The Social Determinants of Polymorphous Prejudice Against Lesbian and Gay Individuals: the Case of Portugal

In the present article, we analyze polymorphous prejudice against lesbians and gays according to a sample of Portuguese heterosexual individuals. We tested the differential importance of demographic-, ideological-, and psychological-level variables to predicting this phenomenon. Our results show that male, Catholic, right-wing respondents with fewer lesbian and gay (LG) friends prove the group exhibiting the highest levels of polymorphous prejudice. Furthermore, the introduction of psychological-level variables into the regression models increased the explained variance in polymorphous prejudice, above and beyond the remaining predictors. Additionally, we report different patterns of results when deploying regression analyses at the level of the sub-scales of polymorphous prejudice. We discuss our results within the light of contemporary sexual prejudice frameworks before reviewing the utility of results to interventions targeting discriminated LG individuals.

Sexual prejudice is a prevalent phenomenon in our societies and has captured the interest of psychology since the 1970s (Herek 2000). Under the guise of homophobia, i.e., Bthe dread of being in close quarters with homosexuals-and in the case of homosexuals themselves, self-loathing^ (Weinberg 1972, 4), this term has been widely applied in psychological research (Clarke et al. 2010) and in lesbian, gay, bisexual, trans, and queer (LGBTQ) activism. However, as pointed out by Herek (2004), this represents a problematic concept because of its pathologizing assumptions. As Kitzinger (1999) also argues, this pathological assumption reduces a social phenomenon to an individual issue (prejudice as resulting from an individual's mental illness), thus de-socializing a collective problem. Homophobia thus mainly applies to understanding individual attitudes towards LGBTQ individuals (Herek 2004), reinforcing the individualistic tone of that line of study, and neglecting the social dimensions to this phenomenon.
Research on sexual prejudice normally encompasses the study of predictors of this form of prejudice (Vincent et al. 2011) and, in a more particular way, the specific predictors of sexual prejudice towards LGBT individuals (e.g., Basow and Johnson 2000;Herek and Capitanio 1996;Lemm 2006). Nevertheless, this research has mainly focused upon English-speaking countries such as the USA. In the present article, we analyze some of the social determinants of prejudice against LG individuals in Portugal, a country that recently expanded the legal rights of LGBTs. Furthermore, in this article, we also present the adaptation, validation, and core psychometric qualities of the Massey (2009) Polymorphous Prejudice Scale (PPS) for the Portuguese context.

Sexual Prejudice
The critiques raised as regards the concept of homophobia have served as the grounds for some researchers to shift their focus towards other conceptualizations, such as heterosexism applied by Herek (2000) to describe the ideology categorizing homosexuality as an inferior sexual orientation vis-à-vis heterosexuality. Due to its marked macro-social emphasis, Herek (2000) and Herek and McLemore (2013) propose recourse to sexual prejudice as a means of referring to the negative attitudes expressed against homosexuals, their behaviors, and communities. Herek (2009) also refers to sexual stigma as a specific instance of sexual prejudice in labeling it as the socially shared knowledge about the devalued status of homosexuality in society as compared to heterosexuality. In his view, sexual stigma creates social roles and behavioral expectations that inform and guide negative attitudes towards LG individuals.
Yet another strand of research, more closely associated with queer theory, identifies the need to study heteronormativity, the norm that adopts heterosexuality as the reference framework for all individuals, classifying LGBTQ individuals as deviations towards such norms (Warner 1993). These norms also hold great relevance to issues surrounding the recognition of humanity as Butler (2004) points out by conveying how the intelligibility of humanity is anchored on gender and sexuality norms. Hegemonic heterosexuality, through these norms, ranks and maintains LGBTQ populations in more fragile and vulnerable positions than other populations. Such a position bears obvious consequences in terms of citizenship and rights as well as for sexual prejudice.
Therefore, there are several ways of analyzing expressions of LGBTQ-targeted prejudice with all of them deployed in psychology (Clarke et al. 2010). Furthermore, one important theoretical discussion, which began during the 1980s (with practical and metric consequences), concerns the dimensionality of sexual prejudice. According to Hegarty and Massey (2006), Herek's (1984) position about the multidimensional scales applied to measuring homophobia or other related constructs triggered this debate. Herek (1984) concluded from research incorporating his BAttitudes Towards Gay Men and Lesbian Scale^(ATGL) that these attitudes were one dimensional and therefore his scale adopts a tolerancecondemnation continuum. This scale has since been used in numerous studies (e.g., Cárdenas and Barrientos 2008;Herek and Gonzalez-Rivera 2006;Stotzer 2009) and reflecting the same one-dimensional structure of the original scale.
However, as argued by Hegarty and Massey (2006), this constitutes a problem when dealing with modern versions of prejudice, on the one hand, and when doing research into the multiple functions and aspects of sexual prejudice (for a discussion, see also Herek 1986) on the other hand. Indeed, while traditional sexual prejudice is commonly associated to moralistic and religious perceptions of LGBT individuals, modern sexual prejudice is subtler and presumably free from normative pressures (Teney and Subramanian 2010) as sexual prejudice has since become normatively proscribed. This implies how the phenomenon of sexual prejudice has itself become increasingly complex (Morrison et al. 2005;Walls 2008) and thus demanding more complex and abridging measures to capture it (Massey 2009). Briefly, the rationale for these changes in research approaches stems from changes in the expression of sexual prejudice in keeping with the more camouflaged and subtler dimensions of this prejudice emerging in the meanwhile.
This argumentation also represented an input into research on sexual prejudice with studies correspondingly deploying multidimensional measures of sexual prejudice not only to cover the traditional prejudice dimensions concerning tolerance but also to include other measures and sub-scales (Walls 2008). For example, in the case of sexism, Glick and Fiske (2001) propose both a more benevolent and subtle form of sexism based on protecting and supporting women and therefore conveying a perception of women as Bpure creatures who ought to be protected, supported, and adored and whose love is necessary to make a man complete^(p.109); and a more hostile sexism, mainly focused on traditional dimensions devaluating women and on maintaining traditional gender roles. This encapsulates the fact that sexism is ambivalent and therefore requires more than one factor to measure its dimensions. Consequently, this led to the proposal of more complex measurements of sexual prejudice in order to reflect the changes in Western societies regarding homosexuality (Seidman 2002) and the multiple entanglements between attitudes towards homosexuality and other related value systems such as pro-gay equality orientation and awareness about gender normalization (Martinez 2011).
Yet another aspect that requires attention as regards modern expressions of sexual prejudice concerns the legal changes endowing equality to lesbian and gay communities rights, specifically the legal right to marry and the explicit prohibition of discrimination based on sexual orientation. Portugal has experienced significant changes in the past four decades following the Carnation Revolution and the ratification of the Democratic Constitution in 1976, in particular enlarging the set of social and civil rights, especially in the case of gender equality. These constitutional principles are, however, contradicted by the persistence of the discriminatory social practices legitimized by institutions such as the Catholic Church. Therefore, the regulatory influence of the Catholic Church, pivotal during the 48 years of fascism, partially serves to explain these social practices (Santos 2012).
In Portugal, homosexuality remained criminalized from 1912 until the Penal Code of 1982. The accession of Portugal to the EU proved instrumental to the enactment of pro-equality policies. There were correspondingly significant changes in legislative terms concerning the equal status of LG communities (see Oliveira et al. 2013). The main legal mechanisms deployed to produce such a change involved the introduction of sexual orientation as one of the explicit constitutional stipulations against discrimination and subsequently followed by the introduction of laws concerning civil unions (2001), same-sex marriage (2010), and a law on gender identity recognition (2011).
These laws entailed an advancement for LGBT rights even if studies on the perceptions of this population concerning heterosexism still identify patterns of felt discrimination and perceived heterosexism (Nogueira and Oliveira 2010) despite these legal advances (Oliveira et al. 2013). As argued by Carneiro and Menezes (2007) and Santos (2012), the landscape of heterosexism in Portugal remains contradictory since legal regulations stipulating equality coexist with discriminatory practices in real daily life.
In rapidly changing contexts, with such an ambivalent coexistence between formal equality and practical discrimination, new measures of sexual prejudice should tap this phenomenon in all its complexity. Such is the case we present in this paper and thereby applying a multidimensional scale of sexual prejudice-Massey's (2009) PPS.

Polymorphous Prejudice and Its Predictors
Polymorphous prejudice (PP) is a construct that derives its meaning from the notion of queer consciousness (Massey 2009). Queer consciousness encompasses attitudes that go beyond the traditional normative terms of heterosexism, heterocentrism, or sexual prejudice (Delgado et al. 2014;Massey 2004). In this sense, Massey (2009) argues that queer conscious individuals reject social norms supporting gender roles, fixed identities, and biological or psychoanalytic explanations for sexual orientation. On the contrary, they support non-essentialist conceptions of sexual orientation, and their political consequences in terms of value differences and diversity (Delgado et al. 2014;Martinez et al. 2011). The PP construct transfers these theoretical assumptions into operational definitions within a context where sexual prejudice proves increasingly complex and subtle and thus demanding new measurement paradigms.
Within this new paradigm, Massey (2009) proposes a scale for Bpolymorphous prejudice^(PPS) entailing a more comprehensive means of conceptualizing sexual prejudice through a much more complete picture of the relationships between sexual prejudice, egalitarian belief systems, heteronormativity, and perception of progress by sexual minorities. Indeed, this measure draws from a rationale encompassing not only the legacy of gay and lesbian psychology but also other propositions more grounded in historical and cultural context and in queer theory.
Seven sub-scales compose the PPS, each pointing to a different and specific aspect of sexual prejudice towards lesbians and gays (LG). These sub-scales are (1) traditional heterosexism, i.e., the evaluation of attitudes towards moral condemnation of homosexuality and those holding that LG rights should not be recognized.
(2) Denial of continuous discrimination measures beliefs that gay and lesbian discrimination does not exist anymore and that the demands of social movements struggling against discrimination are pointless. (3) Aversion towards gays and (4) aversion towards lesbians, i.e., the negative affective reactions (e.g., discomfort) towards gays and lesbians. (5) Value gay progress, through measuring the attitudes towards social diversity and how this diversity benefits the whole of society, (6) resistance to heteronormativity taps feelings of discomfort with heteronormativity and a need to resist traditional gender stereotypes, and (7) positive beliefs measure the support for the unique consequences deriving from being gay or lesbian in a heteronormative society.
Apart from measuring sexual prejudice against LG individuals according to new and multidimensional measurements, analyzing the predictors of PP is of extreme importance, especially in order to determine which demographic, ideological, or more importantly psychological dimensions better help us understand this multidimensional phenomenon. Indeed, while it is true that equality policies were enforced in some EU countries and in the USA, instances of legal discrimination in these countries that reinforce discriminatory social practices still require tackling. As mentioned above, we still observe the coexistence of traditional and modern forms of sexual prejudice in these same societies (e.g., Nogueira and Oliveira 2010). In this sense, the study of sexual prejudice predictors that encompass both forms of prejudice fosters a fuller understanding of this phenomenon as well as informing practitioners and public policy makers on how best to reduce it.
Demographic and ideological predictors are commonly used in studies analyzing sexual prejudice against LGBT individuals. Indeed, laboratory, public opinion surveys, and questionnaire studies have reported how individuals with higher levels of sexual prejudice are more frequently observed among men than among woman (Morrison and Morrison 2003). Additionally, men normally exhibit more traditional heterosexism (Massey 2009). Regarding educational levels, individuals with low levels of education tend to reject homosexuality more than those who are highly educated (Herek 1993;van den Akker et al. 2013), an association also encountered among older people (Herek 1993). Moreover, individuals with strong religious beliefs or belonging to conservative religious movements and that attend religious services more often tend to exhibit higher prejudice levels towards LG individuals (Costa et al. 2015;Ellis et al. 2002;Herek and Capitanio 1996). Regarding political ideologies, studies also find an association between a conservative political positioning or right-wing political ideology and prejudice towards LG individuals (Yang 1998;Stones 2006). Inversely, and as reported in the literature, lower levels of sexual prejudice come associated with individuals experiencing higher contact levels with gay individuals, such as those reporting having more gay friends or family members (Costa et al. 2015;Herek and Capitanio 1996;Herek and Gonzalez-Rivera 2006).
Despite considerable evidence regarding sexual prejudice predictors, in particular demographic and ideological level variables associated to prejudice towards LG individuals, psychological-level predictors remain underreported in the literature and constituting an obstacle to the full understanding of this topic. Some exceptions to this zeitgeist are the studies on the controllability or noncontrollability of homosexual orientation (Frias-Navarro et al. 2015;Haider-Markel and Joslyn 2008) and the openness to experience and masculinity (Barron et al. 2008). Indeed, there is a lack of empirical evidence regarding the predictors associated with heterosexual individual perceptions regarding LG individuals. More specifically, in this paper, we address two relevant predictors in this domain: discrimination deservingness and recognition of discrimination against LGs. As pointed out by Al Ramiah et al. (2010), deservingness is an important issue in the expression of discrimination and in the legitimization of prejudice, based on historical and day-to-day inequalities and social norms (see also Reyna et al. 2005). Accordingly, the more heterosexuals perceive that LGs deserve to be discriminated against, the more likely they are to be prejudiced against.
The other predictor-recognition of discrimination-refers to more or less biased heterosexual perceptions as to the extent LG individuals experience discrimination. Not only is non-recognition of discrimination a profoundly important topic interconnected with heteronormativity (Roseneil et al. 2013) but it also constitutes a relevant and understudied factor in the process of legitimization of prejudice against minorities (for an exception regarding racial prejudice, see Banfield and Dovidio 2013). In this sense, the lack of recognition on behalf of heterosexuals regarding discrimination against LGs opens a path for expressing prejudice against these individuals.
Apart from introducing these two understudied predictors of prejudice against LG individuals, another important gap in the literature refers to the contexts of sexual prejudice studies. In fact, most of these studies were conducted in the USA, and few have analyzed the impact of similar predictors on the expression of sexual prejudice expression among European samples. In the present article, we analyze this phenomenon in a sample of Portuguese heterosexual individuals.
Therefore, this article presents a test of the differential importance of predictors such as demographical (e.g., gender and LG friends), ideological (e.g., political positioning), and psychological (e.g., discrimination deservingness) in the expression of PP by heterosexuals vis-à-vis LG individuals. More in-depth analysis stems from testing the differential impact of these predictors against the PPS scale sub-dimensions. This strategy enables us to analyze and identify those predictors most contributing to understanding this phenomenon and its respective multiple dimensions.

Ethical Statement
All the procedures performed in this article involving human participants took place in accordance both with the ethical standards of the institutional and/or the national research committee and with the 1964 Helsinki declaration and its later amendments or comparable ethical standards. Additionally, the present study was carried out in accordance with the Ethical Guidelines of the host institution. In this sense, the data were (1) collected anonymously, (2) did not involve questions about undesirable personal characteristics, (3) did not involve participants from a population of concern, (4) did not involve deception, (5) did not involve invasive measures, and (6) did not collect personally identifying information. The study was non-invasive, the participants were in no way deceived, and all data were analyzed anonymously. All participants read an informed consent with the description and purpose of the study and were informed that by proceeding, they consented to participating but that they could withdraw at any stage of the study.

Method Participants
One thousand two hundred and twenty self-reported heterosexual Portuguese-speaking respondents participated in the present study, with 75.1 % female in gender. The mean respondent age was 31.6 (SD = 10.31; 6 participants did not reveal their age). Regarding education, the majority of respondents had completed undergraduate study (58.6 %), with 27.1 % holding Master's degrees or PhDs, and 14.4 % completed basic or secondary school (61 respondents did not reveal their level of education, i.e., 5 % of the sample). About 60 % of respondents were single and with 47 % not expressing any religious confession (although 43.4 % defined themselves as Catholic). Concerning their political identifications, the majority of participants positioned themselves as left wing (53.6 %), followed by 26.6 % positioning themselves as in the Bcenter,^and, to a lesser extent, 10.3 % respondents declared they were right-wing supporters; 4 % positioned themselves on the extreme left, and 0.5 % on the extreme right, with 61 respondents not revealing their political identification (5 % of the total sample).
From this sample, two sub-samples were randomly retrieved (random sampling with replacement). Each of these samples corresponds roughly to 50 % of the main sample.
Sub-sample 1 (N = 748) served to analyze the PPS construct validity by deploying principal axis factoring (PAF) analysis. In sub-sample 2 (N = 771), the structure retrieved from the PAF was tested by confirmatory factor analysis (CFA). These sub-samples were again collapsed in order to conduct the remaining analyses, including the analyses of the PP predictors.

Procedure
Participants were recruited via social network websites (Facebook®) via e-mail, making recourse to different mailing lists (e.g., personal research team members' mailing lists). The questionnaire was run on the Internet using Google Docs® platform and complied with the basic standards and procedures established in the literature as good practice for Internet data collection (namely that participants were engaged in a voluntary and confidential selfreport survey and that they could abandon the study at any point by closing the web browser; Barchard and Williams 2008). Participants were provided with an Internet address where they could fill in the Portuguese version of the PPS (i.e., the criterion variable; Massey 2009) and other measures including sociodemographic characterization measures, as well as the remaining variables acting as predictors. At the beginning of the questionnaire, participants were provided with an informed consent statement and also guaranteed full data collection anonymity and confidentiality. At the end, participants were provided with a debriefing text and thanked. Repeated responses were checked for by verifying that no single Internet Protocol (IP) address was associated with more than one questionnaire (Gosling et al. 2004).

Predictors and Criterion Variable
In order to analyze the determinants underlying the expression of PP against LG individuals, we chose a number of variables regarding their theoretical relevance and their proven association to LG prejudice in previous studies (cf . Herek 2000;Delgado et al. 2014). We classified them in different groups: demographic-such as age, education (1 = primary through secondary school, 2 = major or graduation, 3 = MA/MSc. or PhD); LG friends (1 = yes, 2 = no) and gender (1 = male; 2 = female); ideological-political positioning and religion; and psychological-LG discrimination recognition and LG discrimination deservedness. Regarding political positioning, participants were asked about their political positioning using a scale of 5 points that ranged from B1 = extreme right^to B5 = extreme leftŵ ith each of the points of the scale anchored-B2 = right,B 3 = center,^B4 = left^. The religion variable asked participants whether they were BCatholic,^BChristian, non-Catholic,^Bwithout religion,^or Bother religions.^For the purpose of regression analysis, this variable was dummy coded as 1 = Bother religions,^2 = BCatholic.Î n order to measure LG discrimination recognition, we asked participants BFrom the following groups, please identify those you consider that are discriminated against using a scale ranging from 1 = Not discriminated at all to 7 = Very much discriminated against^(examples of the target groups evaluated: lesbians, gays, transsexuals, women, men, unemployed, gypsies). Finally, to assess LG discrimination deservingness, participants were then asked to rate the following item BFrom the following groups, please refer to those that deserve to be the target of discriminationî n accordance with a scale ranging from 1 = deserve being discriminated against to 7 = do not deserve being discriminated against (with the same target groups as in the previous question). Only those ratings regarding lesbian and gay targets were retained for further analyses.
Massey's (2009) PPS (our criterion variable) was originally composed of 70 items measuring different dimensions of expressions of prejudice towards LG individuals. Massey (2009) reports construct validity and adequate scale reliability (test-retest reliability values of the scale's factors ranging from 0.67 to 0.93). All items were rated on a 5-point Likert-type scale ranging from 1 = Btotally disagree^to 5 = Btotally agree^. Nineteen items measured traditional heterosexism, i.e., the evaluation of LG individuals as immoral, sinful, or perverted, and the denial of certain privileges (e.g., BMale homosexuality is a perversion^; Cronbach alpha = 0.95); 9 items measured the denial of continued discrimination (e.g., BMost lesbians and gay men are no longer discriminated against^; Cronbach alpha = 0.83); 16 items measured aversion towards gay men and lesbians (8 items for each dimension) (e.g. BI'm uncomfortable when gay men act feminine^); 8 items measured values attributed to gay progress (e.g., BI see the lesbian and gay movement as a positive thing^; Cronbach alpha = 0.94); 8 items measured resistance to heteronormativity (e.g., BI feel restricted by the sexual rules and norms of society^; Cronbach alpha = 0.90); and finally, 10 items evaluated general positive beliefs (e.g., BStraight men have a lot to learn from gay men about being friends with women^; Cronbach alpha = 0.86).
All 70 items of the PPS were submitted to a translation-back-translation process. The items were translated into Portuguese by a team of psychologists with disagreements resolved through discussion (reaching a 95 % level of agreement). A native Portuguese speaker residing in the UK made the back-translation of the Portuguese items into the original language (English). The final and original items were compared with discrepancies adjusted to ensure convergence with the original scale items.

Construct Validation of Massey's (2009) Polymorphous Prejudice Scale
In order to determine the factorial structure of the Portuguese version of Massey's PPS scale, its 70 items were subject to PAF with Promax rotation using subsample 1. All items were previously linearized for PAF analysis by calculating the natural logarithm of each participant score on each item. This allowed us to smooth skewed distributions, specifically for those items more prone to socially desirable answers. The PAF extraction method was preferred as this proves especially adequate when the normality of data distribution cannot be guaranteed (Costello and Osborne 2005). Also, we applied Promax rotation since this allows for factors to be correlated, which is the case with the PPS (cf. Massey 2009), and has been identified in the literature as a preferable method of factor rotation (Fabrigar et al. 1999).
The final solution obtained from these analytical procedures yielded a six-factor structure integrating 39 of the original 70 items (KMO = 0.93) and explaining 56.96 % of total variance (see Table 1 for a detailed description of factor loadings, eigenvalues, and reliability coefficients). These six factors were retained through the application of the Kaiser rule (i.e., all factors retained returned eigenvalues of greater than 1.00; Kaiser 1960). All factors presented adequate Cronbach alpha coefficients (see Table 1).
Notwithstanding the relative mortality of the original scale items in our analysis, we would note that our final solution closely resembles the original solution proposed by Massey, except for the Baversion towards gay men^and Baversion towards lesbians^factors that now appear collapsed into a sole factor (factor 6; see Table 1). Item mortality mainly stemmed from item cross-loadings in different factors alongside items that were not significantly loaded on any specific factor. However, and compared to the original scale estimates, the items retained in the present solution represent the core of the construct underlying each factor and constituting more than 50 % of the original items.
Following these PAF results, we ran a CFA testing of the six-factor structure of the 39 PPS items through recourse to sub-sample 2 and Mplus 7.0 software (Muthén and Muthén 2012). We tested different PPS structure models and obtained their respective fit indexes: a correlated model (our hypothesized model), a second-order model with a general PP dimension, and an uncorrelated first-order model. Table 2 presents the results of these analytical procedures that we then describe below.
We correspondingly applied the necessary constraints for models' identification and specification, i.e., one indicator path loading for the latent factor was set to 1 with all measurement errors set to 1. By the same token, in the model comprising a second-order factor, the unique variances associated to first-order factors were constrained to 1 (cf. Byrne 2012). We obtained both the relative and absolute goodness of fit indexes of the models: the chi-square fit index (χ 2 ), the relative chi-square fit index (χ 2 /df), the Tucker-Lewis index (TLI; Tucker and Lewis 1973), the comparative fit index (CFI ;Bentler 1990), and the root mean square error of approximation (RMSEA; Browne and Cudeck 1993). The analysis results are presented in Table 2.
Based on the standards established in the literature for fit indexes (i.e., CFI and TLI indices greater than 0.90-0.95, RMSEA lower than 0.08-0.05, SRMR lower than 0.10-0.08 ;Bentler 1990;Browne and Cudeck 1989;Hu and Bentler 1999;Jöreskog and Sörbom 1984;Steiger 1990), and as expected, the model proposing a correlated six-factor structure of PP proved the best model (see Fig. 1). Indeed, an examination of both the absolute and relative fit indexes shows that the model kept within the set standards (e.g., Bentler 1990).
The remaining two models proved of inferior quality as ascertained by the absolute and relative goodness of fit indexes set out in Table 2. We do note that in the second-order model, the correlation between the Bpositive beliefs^factor and the PP second-order factor was non-significant, thus rendering the model theoretically inadequate (ϕ = −0.06, p = 0.28).

Expressions of Polymorphous Prejudice
Now adopting the whole sample (N = 1220), we checked for differences in participant mean scores across each PPS sub-scale. The ANOVA results revealed how participants globally evaluated each dimension differently, F(5,6095) = 2339.41; p < 0.001, η p 2 = 0.66. In order to analyze how the mean participant score for each factor got positioned in reference to the underlying factor rating scale, we performed one-sample t tests against the mid-  In this sense, the results from one-sample t tests conveyed how participants revealed more Bsubtle^PP regarding LG individuals, returning higher scores for positive beliefs, resist heteronormativity, and denial of continued discrimination. This means that participants reporting more negative beliefs towards LG individuals were more heteronormative, and to a greater extent, denied LG individuals are still discriminated against. On the contrary, low levels of Bblatant^PP were observed as implied by low participant scores on the traditional heterosexism and aversion towards gay men/lesbians factors. This means that participants generally expressed less traditional forms of prejudice against LG individuals. However, one exception should be made as regards the Value Gay Progress sub-scale. Indeed, in this subtler dimension of PP, participants scored lower and thus revealed more prodiversity beliefs.

Predictors of General Polymorphous Prejudice Towards LG individuals
The data analysis strategy applied in the present article aimed at investigating the contribution of different groups of variables towards predicting both general PP against LG individuals and its sub-dimensions. More specifically, we sought to analyze the importance of psychological-level variables (i.e., LG discrimination recognition and LG discrimination deservedness) in predicting PP while controlling for the remaining groups of variables. As these variables were entered hierarchically in the regression successive models, their contribution to the increment of the explained variance of this criterion variable represented a strong test of their predictive power. Table 3 presents a correlation matrix of the PPS total and sub-dimension scores alongside the predictor variables.
As referred above, the groups of variables described previously were entered hierarchically into different regression steps. Thus, in a first step, demographic variables were entered and their predicting value tested against PP. In the second step, ideological variables were added, while the third step added psychological variables to the other two groups of variables. Variance inflation factors were calculated for every step of the regression models to account for possible multicollinearity between predictors (O'Brien 2007). A summary of the hierarchical regression for the total PP score features in Table 4. 1 The results presented in Table 4 generally convey how the introduction of psychological-level predictors, i.e., discrimination recognition against LG individuals and LG discrimination deservingness, returns the highest increase in the hierarchical model explained variance. In this sense, the more participants believed that LG individuals deserved to be discriminated against, the more they tended to score higher on PP. Furthermore, the lower the level of participant recognition that LG individuals were actually discriminated against, the more they tended to express PP.
Nevertheless, and looking more closely to the hierarchical regression results presented in Table 4, we found that PP scores rose in the case of male participants, in participants who had fewer LG friends, who positioned themselves as right-wing, and who were Catholics. The results also showed that age and education did not constitute significant predictors of participants' scores on PP.
Switching now to a more detailed analysis, specifically at the level of the six sub-scales underlying the PP construct, the different hierarchical regressions ran on each of these sub-dimensions portrayed a clearer picture regarding the predictors of this type of prejudice. In fact, Table 5 displays In the second-order model, the standardized regression weight of the path of the second-order latent factor to the first-order factor positive beliefs is non-significant (ϕ = −0.06, p = 0.28) how more blatant forms of PP tended to share the same sets of predictors and similar patterns of increase in the explained variance of the criterion variable. In this sense, the prediction of the sub-scale scores of Btraditional heterosexism^and Baversion towards gay men/ lesbians^benefited from the inclusion of the psychological variables. Indeed, the explained variance of the regression models reached a maximum value when these factors were added to the equations. In this sense, participants that believed LG individuals deserve to be discriminated against and that did not recognize that they are exhibited higher scores in these sub-scales.
Moreover, participants scores in these sub-scales were also impacted on by other predictors both ideological and sociodemographic in nature. In this sense, male participants, with less LG friends, right-wing, and Catholics did exhibit more traditional heterosexism and aversion towards gay men/lesbians. We would note that, regarding this last sub-  I1  I13  I11  I12  I10  I9  I8  I7  I6  I5  I4  I3  I2 .   LG discrimination recognition −0.31*** −0.18*** −0.09** −0.23*** 0.13*** −0.28*** −0.21*** PP polymorphous prejudice, TH traditional heterosexism, PB positive beliefs, RH resist heteronormativity, DCP denial of continued discrimination, VGP value gay progress, AVGM/L aversion towards gay men/lesbians *p < 0.05, **p < 0.001, ***p < 0.000 scale, participant level of education also predicted their scores for aversion. Consequently, participants with lower levels of education exhibited more aversion than participants with higher levels of education. Turning now to more subtle forms of prejudice against LG individuals, we may observe in Table 5 that Bpositive beliefs,B resist heteronormativity,^and Bdenial of continued discrimination^sub-scales shared similar predictors and similar increases in explained variance. However, these more subtle dimensions of PP did not benefit much from the introduction of the ideological or psychological predictors, over and above the sociodemographic ones.
Regarding the positive beliefs sub-scale, we found that younger participants, with higher levels of education, non-Catholics, and that did not recognize that LG individuals are discriminated against were those exhibiting higher levels of denial of expression of positive beliefs. In the same vein, participants that resisted less to heteronormativity were those expressing having less LG friends, that positioned themselves politically as more conservative (right-wingers), that were Catholics, and that did not recognize LG individuals are discriminated against. Finally, participants with higher levels of education, right-wing, Catholics, and that did not recognize that LGs are discriminated against were among those that mostly denied the continued discrimination of LG individuals.
As we previously noted, the Bvalue gay progress^sub-scale appeared once again as a special case in our analyses. Indeed, being a more subtle form of expressing prejudice against LG individuals, the predictors of this particular sub-scale, as well as increasing the explained variance obtained by the different steps in the hierarchical regression, resembled the pattern of results obtained for the more Bopen^or Bblatant^prejudice sub-scales. The prediction of scores in this sub-scale benefited significantly from the introduction of the psychological predictors. In fact, the more participants tended to think that LG individuals deserve to be discriminated against and the more they did not recognize that they are, the higher the devaluation of gay progress. Moreover, other predictors helped in explaining the scores in this particular sub-scale: participants with less LG friends, right-wing, and Catholics were those that most devalued this particular PPS dimension.

Discussion
In this article, we presented the social and psychological determinants of PP against Portuguese LG individuals. We also presented the construct validation of Massey's (2009) PPS in accordance with a sample of Portuguese heterosexuals. In general terms, the results obtained not only added empirical evidence to the existing literature on the predictors of sexual prejudice against LG individuals but also shed light on the explanation and comprehension of this phenomenon.
As we pointed out in the introduction, research on sexual prejudice towards LG individuals has evolved from studies using unidimensional measurement instruments to multidimensional approaches. In fact, these changes reflect a zeitgeist of changes ongoing in Western societies concerning LG individuals (Seidman 2002), especially in terms of public visibility and in legal changes. Indeed, research has also been pointing to multiple entanglements between attitudes towards homosexuals and other value systems, such as pro-gay equality and gender normalization (Martinez 2011). All these aspects called for a clear change in the way sexual prejudice attitudes towards LG individuals are measured. Massey (2009) PPS provided one of the efforts at the measurement level designed to surpass this methodological vacuum and ensure an adequate and up-to-date measurement instrument for these types of sexual prejudice attitudes.
As such, the adaptation and validation of this scale to the Portuguese context, as well as the analysis of the PP predictors, seems a highly relevant objective given the recent significant changes enacted for the panorama of equal rights for the LG community in Portugal (Oliveira et al. 2013), such as the introduction of same-sex marriage law (Oliveira et al. 2014).

Predictors
Step 1 a Step 2 b Step 3  LG lesbian/gay *p < 0.001 a Collinearity statistics, as represented by the variance inflation factor (VIF), revealed the absence of collinearity between predictors (VIFs ranging from 1.02 to 1.06) b VIFs ranging from 1.03 to 1.11, revealing the absence of collinearity between predictors c VIFs ranging from 1.04 to 1.14, revealing the absence of collinearity between predictors Step 1 Step 2 Step 3 Step 1 Step 2 Step 3 Step 1 Step 2 Step 3 Step 1 Step 2 Step 3 Step 1 Step 2 Step 3 Step 1 Step 2 Step 3 Sociodemographic: LG lesbian/gay ***p < 0.001, **p < 0.01, *p < 0.05 a VIFs ranging from 1.018 to 1.139, revealing absence of multicollinearity between predictors

Age
The results presented in this article stem from two types of analysis. We started out by adapting Massey's (2009) PPS and determining its construct validity for a sample of Portuguese heterosexuals. PAF analyses presented a sixfactor structure similar to that obtained in Massey's original study. However, some of the original items were lost (less than 50 %), and the original dimensions of aversion towards lesbians and gay men were collapsed into a sole factor in our analysis. Nonetheless, the original structure of Massey's (2009) PPS was maintained and the core meaning of the dimensions and of the polymorphous prejudice construct both remained intact. In the end, our adapted version of the PPS resulted in a shorter scale that allows researchers to measure polymorphous prejudice in a more parsimonious approach.
This six-factor solution was further tested by means of a CFA. From the three models tested, the correlated six-factor structure proved the model with the highest fit. This means that although the PPS is a multidimensional scale, its subscales are relatively correlated and share the variance of a common underlying construct, namely prejudice towards LG individuals.
After this, we continued with our analysis by identifying the predictors of PP towards LG individuals. In this sense, the results from the hierarchical regressions deployed revealed, in a general way, that male participants, with less LG friends, that positioned themselves as right-wing, and that were Catholic, were those exhibiting higher PP. These results also back up other previously published empirical evidence (Herek 1993;Herek and Capitanio 1996;Herek and Gonzalez-Rivera 2006;Yang 1998).
However, these regression models also conveyed how the introduction of psychological predictors (e.g., LG discrimination deservingness) contributed more to the explained variance of PP above and beyond the remaining predictors added in these regression models. In this sense, the more participants thought that LG people deserve to be discriminated against, the more they tended to express PP. Inversely, the less they recognized that LGs are actually discriminated against, the more they tended to express this type of prejudice. These results are new and signal the importance of discrimination deservingness and discrimination recognition in the explanation of polymorphous prejudice against LG individuals. Indeed, the importance of the psychological-level predictors has been outlined in numerous studies on prejudice (e.g., racism; Vala et al. 1999) but less often demonstrated, as far as we know, in the domain of sexual prejudice.
Also of added interest were the results from the different hierarchical regressions deployed for the PPS sub-scales. Indeed, more blatant prejudice sub-scales (i.e., traditional heterosexism and aversion towards gay men/lesbians) shared the same set of predictors with scores in these sub-scales mostly shaped by psychological-level factors. Notwithstanding, these blatant sub-scales were also predicted by sociodemographic and ideological factors.
A completely different picture was obtained when deploying the regression models predicting more subtle subscales (i.e., positive beliefs, resist heteronormativity, and denial of continued discrimination). In this case, the ideological or psychological-level factors did not add much to the explained variance of these sub-scales above and beyond that obtained by the sociodemographic set of predictors.
These two patterns of results are extremely interesting as the more blatant sub-scales are associated to a more complex set of variables, while the more subtle subscales are mainly associated to sociodemographic predictors. In reality, studies with other forms of prejudice tend to show a reversed pattern (e.g., Vala et al. 1999), where blatant scales are mainly associated with sociodemographic variables, and subtle scales to ideological and psychological factors. The nature of the subtle sub-scales of the PPS might help us understand this reversed pattern. On the one hand, the expression of more positive beliefs towards LGs, and the recognition that they are discriminated against and that society continually imposes an heterosexual normative context might be subjects that are only cognitively available to participants depending on age, educational level, friendship with LGs, and even their political or religious positioning. On the other hand, more blatant forms of polymorphous prejudice might be mainly anchored on participant perceptions of justice and equity between social groups (deservingness) and the lack of recognition that LGs are the targets of prejudice or stigmatization.
The only exception to this pattern of results was the value gay progress sub-scale. Although a more subtle sub-scale in its nature, the set of predictors revealed by the regression models coincided with those obtained by the more blatant sub-scales. Indeed, respondents that tended to think that LG people deserve discrimination, and that did not recognize that LGs are actually discriminated against, were among those that most devalued gay progress. Indeed, the wordings of some of the items of this particular sub-scale serve the purposes of both blatant and subtle prejudice, i.e., raising sensitive issues for both types of prejudiced individuals. In this sense, it is interesting to note that the score on this particular dimension was significantly below the mid-point of the scale (indicating less support for gay progress) hand in hand with the blatant dimensions of the PPS.
These results shed new empirical evidence and stress the importance of studying the phenomenon of LG prejudice in multidimensional approaches as the results obtained from the total PP score do not enable us to obtain an accurate and overall picture. Indeed, general PP comes qualified by a set of predictors that do not differentiate it from other forms of prejudice, such as prejudice against black individuals (Vala et al. 1999). However, a totally different picture is obtained when analyzing PPS sub-scales. In terms of their predictors, more blatant sub-scales verge towards the pattern of results obtained from the total PPS score. But regarding more subtle sub-scales, we found that the sociodemographic predictors helped us in accounting for most of the construct that underlies each of these subscales. Briefly, these results supported the contemporary argumentation regarding the need for more complex and multidimensional measures of sexual prejudice (Hegarty and Massey 2006), in particular, instruments that aid in encapsulating not only attitudes towards homosexuality but also values and other belief systems (Martinez 2011).
The results presented in this article also provide evidence helping with the development of practical measures to fight prejudice against LG individuals. Indeed, different types of actions might be taken to alert individuals of their prejudiced opinions regarding LG individuals depending on the more or less blatant or subtle nature of their PP. In this sense, anti-prejudice campaigns might target specific types of individuals with the knowledge acquired from the different predictors deriving from the present study. Additionally, professionals working in the field of sexual prejudice or stigma may correspondingly tailor specific information to different target individuals during consciousness raising campaigns.

Limitations
Our study is not without limitations. Firstly, some psychological-level predictors were not covered by our study, such as beliefs about homosexuality controllability, openness to change, biographical variables, and personality traits. Future studies might well incorporate the analysis of the role of these variables in predicting LG prejudice. Secondly, the results regarding the association between the recognition of discrimination (used in our study as a predictor) and the PPS sub-dimension denial of continued discrimination (used as a criterion) should be interpreted with caution as they might point to a similar construct. While it remains true that these measures do seem mutually complementary, the content of each one nevertheless departs in somehow different directions. While discrimination recognition asks participants a general and simple question (i.e., if they think lesbians/gays are discriminated against), denial of continued discrimination goes much further and taps participants' opinions regarding job opportunities that are missed out on to LG discrimination, equal treatment between LG and straights, etcetera. Additionally, at the results level, the zero-order correlations between these two measures were low with the predictive power of discrimination recognition over denial of continued discrimination not proving substantive. Briefly, these constructs merit further analysis by future studies with full metric analysis helping to disentangle these measures being highly advisable.
Thirdly, our data collection method (web-surveys) might have raised issues of participants' confidentiality or privacy. There are well-known limitations regarding websurvey data collection, especially in terms of confidentiality. However, in our questionnaire, participants were never asked to disclose personal or potentially identifiable information. Moreover, web-surveys also attain particular relevance for the collection of data on socially sensitive issues (Couper 2000) such as that addressed in this present article. Fourthly, there are limitations inherent to the sample studied as regards the high level of education of participants (with the majority holding university degrees) deemed to normally exhibit lower levels of prejudice when contrasted with the population in general. To overcome this possible bias, future studies should target participants with different levels of education.

Future Studies and Applications
Apart from the suggestions already outlined above, future lines of research should explore even further the predictors of PP, in particular, by adding more psychological-level predictors. For example, predictors such as perceived relative deprivation and emotions should be added to the regression models in order to analyze their utility in terms of the amount of variance obtained in explaining this phenomenon. Furthermore, the inclusion of macro-social and political scenarios in the general framework of theories of sexual prejudice, such as neo-liberalism, the respondent economic and geo-political situations might return findings useful to situating and contextualizing modern forms of prejudice against LG individuals. Moreover, analysis of polymorphous prejudice in other stigmatized populations, such as bisexuals, could provide us with a greater understanding of the sexual prejudice phenomenon and enable comparisons across groups with different types of stigma attached.
To sum up, the psychometric evidence presented in this article adds up to that already obtained concerning the psychometric qualities of the Massey (2009) PPS ensuring its appropriate utilization as a good instrument both in research and other applied fields. Moreover, the results obtained from this scale, as well as its predictors, add new evidence to the literature on sexual prejudice, specifically prejudice against LG individuals. More importantly, our results also help us construct more focalized intervention programs striving to reduce prejudice against LG individuals and, as such, of great relevance to those directly working in the field with this type of phenomena.

Compliance with Ethical Standards
Ethical Statement All the procedures performed in this article involving human participants took place in accordance both with the ethical standards of the institutional and/or the national research committee and with the 1964 Helsinki declaration and its later amendments or comparable ethical standards. Additionally, the present study was carried out in accordance with the Ethical Guidelines of the host institution.
Informed Consent All participants read an informed consent with the description and purpose of the study and were informed that by proceeding, they consented to participating but that they could withdraw at any stage of the study.

Conflict of Interest
The authors declare that they have no conflict of interest.