was set at 0.05 and the calculations were made with the programs SPSS 24, module R version 2.4 for SPSS 24, AMOS 16, and Real Statistics Resource Pack for Excel 2013.
Testing the fit and invariance of hypothetical models across psychology and medical students, convergent validity of each factor, and discriminant validity between factors
Table 2 shows the fit indices of the four hypothesized models for the ASS20 (1F = single factor, CF3 = three correlated factors, HM3 = hierarchical model with one general higherorder factor and three hierarchical lowerorder factors, and BM3 = bifactor model with one general factor and three specific factors). Since a multigroup analysis was performed, models nested in constrains were specified within each of the four hypothesized models. The four common nested models in the four hypothesized models were: the unconstrained model (UC), with constraints on measurement weights (MW), with constraints on the structural covariances (SC), and with constraints on measurement error variances (ME). The hierarchical model (HM3) had two additional nested models: on the structural weights (SW) and on the structural error variances (SE), which were associated with the higherorder submodel or direct effect of the general higherorder factor on the three hierarchical lowerorder factors. Table 2 shows the comparison between goodnessoffit indices between the factorial model with the best fit (BM3) and the other factorial models (1F, CF3, and HM3). Table 3 shows the convergent validity indices of the singlefactor model (1F) and three correlated factors (CF3), and the discriminant validity indices between the factors of this last model. Table 4 shows the convergent and discriminant validity indices of the hierarchical model (HM3) and bifactor model (BM3).
The onefactor model
The onefactor model (1F) showed that all its parameters were significant in the four nested models when being estimated in both student samples. In the nested with constraints on measurement error variances (ME), whose estimates are exactly the same for both samples, the single factor with 20 indicators showed convergent validity: M_{λ} = 0.53, with a minimum of 0.41 and a maximum of 0.76, AVE = 0.30 > 0.25 and ω = 0.99 > 0.70 (Table 3).
The fit of the unconstrained model (UC) was close through four indices (GFI, AGFI, CFI, and RMSEA) and acceptable through four (χ^{2}/df, NFI, RFI, and SRMR) (Table 2). In this model, there were significant differences in 8 out of the 20 (40%) measurement weights, the structural or factor variance, and 5 out of the 20 (25%) error variances between the two samples. The variance of the factor (Z = 2.97, p = 0.003) and the measurement weights of items 9 (Z = 3.04, p = 0.002), 14 (Z = 2.08, p = 0.038), and 19 (Z = 2.28, p = 0.023) were higher in medical students than in psychology students. In addition, the measurement weights of items 4 (Z = 2.26, p = 0.024), 6 (Z = 4.14, p < 0.001), 15 (Z = 3.14, p = 0.002), and 17 (Z = 3.67, p < 0.001) were higher and their error variances were lower in medical students than in psychology students. On the contrary, the measurement weight of item 13 (Z = 2.17, p = 0.030) was higher and its error variance was lower (Z = 4.86, p < 0.001) in psychology students than in medical students. The goodness of fit of the unconstrained model (UC) was higher when is compared to the other three nested models for singlefactor model (Δχ^{2}/df > 3, ΔAIC > 10, ΔGFI, ΔAGFI, ΔNFI, ΔCFI, and ΔRFI > 01). In the nested model with constraints on measurement error variances (ME), which presented the worst fit among the nested models for the onefactor model, the fit was acceptable through seven indices, although bad through one (RMSEA > 0.08) (Table 2). The goodness of fit of the singlefactor model (1F), compared to the other three hypothesized models (CF3, HM3, and BM3), was the lowest in each of the four nested models through the indices Δχ^{2}/df > 3, ΔAIC > 10, ΔGFI, ΔAGFI, ΔNFI, ΔCFI, ΔRFI, ΔRMSEA, and ΔSRMR > 01 (Table 2).
The three correlated factor model
The model of three correlated factors (CF3) showed that all its parameters were significant in the four nested models when they were estimated in the two samples of students. In the nested with constraints on measurement error variances (ME), whose estimated values are the same in the two samples, the three factors showed convergent validity (M_{λ} from 0.53 to 0.67 > 0.50, AVE from 0.29 to 0.45 > 0.28, ω from 0.95 to 0.99 > 0.70) and discriminant (HTMT from 0.61 to 0.86 < 0.90) (Table 3).
The fit of the unconstrained model (UC) was good through seven indices and acceptable through one (SRMR = 0.06) (Table 2). In this model, there were significant differences in 6 out of the 20 (30%) measurement weights, in one out of the three (33.3%) structural variances, one out of the three (33.3%) structural correlations, and 5 out of the 20 (25%) measurement error variances. The measurement weights of items 6 (Z = 4.13, p < 0.001), 15 (Z = 3.01, p = 0.003), and 17 (Z = 3.27, p = 0.001) were higher and their error variances were lower (Z = 2.81, p = 0.005 in item 6; Z = 3.07, p = 0.002 in item 15; and Z = 3.01, p = 0.003 in item 17) in medical students than in psychology students. In addition, the correlation between the first two factors (Z = 2.21, p = 0.027) and the measurement weights of items 9 (Z = 2.85, p = 0.004) and 19 (Z = 1.97, p = 0.049) were higher in medical students than in psychology students. On the contrary, the measurement weight of item 13 (Z = 2.66, p = 0.008) was higher and its error variance lower (Z = 5.15, p < 0.001) in psychology students than in medical students. The variance of the third factor of sexual shyness, shame and modesty was higher in psychology students than in medical students (Z = 2.32, p = 0.020).
The goodness of fit of the unconstrained model (UC) was greater compared to the other three nested models for correlatedfactor model (Δχ^{2}/df > 3, ΔAIC > 10, ΔGFI, ΔAGFI, ΔNFI, ΔCFI, ΔRFI, and ΔSRMR > 01). In the nested model with constraints on measurement errors (ME), which had the worst fit among the nested models for the correlatedfactor model, the fit was good through seven indices and acceptable through one (SRMR = 0.06) (Table 2 and Figure 1). The goodness of fit of the three correlated factor model (CF3) was equivalent to that of the hierarchical model (HM3), except in the nested model with constraints on structural covariances, which was better in the hierarchical model through the indices χ^{2}/df = 19.14 > 3 and ΔAIC = 51.43 > 7. With respect to the bifactor model (BM3), the only index that showed a difference was Akaike’s index. This index favored the three correlated factor model in the nested model with constraints on the measurement weights (ΔAIC = 8.60), but favored the bifactor model in the nested models with constraints on the structural covariances (ΔAIC = 15.60) and on error variances (ΔAIC = 15.60) (Table 2).
The hierarchical model
The hierarchical model (HM3) showed that all its parameters were significant in the six nested models when being estimated in the two samples of students. In the higherorder submodel, the general factor showed convergent validity (AVE = 0.71 > 0.44 and ω = 0.88 > 0.70). In the lowerorder submodel, the three hierarchical factors also showed convergent validity with a measurement weight mean > 0.50 (M_{λ} from 0.53 to 0.67), the AVE > 0.28 (AVE_total from 0.29 to 0.45) and ωh > 0.70 (ωh_total from 0.83 to 0.73). However, the direct effect of the hierarchical factor was poor both in the masturbation factor and in that of sexual shyness, shame and modesty (AVE_HF < 0.08, ωh_HF and ECV_HF < 0.30); even in the total evaluation the effect of the hierarchical factors was poor (ωh_HF = 0.26 and mean of ECV_HF = 0.29 < 0.30).
The fit of the unconstrained model (UC) was close through seven indices and acceptable through one (SRMR = 0.06) (Table 2). In this model, in the higherorder submodel the structural weights and structural error variances were equivalent between the two samples. However, in the lowerorder model, there were significant differences in six out of the 20 (30%) (total) measurement weights and five out of the 20 (20%) measurement error variances. The measurement weights of virginity and condemnation of pornography factor on items 6 (Z = 4.13, p < 0.001), 15 (Z = 3.01, p = 0.003), and 19 (Z = 1.97, p = 0.049), as well as the measurement weights of the masturbation factor on items 9 (Z = 2.85, p = 0.004) and 17 (Z = 3.27, p = 0.001) were higher in medical students than in psychology students. In addition, the measurement error variances of items 4 (Z = 3.09, p = 0.002), 6 (Z = 2.81, p = 0.005), 15 (Z = 3.07, p = 0.002), and 17 (Z = 3.01, p = 0.003) were lower in medical students than in psychology students. On the contrary, the weight of the masturbation factor on item 13 was higher (Z = 2.66, p = 0.008) and its error variance was lower (Z = 5.15, p < 0.001) in psychology students than in medical students.
The goodness of fit of the unconstrained model (UC) was better compared to the other five nested models for hierarchical model through the indices Δχ^{2}/df > 3, ΔAIC > 10, ΔGFI, ΔAGFI, ΔCFI, ΔRFI and ΔRMSEA > 01 In the model with constraints on measurement error variances (ME), which had the worst fit among the nested models for the hierarchical model, the fit was close through five indices and acceptable through three (NFI, RFI, and SRMR) (Table 2 and Figure 2). The goodness of fit of the hierarchical model (HM3) was equivalent to that of the correlatedfactor model (CF3), except in the nested model with constraints on structural covariances, which was better in the hierarchical model through the indices χ^{2}/df and ΔAIC, as previously mentioned. With respect to the bifactor model (BM3), the only index that showed a difference was Akaike’s index. This index favored the hierarchical model in the nested models with constraints on the measurement weights (ΔAIC = 8.60) and on the structural variances (ΔAIC = 35.82), but it favored the bifactor model in the nested model with constraints on measurement error variances (ΔAIC = 15.60) (Table 2).
The bifactor model
In the unconstrained model (UC), 10 out of the 40 measurement weights were not significant. The weight of the specific factor of masturbation was not significant for items 1 and 7 in both samples and for items 9, 13, and 14 in the sample of medical students. The weight of the specific factor of sexual shyness, shame and modesty was not significant for item 12 in both samples and for items 3, 5, 10, and 16 in the sample of medical students. When all the parameters are restricted (ME), 5 out of the 40 measurement weights were not significant: the weight of the masturbation factor for items 1, 7, and 14, and the weights of the sexual shyness, shame and modesty factor for items 3 and 5.
The three specific factors and the general factor showed convergent validity (Mλ_total from 0.54 to 0.69 > 0.50, AVE_total from 0.30 to 0.48 > 0.28, ω_total from 0.74 to 0.92 > 0.70). The effect of the specific factor was poor both on the six masturbation items (ECV_SF = 0.19 and ωh = 0.12 < 0.30) and in the seven items of sexual shyness, shame and modesty (AVE_SF < 0.08, ECV_SF = 0.27 and ωh = 0.19 < 0.30). On the contrary, the effect of the general factor was excessive on these two factors (ECV_GF = 0.81 and ωh_GF = 0.72 in masturbation and ECV_GF = 0.73 in sexual shyness, shame and modesty) (Table 4).
The goodness of fit of the unconstrained model (UC) was close through the eight indices (Table 2). In this nested model, there were differences in 9 out of the 40 (22.5%) measurement weights and in 4 out of the 20 (20%) variances of measurement errors. The general factor had higher weight on items 3 (Z = 6.59, p < 0.001), 6 (Z = 3.33, p = 0.001), 7 (Z = 3.21, p = 0.001), 9 (Z = 2.27, p = 0.023), 14 (Z = 2.41, p = 0.016), and 19 (Z = 2.01, p = 0.045) in medical students than in psychology students. On the contrary, the weights of the masturbation factor on items 9 (Z = 5.15, p < 0.001) and 13 (Z = 2.66, p = 0.008) were higher in psychology students than in medical students. However, the general factor had higher weight on item 17 (Z = 2.89, p = 0.004) in psychology students than in medical students. The measurement error variances of item 8 (Z = 2.13, p = 0.033) and 13 (Z = 2.87, p = 0.004) were higher in medical students than in psychology students. On the contrary, the measurement error variances of items 4 (Z = 4.41, p < 0.001) and 15 (Z = 4.54, p < 0.001) were higher in psychology students than in medical students.
The goodness of fit of the unconstrained model (UC) was better compared to the other three nested models (Δχ^{2}/df > 3, ΔAIC > 10, ΔGFI, ΔAGFI, ΔCFI, ΔRFI, and ΔRMSEA > 0.01). In the model with constraints on measurement error variances (ME), which had the worst fit among the four nested models for the bifactor model, the fit was close through five indices and acceptable through three (NFI, RFI, and SRMR) (Table 2 and Figure 3). The goodness of fit of the bifactor model (BM3) was better than that of the singlefactor model (1F), but was not clearly differential with respect to the correlatedfactor model (CF3) and the hierarchical model (HM3), since only one index was differential (ΔAIC) and this was contradictory from one nested model to another.
Taking into account all the properties, the bifactor model (BM3) was not invariant between the two samples, like the previous hypothetical models. Although it had the best fit to the data, this was not clearly better than the fit of the correlatedfactor (CF3) and hierarchical (HM3) models. It had nonsignificant parameters (measurement weights of two specific factors) and presented discriminant validity problems in the content domains of masturbation and sexual shyness, shame and modesty with an excessive contribution of the general factor to the detriment of the specific factor. Consequently, it was not a good model. The hierarchical model (HM3) shared the same discriminant validity problem between the general higherorder factor and the hierarchical lowerorder factor in the domains of masturbation and sexual shyness, shame and modesty as the bifactor model.
The correlatedfactor model (CF3) yields the model with the best properties. All its parameters were significant and its three factors showed convergent and discriminant validity. Its fit in the unconstrained model between both student samples was close through seven indices and acceptable through one, as well as equivalent to that of the other two threefactor models (HM3 and BM3) and better than the singlefactor model (1F). Even though its fit worsened when parameters were constrained between the two samples, this ranged from good through five indices to acceptable through three in the model with all the constrained parameters (ME), and never was bad. Therefore, it was a good model, but not invariant between psychology and medical students.
Internal consistency reliability
The overall internal consistency reliability was good (ordinal α = 0.89; 95% CI: 0.87, 0.90), without difference between both samples (F[401, 197] = 1.27; 95% CI: 0.99, 1.61; right tail p = 0.055). Also, the reliability of the masturbation factor was good (ordinal α = 0.84; 95% CI: 0.82, 0.86) and without difference between both samples (F[401, 197] = 1.16; 95% CI: 0.91, 1.47; right tail p = 0.229). The reliability of the sexual shyness, shame and modesty factor was acceptable (ordinal α = 0.70; 95% CI: 0.66, 0.74) and without difference between both samples (F[401, 197] = 1.01; 95% CI: 0.79, 1.28; right tail p = 0.924).
Regarding the appraisement of virginity and condemnation of pornography factor, there was a significant difference (F[401, 197] = 1.39; 95% CI: 1.08, 1.76; right tail p = 0.010). Its reliability was acceptable among psychology students (ordinal α = 0.79; 95% CI: 0.76, 0.82) and good among medical students (ordinal α = 0.85; 95% CI: 0.82, 0.88). The reliability of this factor in the pooled sample was good (ordinal α = 0.81; 95% CI: 0.79, 0.83) (Table 3).
Description of the distribution of the ASS20 and its three factors
In Table 5, the descriptive statistics and normality tests are presented for the ASS20 total score and its three factors in both student samples and in the pooled sample. The scores on the ASS20 and appraisement of virginity and condemnation of pornography factor followed a normal distribution in both student samples and in the pooled sample. The distribution of scores on the rejection of masturbation and sex factor showed positive skewness or long tail to the right in psychology students (Z_{Sk} = 4.09 > 1.96), platykurtosis or shortened tails in medical students (Z_{K} = 2.04 <1.96) and both positive skewness and platykurtosis in the pooled sample (Z_{Sk} = 4.68 > 1.96 and Z_{K} = 2.25 < 1.96, respectively); therefore it did not follow a normal distribution model. Scores on attitude towards sexual shyness, shame and modesty factor showed platykurtosis or shortened tails in psychology students and in the total sample, and thus they also deviated from normality. As in previous cases, the deviation from normality was slight. In medical students, the correlation between the theoretical (t) and empirical (e) quantiles in the normal quantilequantile plot was unitary (r_{et} = 0.99, [0.97, 1.01]), the null hypothesis of normality is maintained with a p value greater than 0.01 with the D’AgostinoPearson test using D’Agostino et al.’s (1990) population formulas (K^{2} = 7.41, right tail p = 0.025) and er than 0.05 with Fisher’s (1930) unbiased formulas (χ^{2}[2] = 4.56, right tail p = 0.102). Its mean (M = 2.34; 95% CI: 2.25, 2.43), median (Mdn = 2.43), and mode (Mo = 2.43, 11.1% of the scores) had very close values, and the histogram profile described a bellshaped curve; consequently, its approximation to normality was acceptable.
Attitudinal levels and mean differences among factors and between men and women
In the psychology students, the mean on appraisement of virginity and the condemnation of pornography showed an ambiguous attitude (2.6 < M = 2.92 < 3.4). The attitudes were liberal in the ASS20 total score and the other two factors (Table 5). When comparing the means of the three factors, there was a significant difference (F[1.91, 764.60] = 534.96, p < 0.001; without assuming homogeneity of variances: Mauschly’s W = 0.93, χ^{2}[2, N = 198] = 14.71, p = 0.001 and using the GreenhouseGeisser correction for the degrees of freedom). The effect size of each factor on the attitudinal level was very large (η_{p}^{2} = 0.57 and ε_{p}^{2} = 0.72 > 0.14).
Likewise, in the medical students, the mean on appraisement of virginity and the condemnation of pornography showed an ambiguous attitude (2.6 < M = 2.88 < 3.4) and the attitudes were liberal in the total score and the other two factors (1.8 < M < 2.6) (Table 5). When comparing the means of the three factors, there was a significant difference (F[1.865, 367.430] = 169.09, p < 0.001; without assuming homogeneity of variances: Mauschly’s W = 0.95, χ2 [2, N = 402] = 20.06, p = 0.001 and using the GreenhouseGeisser correction for the degrees of freedom). The effect size of each factor on the attitudinal level was very large (η_{p}^{2} = 0.46 and ε_{p}^{2} = 0.61 > 0.14). When making pairwise comparisons by Student’s paired t test with the Bonferroni correction for significance level (α_{c} = 0.017), there was a significant difference between the three pairs both in psychology and medical students. The most liberal attitude was towards masturbation and the least liberal was in appraisement of virginity and condemnation of pornography.
When comparing the means in the three factors between both samples by Hotelling’s T^{2} test, there was a significant difference (T^{2} = 22.54, F[3, 594] = 7.49, p < 0.001; assuming homogeneity of covariances between both samples by Box’s M test: M = 10.489, F[6, 1021 920.86] = 1.74, p = 0.108). When making pairwise comparisons, the mean attitude towards masturbation was significantly higher in medical students than in psychology students (MD = 0.21; 95% CI: 0.36, 0.05). In the other two factors, the difference was not significant.
The multivariate distribution of the three factors presented mesokurtosis by the Mardia statistic both in the sample of psychology students (b_{2} = 14.95, Z_{b}^{2} = 0.10, two tail p = 0.921) and medical students (b_{2} = 14.22, Z_{b2} =  0.01, two tail p = 0.994) as in the pooled sample (b_{2} = 14.57, Z_{b2} = 0.964, two tail p = 0.335). It also presented symmetry in the sample of psychology students (U[3] = 4.70, right tail p = 0.195, Z_{U} = 0.87 with the WilsonHilferty transformation of a chisquare variable to normality), but it showed very slight skewness in the sample of medical students (U[3] = 9.77, right tail p = 0.021, Z_{U} = 2.04) and in the pooled sample (U[3] = 9.08, right tail p = 0.028, Z_{U} = 1.91). By multivariate version of JarqueBera test (χ^{2}[3+1 = 4] = U + Z_{b2}^{2}), the null hypothesis of multivariate normality would hold in the sample of psychology students with a significance level of 0.05 (χ^{2}[4] = 4.71, right tail p = 0.318). In the sample of medical students (χ^{2}[4] = 9.77, right tail p = 0.044) and in the pooled sample (χ^{2}[4] = 10.01, right tail p = 0.040), the 95% confidence level should be increased to 96% to maintain the null hypothesis of normality. If the test statistic is calculated using the WilsonHilferty transformation to normality applied to the chisquare statistic of skewness (χ^{2}[2] = Z_{U}^{2} + Z_{b2}^{2}), the assumption of multivariate normality would hold in all three samples (χ^{2}[2] = 0.76, right tail p = 0.684 in psychology students, χ^{2}[2] = 4.18, p = 0.124 in medical students, and χ^{2}[2] = 4.59, p = 0.101 in pooled sample). Consequently, there was a good approximation to multivariate normality as required by the Hotelling’s T^{2} and Box’s M tests.
Concurrent validity in relation to the attitude towards gay men
The AHS10 was only administrated in the sample of psychology students. Its mean (M = 2.24; 95% CI: 2.17, 2.31) and median (Mdn = 2.20) overlapped, reflecting an attitude of acceptance [1.8, 2.6). The distribution of AHS10 total scores was mesokurtic (Z_{K} = 0.55), but was skewed slightly to right (Z_{Sk} = 3.48), and thus it did not adjust to normality (D’Agostino K^{2} = 11.93, right tail p = 0.003; with Fisher’s unbiased formulas: χ^{2}[2] = 12.44, p = 0.002). However, the correlation between the theoretical (t) and empirical (e) quantiles in the normal quantilequantile plot was unitary (r_{et} = 0.997; 95% CI: 0.989, 1.005) and the histogram profile described a bellshaped curve. Consequently, the distribution of AHS10 total scores approached a normal distribution.
ATGS was administered only in the sample of medical students. Its mean (M = 4.62; 95% CI: 4.35, 4.89) and median (Mdn = 4.60) overlapped, reflecting an ambiguous attitude [4.2, 5.8). The histogram profile described a bellshaped curve. The distribution of ATGS total scores was symmetric (Z_{Sk} = 0.73) and mesokurtic (Z_{K} = 1.69), and followed to normal distribution (D’Agostino K^{2} = 5.65, right tail p = 0.059; with Fisher’s formulas: χ^{2}[2] = 3.39, right tail p = 0.183). The correlation between the theoretical (t) and empirical (e) quantiles in the normal quantilequantile plot was unitary (r_{et} = 0.99; 95% CI: 0.98, 1.01). Consequently, the distribution of ATGS total scores followed a normal distribution. Table 6 shows the correlations between ASS20 and the two scales of attitude towards male homosexuality. All the correlations were significant, direct, and statistically equivalent between both samples. The strength of the association of the ASS20 and its first two factors was medium with the two scales of attitude towards homosexuality; on the other hand, strength of the association of the sexual shyness, shame and modesty factor was small; the greater the rejection of sexuality, the greater the rejection of gay men.
When comparing the four correlations within the sample of psychology students, considering the Bonferroni’s correction for the significance level due to multiple comparisons (α = 0.05/6 = 0.0083), the correlation of AHS10 with the ASS20 total score was significantly higher than with the VCP and SSM factors (Steiger’s Z = 2.79, two tail p = 0.005; and Steiger’s Z = 4.85, two tail p < 0.001, respectively). There was no significant difference between the correlations of three factors with AHS10 (Table 7). Within the sample of medical students, the correlation of ATGS with SSM was significantly lower than with the ASS20 total score (Steiger’s Z = 4.87, two tail p < 0.001) and with the VCP factor (Steiger’s Z = 3.11, two tail p = 0.002) (Table 7). Additionally, the correlation of AHS10 or ATGS with MAS would be significantly higher than with SSM in each of the samples, but without the Bonferroni’s correction (Steiger’s Z = 2.58, two tail p = 0.010 in psychology students; and Steiger’s Z = 2.407, two tail p = 0.016 in medical students) (Table 7).
Factor structure of ASS20
The first objective of the study was to verify whether the threefactor model proposed by Moral and Ortega (2008) was valid in psychology and medical students and even invariant across the two types of students. Previously, only a correlatedfactor model had been tested (Moral and Ortega, 2009). In this study, two additional models are specified that theoretically justify the calculation of a total score, apart from the factor scores. On the one hand, there is the bifactor model in which each item is directly determined by a general factor of attitude towards sexuality and one of the three specific factors (DomínguezLara and Rodríguez, 2017; Reise, 2012). On the other hand, there is the hierarchical model in which each item is directly determined by one of the three hierarchical lowerorder factors and indirectly by the general higherorder factor of attitude towards sexuality (lowerorder submodel); in turn, the three hierarchical lowerorder factors are directly determined by the general higherorder factor (higherorder submodel) (Brunner et al., 2012).
The bifactor model was expected to present the best fit and invariance properties between both types of students (Cucina and Byle, 2017; Reise, 2012). Furthermore, its three specific factors were also expected to have convergent and discriminant validity. This expectation was not fulfilled. Its goodness of fit was equivalent to the correlatedfactor model. It had some nonsignificant parameters and showed discriminant validity problems with an excessive weight of the general factor to the detriment of the specific factor on the six masturbation items and on the seven items of sexual shyness, shame and modesty. Consequently, it was a bad model. The hierarchical model also showed discriminant validity problems with a poor hierarchical factor effect on the six masturbation items and on the seven sexual shyness, shame and modesty items. The correlatedfactor model was the best model. Its goodness of fit was equivalent to the bifactor and hierarchical models, and showed convergent in its three factors and discriminant validity between them.
When stating that the correlatedfactor model better represents the interrelation between the 20 items, the calculation of a total score is not theoretically justified as in the bifactor and hierarchical models. The justification for its calculation becomes merely practical, based on the positive interrelation between the factors, as well as on its usefulness, such as having cutoff points to classify people in ordered categories (liberal, ambiguous, and conservative attitude). The unconstrained model for correlatedfactor model had the best fit compared to the other nested models, which was close, and had parameters with significant differences between both samples, as it also happened with the bifactor and hierarchical models. Therefore, it is a valid model for both types of students, but not invariant. What differences does the correlatedfactor model show?
Masturbation is more frequently defined in terms of “mental dirt” (item 13) among psychology students than among medical students; on the other hand, masturbation is also frequently defined in terms of a psychopathological phenomenon (item 9) and practice that deteriorates health (item 17) among medical students than among psychology students. Another difference is that the positive linear association between the masturbation factor and the appraisement of virginity and condemnation of pornography factor was stronger among medical students than among psychology students (six tenths of the explained variance versus four tenths). Moreover, the importance given to remain virgin until marriage (item 6), to assume that premarital sex is immoral (item 15) and to consider pornography as a corrupting influence to mind (item 19) are more defined indicators of the appraisement of virginity and condemnation of pornography among medical students than among psychology students. Another difference is that the sexual shyness, shame and modesty factor had lower variability (more uniformity) among medical students than among psychology students.
Reliability of ASS20 and its factors
The second objective of the study was to estimate the internal consistency reliability of ASS20 and its three factors. In accordance with expectations (Moral and Ortega, 2008, 2009), the level of overall internal consistency reliability in the pooled sample was good, and it was also so for the first two factors (the factor that assesses appraisement of virginity and condemnation of pornography, and the other one related to rejection of masturbation and sex). There was no difference between both samples in the internal consistency values of the scale and the first factor (appraisement of virginity and condemnation of pornography); there was a difference in the second factor (rejection of masturbation and sex), which showed higher internal consistency among medical students (good) than among psychology students (acceptable). This higher consistency, together with the more defined weights of items 9 (masturbating is for sick minds) and 17 (masturbation is bad), indicate that the image of masturbation as a psychopathological phenomenon is widely shared among medical students. Also, according to expectations, the factor with the lower internal consistency, at an acceptable level, was sexual shyness, shame, and modesty. As it was the case for the scale and the first factor, this third factor did not show any significant difference between both samples.
Distributions and averages of the scores on ASS20 and its factors
The third objective of this study was to describe the distributions of scores on ASS20 and its three factors. According to expectations (Moral and Ortega, 2009), the scores on the scale and the factor for assessing appraisement of virginity and condemnation of pornography followed a normal distribution. The other two factors deviated slightly from normality. In the case of the factor related to rejection of masturbation and sex, the deviation was due to slight skewness with a long tail to the right (majority of cases on the liberal pole, with a few cases, very distant from each other, on the conservative pole) among psychology students and in the pooled sample; among medical students, this skewness was due to slight platykurtosis (lower number of cases in the tails than the expected for a normal distribution). This higher number of cases in the central area (between the two shoulders or points of inflection) than that corresponding to a normal distribution, together with the symmetry of a bellshaped curve profile, reflects a very representative mean, which is significantly higher among medical students than among psychology students, and reflects a less liberal attitude. In addition, the higher internal consistency, the higher measurement weights, and the smaller measurement residuals in the masturbation factor among medical students than among psychology students show a great homogeneity in this attitudinal aspect among medical students. In the case of the factor of sexual shyness, shame, and modesty, the deviation from normality was due to platykurtosis in the pooled sample and in each one of the two student samples. However, leptokurtosis was very slight among medical students, only detectable with the AnscombreGlynn transformation, but not with the Fisher’s unbiased estimators, thus allowing to support the null hypothesis of normal distribution by the three statistical tests (K^{2} with population or unbiased statistics and normal QQ plot) among medical students.
It should be noted that these deviations from normality in the scores of the last two factors of the ASS20 are mild, so that there is a good approximation to multivariate normality. Consequently, these distributions show that there is freedom in the attitudinal expression of these young university students (Lyon, 2013). Their attitudes are determined by multiple factors with linear effects, and none of these factors has an excessive weight that could skew and concentrate the scores on the scale and distort the bellshaped curve profile corresponding to the normal probability law (Frank, 2009; Gould, 2002).
The average attitude was liberal both on the scale as well as on the factor related to rejection of masturbation and sex and on the factor related to sexual shyness, shame, and modesty, such as it has been reported previously in other studies in the population of university students (Moral and Ortega, 2008; Menshawy et al., 2020; Salameh et al., 2016). Nevertheless, the average valuation in the appraisement of virginity and condemnation of pornography was in the range of ambiguity in both samples of students. The averages were differential between the three factors within each sample. These attitudinal levels can be ordered according to the level of liberality in the following sequence: attitude towards masturbation, attitude towards nudity and carnality, and attitude towards virginity and pornography. The fact that the attitude towards the appraisement of virginity and the condemnation of pornography is the least liberal aspect is concordant with studies about sociohistorical premises in Mexico (VelascoMoncada and HernándezGonzález, 2017; DíazLoving et al., 2015) and Latin America (Alarcón, 2005). These studies show that female virginity, until the wedding night, is one of the existing and still valued sociohistorical premises in Latino culture.
The most liberal average appeared in the attitude towards masturbation, and this attitude was significantly more liberal among psychology students than among medical students. There was also a significant difference in the attitude towards gay men, which was liberal among psychology students and ambiguous among medical students. These differences could be attributed to the depathologization of both sexual behaviors in the scientificacademic context outside the nonpsychoanalytic paradigms (Frank, 2016; Regnerus et al., 2017). Nevertheless, the classification of these behaviors as perverse still persists within the classical psychoanalytic currents, as in Freudian, Kleinian, and Lacanian schools (Frank, 2016; Kunzel, 2020). Precisely, these psychoanalytic currents originate and prevail in the medicalpsychotherapeutic field (Paris, 2017), while cognitivebehavioral, constructivist, and psychosocial currents, which conceptualize both sexual behaviors as natural, prevail in the different fields of psychology (David et al., 2018).
Concurrent validity of ASS20 regarding male homosexuality
The fourth objective of this study was to verify convergent validity of ASS20 in relation to the attitude towards male homosexuality. The expectation was to find a positive and significant association, with no difference between both samples/scales (Moral and Valle, 2014), and this expectation was met. According to the content, the strength of association of the two scales of attitude towards homosexuality varied from medium (with the ASS20 total score and the first two factors) to small with the third factor (sexual shyness, shame and modesty). The two most related aspects were the appraisement of virginity and condemnation of pornography and the subtle rejection of male homosexuality, with almost a quarter of shared variance. As previously mentioned, this factor reflects the more culturally conservative aspect of ASS20. It should be noted that homosexuality has traditionally been rejected in Western culture and only in recent decades has a change occurred (Redman, 2018). First, this change was towards their tolerance and, progressively, towards their acceptance as a natural and valid expression of sexual orientation (Kite and BryantLees, 2016); thus, the rejection is essentially subtle, symbolic, or overlapped.
In the present study we chose to measure the attitude towards gay men, since the adaptation of the ATLG in Mexico (Moral and Valle, 2011) allowed us to define a subtle rejection factor and because the rejection towards homosexuality in men is a more ingrained cultural aspect (Kite and BryantLees, 2016). Nevertheless, the rejection towards gay men and towards lesbians is strongly related (Moral and Valle, 2011), hence it should not make a big difference if the Attitude Toward Lesbian Scale (ATL) is used to assess concurrent validity.
We used an incidental nonprobability sampling, and thus the conclusions should be taken with due caution. There are no paired data of the participants in a short term (days or weeks) or a medium (months) or long term (years); consequently, the temporal reliability and factor model stability could not be tested. The design was nonexperimental, hence we only could talk about associations between attitudes, and we could not make any causal inference.
Among these Mexican students of psychology from a public university and among these medical students from a private university, the model composed of three correlated factors that was originally proposed shows a close fit, convergent validity in each factor, and discriminant validity between its factors. Nevertheless, it is not invariant across the two types of students. Items 6, 15, and 17 are better explained (higher weight and lower residual) by their corresponding factor, and items 9 and 19 have a greater measurement weight among medical students than among psychology students. These five items belong to the factors related to the appraisement of virginity and condemnation of pornography and to the factor related to rejection of masturbation and sex, which show a higher correlation among medical students than among psychology students. Among psychology students, item 13 (sex is dirty) is better explained by its factor (rejection of masturbation and sex) than among medical students. The level of internal consistency reliability of ASS20 and its first factor (appraisement of virginity and condemnation of pornography) are good, whereas it was acceptable for the factor related to sexual shyness, shame, and modesty, without any significant difference between both samples. There is a significant difference in the level of reliability of the second factor (rejection of masturbation and sex); its level is good among medical students and acceptable among psychology students. Scores on ASS20 and the factor related to the appraisement of virginity and condemnation of pornography follow a normal distribution. The deviations from normality in the other two factors are very mild, so that there is a good approximation to multivariate normality. The average attitude, interpreted in an absolute sense, is liberal, except for the appraisement of virginity and condemnation pornography, which is ambiguous. The only difference in means between the two samples is the more liberal attitude towards masturbation among psychology students than among medical students. ASS20 shows convergent validity in relation to the attitude towards male homosexuality, assessed with two different instruments. The correlation of those two instruments with the ASS20 total score and its first two factors are medium, whereas the correlation with the third factor is low; there are no differences in these correlations between both samples or both scales of attitude towards homosexuality.
The use of this scale is recommended to assess and study the attitude towards sexuality in Mexican health science university students. ASS20 can be scaled using the mean and standard deviation estimated through a probability sampling method. Likewise, the study of ASS20 is suggested at the levels of compulsory secondary school and high school, as well as among university students of other careers. Additionally, the factorial invariance across sexes, age groups, or levels of schooling should be tested. From the present data, it can be inferred that, by promoting virginity and the condemnation of pornography (cultural values that still pervade our society), rejection towards homosexuality is being encouraged, especially the implicit type of rejection; these two aspects are not independent and have a rather close association (almost a quarter of shared variance).
The authors have not declared any conflict of interests.
ASS20, Twentyitem Attitude towards Sexuality Scale; AHS10, Tenitem Attitude towards Homosexuality Scale; ATGS, Subtle rejection Attitude Towards Gay men subscale from the Mexican adaptation of Attitudes Towards Lesbians and Gay men (ATLG) scale; MAS, ASS20 rejection of Masturbation and Sex factor; SSM, AAS20 sexual shyness, shame, and modesty factor; VCP, AAS20 appraisement of Virginity and Condemnation of Pornography factor.
REFERENCES
Alarcón R (2005). Premisas históricosocioculturales de la juventud peruana: obediencia filial y virginidad [Historicalsociocultural premises of Peruvian young adults: Parental obedience and virginity]. Revista Latinoamericana de Psicología 37(1):8194.


American Psychological Association (2017). Ethical principles of psychologists and code of conduct. With the 2016 amendment to standard 3.04. Washington, DC: Autor. Available at:
View



Anscombe FJ, Glynn WJ (1983). The distribution of kurtosis statistic b2 for normal statistics. Biometrika 70(1):227234.
Crossref



Brunner M, Nagy G, Wilhelm O (2012). A tutorial on hierarchically structured constructs. Journal of Personality 80(4):796846.
Crossref



Burt MR (1980). Cultural myths and supports for rape. Journal of Personality and Social Psychology 38(2):217230.
Crossref



Byrne B (2016). Structural equation modeling with AMOS (3rd ed.). New York, NY: Routledge.
Crossref



Clark A (2019). Desire. A history of European sexuality (2nd Ed.). London: Routledge.
Crossref



Cucina J, Byle K (2017). The bifactor model fits better than the higherorder model in more than 90% of comparisons for mental abilities test batteries. Journal of Intelligence 5(3):27.
Crossref



D'Agostino RB (1970). Transformation to normality of the null distribution of g1. Biometrika 57(3):679681.
Crossref



D'Agostino RB, Belanger A, Ralph B, D'Agostino Jr RB (1990). A suggestion for using powerful and informative tests of normality. The American Statistician 44(4):316321.
Crossref



D'Agostino RB, Pearson ES (1973). Testing for departures from normality. Empirical results for the distribution of b2 and √b1. Biometrika 60(3):613622.
Crossref



David D, Cristea I, Hofmann SG (2018). Why cognitive behavioral therapy is the current gold standard of psychotherapy. Frontiers in Psychiatry 9(4):13.
Crossref



DíazLoving R, Saldivar A, ArmentaHurtarte C, Reyes NE, López F, Moreno M, Romero A, Hernández JE, Domínguez M, Cru C, Correa FE (2015). Creencias y normas en México: una actualización del estudio de las premisas psicosocioculturales [Beliefs and norms in Mexico: An update of the study of psychosociocultural premises]. Psykhe 24(2):125.
Crossref



DomínguezLara S, Rodríguez A (2017). Índices estadísticos de modelos bifactor [Statistical indices from the bifactor models]. Interacciones 3(2):5965.
Crossref



Feldt LS (1969). A test of the hypothesis that Cronbach's alpha or KuderRichardson coefficient twenty is the same for two tests. Psychometrika 34(3):363373.
Crossref



Feldt LS, Woodruff DJ, Salih FA (1987). Statistical inference for coefficient alpha. Applied Psychological Measurement 11(1):93103.
Crossref



Fisher RA (1930). The moments of the distribution for normal samples of measures of departure from normality. Proceedings of the Royal Society A 130(812):1628.
Crossref



Frank SA (2009). The common patterns of nature. Journal of Evolutionary Biology 22(8):15631585.
Crossref



Frank E (2016). Masturbation. In: Wong A, Wickramasinghe M, Hoogland R, Naples NA (eds.), The Wiley Blackwell encyclopedia of gender and sexuality studies. Hoboken, NJ: Wiley Online Library 2:603614.
Crossref



Gignac GE, Kretzschmar A (2017). Evaluating dimensional distinctness with correlatedfactor models: Limitations and suggestions. Intelligence 62(1):138147.
Crossref



Gould SJ (2002). The structure of evolutionary theory. New York: Harvard University Press.
Crossref



Hendrick C, Hendrick SS, Reich DA (2006). The brief sexual attitudes scale. Journal of Sex Research 43(1):7686.
Crossref



Henseler J, Ringle CM, Sarstedt M (2015). A new criterion for assessing discriminant validity in variancebased structural equation modeling. Journal of the Academy of Marketing Science 43(1):115135.
Crossref



Herek GM (1984). Attitudes toward lesbians and gay men: A factor analytic study. Journal of Homosexuality 10(1/2):3951.
Crossref



Honold JA (2005). Diseño de un instrumento para la evaluación de información y actitudes ante la sexualidad [Design of an instrument for the evaluation of information and attitudes towards sexuality]. (Bachelor thesis). Universidad Nacional Autónoma de México, Ciudad de México. Available at:
View



Imhoff D, Dreizik, M, Brussino S (2020). Análisis psicosocial del prejuicio hacia trabajadoras sexuales [Psychosocial analysis of prejudice towards sex workers]. Revista CS 30(1):173196.
Crossref



Joenssen DW, Vogel J (2014). A power study of goodnessoffit tests for multivariate normality implemented in R. Journal of Statistical Computation and Simulation 84(5):10551078.
Crossref



Kankainen A, Taskinen S, Oja H (2007). Tests of multinormality based on location vectors and scatter matrices. Statistical Methods and Applications 16(1):357379.
Crossref



Kite ME, BryantLees KB (2016). Historical and contemporary attitudes toward homosexuality. Teaching of Psychology 43(2):164170.
Crossref



Kite ME, Togans LJ, Schultz TJ (2020). Stability or change? A crossâ€cultural look at attitudes toward sexual and gender identity minorities. In: Keith KD (ed.), Crossâ€Cultural psychology: contemporary themes and perspectives (2nd Ed.). Hoboken, NJ: Wiley Online Library pp. 427488.
Crossref



Koizumi K, Okamoto N, Seo T (2009). On JarqueBera tests for assessing multivariate normality. Journal of Statistics: Advances in Theory and Applications 1(2):207220.



Kunzel R (2020). The "durable homophobia" of psychoanalysis. Modern Intellectual History 17(1):215219.
Crossref



LimaSerrano M, SáezBueno A, CáceresRodríguez, B, LimaRodríguez JS (2013). Diseño y validación de escalas para medir la actitud adolescente hacia: sexualidad, sustancias adictivas y seguridad vial: ¿Se relacionan con los comportamientos? [Design and validation of scales to measure adolescent attitudes toward sexuality, addictive substances and road safety. Are they related to behaviour?]. Anales del Sistema Sanitario de Navarra 36(2):203215.
Crossref



Lyon A (2013). Why are normal distributions normal? The British Journal for the Philosophy of Science 65(3):621649.
Crossref



Mardia KV (1970). Measures of multivariate skewness and kurtosis with applications. Biometrika 57(3):519530.
Crossref



Menshawy A, Abushouk AI, Ghanem E, Senna M, Lotfy N, AbdelMaboud M, Elmaraezy A, Elsherbeny MY, the SABEAD study group (2020). Break the silence: knowledge and attitude towards sexual and reproductive health among Egyptian youth. Community Mental Health Journal, pp. 19.
Crossref



Moral J (2010). Religión, significados y actitudes hacia la sexualidad: un enfoque psicosocial [Religion, meanings and attitudes toward sexuality: a psychosocial approach]. Revista Colombiana de Psicología 19(1):4559.



Moral J (2019). Revisión de los criterios para validez convergente estimada a través de la Varianza Media Extraída [Review of the criteria for convergent validity estimated through the Average Variance Extracted]. Psychologia 13(2):2541.
Crossref



Moral J, MartínezSulvarán JO (2011). Escala de actitud hacia la homosexualidad: propiedades psicométricas y aspectos diferenciales por sexos [Attitude towards homosexuality scale: psychometric properties and differential aspects by sex]. SOCIOTAM 21(1):105124.



Moral J, Ortega ME (2008). Diferencias de género en significados, actitudes y conductas asociados a la sexualidad en estudiantes universitarios [Gender differences in meanings, attitudes and behaviors associated with sexuality in university students]. Estudios sobre las Culturas Contemporáneas 14(28):97119.



Moral J, Ortega ME (2009). Representación social de la sexualidad y actitudes en estudiantes universitarios mexicanos [Social representation of sexuality and attitudes in Mexican university students]. Revista de Psicología Social 24(1):6580.
Crossref



Moral J, Valle A (2011). Escala de Actitudes hacia Lesbianas y Hombres Homosexuales en México 1. Estructura factorial y consistencia interna [Attitudes Toward Lesbians and Gays (ATLG) scale in Mexico 1. Factor structure and internal consistency]. Revista Electrónica Nova Scientia 3(6):139157.
Crossref



Moral J, Valle A (2014). Measurement of attitudes toward lesbians and gay men in students of health sciences from northeast Mexico. Journal of Behavior, Health and Social Issues 6(1):5165.



Paris J (2017). Is psychoanalysis still relevant to psychiatry? The Canadian Journal of Psychiatry 62(5):308312.
Crossref



PinosAbad GM, PinosVélez VP, PalaciosCordero MP, LópezAlvarado SL, CastilloNúñez JE, OrtizOchoa WA, JervesHermida EM, Enzlin P (2017). Conocimientos y actitudes hacia la sexualidad y educación sexual en docentes de colegios públicos [Knowledge and attitudes toward sexuality and sexual education in public schools teachers]. Actualidades Investigativas en Educación 17(2):122.
Crossref



Redman SM (2018). Effects of samesex legislation on attitudes toward homosexuality. Political Research Quarterly 71(3):628641.
Crossref



Regnerus M, Price J, Gordon D (2017). Masturbation and partnered sex: substitutes or complements? Archives of Sexual Behavior 46(7):21112121.
Crossref



Reise SP (2012). The rediscovery of bifactor measurement models. Multivariate Behavioral Research 47(5):667696.
Crossref



Salameh P, Zeenny R, Salamé J, Waked M, Barbour B, Zeidan N, Baldi I (2016). Attitudes towards and practice of sexuality among university students in Lebanon. Journal of Biosocial Science, 48(2):233248.
Crossref



Sanabria H (2019). Gender, sexuality, and reproduction. In The Anthropology of Latin America and the Caribbean (2nd ed.) (pp. 150183). London: Routledge.
Crossref



TrejoPérez F, DíazLoving R (2013). En torno a la sexualidad: actitudes y orientación sociosexual en una muestra mexicana [In regards to sexuality: attitudes and sociosexual orientation in a Mexican sample]. Psicología Iberoamericana 21(1):715.



VelascoMoncada AP, HernándezGonzález E (2017). Tradicionalismo en mujeres adultas de Jamay, Jalisco desde las premisas psicosocioculturales de la familia mexicana [Tradicionalism in adult women of Jamay, Jalisco from psychosociocultural premises of Mexican family]. Revista Electrónica de Psicología Iztacala 20(4):12961313.



Viladrich C, Angulo A, Doval E (2017). A journey around alpha and omega to estimate internal consistency reliability. Annals of Psychology 33(3):755782.
Crossref



Wilson EB, Hilferty MM (1931). The distribution of chisquared. Proceedings of the National Academy of Sciences of the United States of America 17(12):684688.
Crossref

