Introduction

Sexual prejudice refers to a negative attitude toward an individual because of her or his sexual orientation. Given the widely shared perception of heterosexuality as the “correct” way of intending sexual behavior, this type of prejudice is often specifically directed toward people who engage in homosexual behaviors, and thus toward gay men, lesbians, or bisexual individuals [1].

In the last decade, research on prejudice based on sexual orientation has taken advantage of the diffusion of implicit measures. These measures aim at capturing positive and negative features of the target object that might rely outside of conscious awareness, or that might not be accurately reported through introspection. For these reasons, implicit measures are more resistant than explicit measures to response biases, such as social desirability, impression management strategies and limited introspective access [2].

The Sexuality Implicit Association Test (Sexuality IAT) [3] has been the most widely used implicit measure of sexual attitude. It is a computerized two-choice discrimination task in which stimuli have to be categorized as belonging to the target categories Heterosexuals and Homosexuals, or to the attribute categories Good and Bad by pressing, as quickly and accurately as possible, one of two response keys. Categorizing the stimuli more quickly when Heterosexuals shares the response key with Good – and Homosexuals with Bad – than when it shares the response key with Bad – and Homosexuals with Good – is taken to indicate an implicit preference for heterosexuals relative to homosexuals.

Previous research using the Sexuality IAT found that individuals tend to prefer heterosexuals to homosexuals [3]–[6]. Looking at participants with different sexual orientation, a preference for one's own sexual orientation was observed in both heterosexual and gay individuals [7], [8], although such a preference was weaker for the latter [9] (see [10] for a review).

To date, however, there are no studies specifically focused on the investigation of the implicit sexual attitude of bisexual individuals. When present [4], [5], data concerning bisexuals were aggregated with those regarding gays. In this paper, we investigated the implicit sexual attitude of heterosexual, gay and bisexual individuals, keeping the three groups distinct.

As an additional novelty, we analyzed the contribution of specific associations between target groups and positive vs. negative attributes, using a methodology that allowed us to decompose the results of a Sexuality IAT. The IAT effect is the comparison between the response latencies in two mappings that differ in the target and attribute categories that share the same response key. There is a growing interest on analyzing and decomposing the IAT effect. For instance, models such as the QUAD model [11], the diffusion model [12], and the discrimination-association model [13] have been proposed that disentangle multiple, qualitatively different processes underlying the responses to the IAT. In the present paper, we disentangled the contribution of specific associations to the overall measure by investigating how the speed of categorization of individual stimuli changed according to the associative condition they were presented in. It is worth noting that the proposed procedure is different from sorting the trials into subsets and computing separate IAT effects for the two targets. Such a procedure is not advisable in analyzing IAT data [14]. Instead, we assessed the differential contribution of individual stimuli to the overall IAT effect, which remains a measure of a target relative to the other. The analysis was conducted through a Many-Facet Rasch Measurement (MFRM) [15]. Being a Rasch model [16], the MFRM provides us with a rigorous measurement system [17]. The speed of categorization of the stimuli is expressed by interval measures characterized by a measurement unit that, if the data fit the model, maintains the same size over the entire continuum. It follows that the measurement and comparison of different elements is more precise. Moreover, the MFRM allows us to investigate whether the speed of categorization of individual stimuli differs in the two associative conditions.

The proposed procedure has a strong potential for understanding the meaning of the implicit measure itself. For instance, consider the implicit preference for white people relative to black people that is usually observed in the literature [3], [18]. If the stimuli that contribute most to the measure are the positive ones, rather than the negative ones, then the implicit preference should be interpreted more as an expression of pro-Whites bias, rather than anti-Blacks bias. If the opposite occurs, the same implicit preference should be interpreted more as an expression of anti-Blacks bias, rather than pro-Whites bias. Taking into account the contribution of different stimuli provides researchers with a detailed depiction of the implicit associations that mostly underlie the measure (and that might differ across individuals), and allows them to precisely define the construct under investigation. Moreover, it prevents researchers from misreading the real meaning of the implicit measure. An example of such a misunderstanding has been described in the analysis of racial attitude. By means of the proposed procedure, it was found that the implicit preference for white people displayed by white individuals especially resulted from the attribution of positive traits to Whites, rather than of negative traits to Blacks [19]. Therefore, it was argued that the implicit measure of racial attitude might not necessarily imply black derogation, but could be mostly related to white favoritism. Independent contributions reached similar conclusions with different strategies of analysis (see, e.g., [20], [21]).

Disentangling the contribution of individual stimuli could be particularly useful when sexual prejudice is measured across different sexual orientations. For instance, it might help to answer questions concerning the direction of the preference observed in heterosexuals (e.g., whether it is most related to a preference for heterosexuals or to a derogation of homosexuals), the possible different meaning of the preference for one's own sexual orientation usually showed by heterosexuals and gays, and the reasons underlying implicit preferences showed by bisexual individuals.

Methods

Results

Both infit and outfit were excellent for the facets sexual orientation (0.97≤infit/outfit≤1.28), attribute stimuli (0.83≤infit/outfit≤1.13), and associative condition (0.95≤infit/outfit≤1.06). Only 14 out of the 789 respondents (1.78%) had an infit or outfit outside the recommended range.

Respondents completed the IAT with different speeds (α ranges from −3.22 to 2.99; R = 0.87; χ²(788) = 4857.1, p<0.001). The stimuli were categorized with different speeds (γ ranges from −0.45 to 0.30; R = 0.97; χ²(14) = 423.9, p<0.001). The condition Heterosexuals-Good/Homosexuals-Bad (HE-G/HO-B) was easier than the condition Homosexuals-Good/Heterosexuals-Bad (HO-G/HE-B; ε_HE-G/HO-B = 0.47; ε_HO-G/HE-B = −0.47; R = 1; χ²(1) = 2494.9, p<0.001). The distance Δ = 0.94 between the two conditions represents the size of the IAT effect. It is significantly different from 0 (z = 66.47, p<0.001), meaning that, taken all together, respondents implicitly preferred heterosexuals to homosexuals.

Analyzing the IAT effect computed on respondents of different sexual orientation, we found that both heterosexual and bisexual respondents significantly preferred heterosexuals to homosexuals (Δ_heterosex = 1.08, z_heterosex = 76.37, p<0.001; Δ_bisex = 0.29, z_bisex = 4.10, p<0.001), but heterosexual respondents held a stronger preference than bisexual respondents (z_{heterosex-bisex} = 11.07, p<0.001). Gay respondents significantly preferred homosexuals to heterosexuals (Δ_gay = −0.20, z_gay = −2.83, p<0.01). To compare the strength of the preference observed in heterosexual and gay respondents, we compared the size of their IAT effects in absolute terms: heterosexual respondents held a stronger preference for their own sexual orientation than gay respondents (z_{|heterosex|-|gay|} = 12.21, p<0.001). For the sake of comparison with classical analysis, we also computed the traditional D₂ score [28], obtaining the same pattern of results: mean D_heterosex = 0.52, SD_heterosex = 0.41; mean D_bisex = 0.16, SD_bisex = 0.48; mean D_gay = −0.16, SD_gay = 0.39.

Table 1 provides information concerning the differential stimulus functioning, separately for the three groups of respondents. For each individual stimulus, it is shown whether its overall speed of categorization (i.e., estimated across the two associative conditions) changes according to the specific associative condition the stimulus is presented in. This allowed us to investigate the contribution of each individual stimulus to the overall implicit measure.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Speed of categorization of the stimuli in the two associative conditions for respondents of each sexual orientation.

https://doi.org/10.1371/journal.pone.0078990.t001

The speed with which heterosexual respondents categorized the positive stimuli pleasure, marvelous and wonderful increased in the condition Heterosexuals-Good/Homosexuals-Bad and decreased in the condition Homosexuals-Good/Heterosexuals-Bad, compared with the overall speed (i.e., across the two associative conditions) of these stimuli. Thus, heterosexual respondents associated these stimuli more easily with heterosexuals than with homosexuals. Differently, the speed with which heterosexual respondents categorized the negative stimuli tragic, nasty and painful increased in the condition Homosexuals-Good/Heterosexuals-Bad and decreased in the condition Heterosexuals-Good/Homosexuals-Bad. These were the stimuli that heterosexual respondents associated more similarly with heterosexuals and homosexuals. Therefore, pleasure, marvelous and wonderful were the stimuli that most contributed to increasing the implicit preference for heterosexuals observed in heterosexual respondents, whereas tragic, nasty and painful were the stimuli that most contributed to decreasing it. In gay respondents, only the stimulus glorious exhibited differential stimulus functioning. The speed of categorization of this stimulus increased in the condition Heterosexuals-Good/Homosexuals-Bad and decreased in the condition Homosexuals-Good/Heterosexuals-Bad. Gay respondents associated glorious more easily with heterosexuals than with homosexuals. With this single exception, the pattern of the different stimuli was similar and, in particular, there was no specific stimulus that mainly contributed to increasing the implicit preference for homosexuals shown by gay respondents.

Concerning bisexual respondents, the speed with which they categorized the negative stimuli humiliate and horrible increased in the condition Heterosexuals-Good/Homosexuals-Bad and decreased in the condition Homosexuals-Good/Heterosexuals-Bad: Bisexual respondents associated these stimuli more easily with homosexuals than with heterosexuals. Thus, humiliate and horrible were the stimuli that contributed most to increasing the implicit preference for heterosexuals relative to homosexuals observed in the bisexual respondents.

Discussion

This article investigated the implicit sexual attitude of heterosexual, gay and bisexual individuals, analyzing the attributes that mostly drive implicit preferences. Consistently with the literature [7], [8], we observed that both heterosexual and gay respondents preferred individuals with their own sexual orientation, and that this preference was stronger for the former than for the latter. In addition, we found that bisexual respondents implicitly preferred heterosexuals to homosexuals. Looking at the contribution of individual stimuli, we found that the strong preference for their own sexual orientation observed in heterosexual participants was mostly driven by the attribution of positive traits to heterosexuals, rather than of negative traits to homosexuals. Differently, the weaker preference for homosexuals observed in gay respondents was not particularly guided by either positive or negative attributes. Finally, the preference for heterosexuals displayed by bisexual participants especially resulted from the attribution of negative traits to homosexuals, rather than of positive attributes to heterosexuals.

These results can be interpreted taking into account, at the same time, intergroup relations and shared social norms. The bias in sexual attitudes could be conceived as an intergroup phenomenon, in which people tend to evaluate their membership group more positively than other groups, with the aim of obtaining a positive social identity and, thus, a positive self-image [29]. This asymmetry is generally based on a particularly positive evaluation of the ingroup, while a negative evaluation of the outgroup is likely to occur only when this group is perceived as threatening [30], [31]. Concerning social norms, the widely shared beliefs regarding sexual orientations are claimed to lead to the perception of heteronormativity, defined as an internalized view according to which heterosexuality is the standard for legitimate and expected social and sexual relations [32], [33].

For heterosexuals, these two social processes point at the same direction: the self-image is protected and enhanced by a positive image of heterosexuality compared to homosexuality, and the shared social view based on heteronormativity is consistent with this perception. This could explain why heterosexual respondents showed a strong implicit preference for heterosexuality. Moreover, given that the perception of heterosexuality as natural and normal is considered stable, justified and not threatened [34], heterosexuals did not need to derogate homosexuals to obtain a positive image of their own sexual orientation.

For gay perceivers, however, the two processes may conflict. On the one hand, the homosexual ingroup could be preferred, as a way to increase self-esteem. On the other hand, shared beliefs related to heteronormativity could lead to biases in favor of heterosexuality: in this case, phenomena of self-stigma are possible, as individuals may internalize a negative view of their own group, and thus of themselves [35]. In our data, the need for a positive self-esteem could have led to ingroup preference, while self-stigma and heteronormativity perceptions could have weakened such a preference. The presence of opposite motives may explain why the global implicit preference was not particularly driven by individual attributes, either positive or negative.

Finally, for bisexual respondents, heteronormativity should not be counterbalanced by ingroup positivity, as in this case the very definition of the ingroup is unclear [36]. On the other hand, the presence of internalized self-stigma related to homosexual behaviors is likely [37]. This combination of motives might explain the implicit preference for heterosexuals, and the fact that it is driven by attributing the negative stimuli humiliate and horrible to homosexuals.

Although our findings are clear, some limitations about the present study should be noted. First, the physical absence of the experimenter in an online study increases the difficulty of ensuring that procedural instructions are clear to all respondents and that the task is performed in the proper way. However, it is reassuring that several online studies, employing different IATs, replicated the effects found in laboratory settings [3], [22]. Second, because of the method of data collection, respondents to the Sexuality IAT were self-selected and are not representative of any particular population. Second, the number of gay and bisexual respondents was not very large, in particular compared to the heterosexual sample. As a consequence, the obtained results should be replicated in further research.

Overall, our findings demonstrate the usefulness of considering, besides the usual global IAT scores, the specific contributions of positive and negative stimuli to the expression of implicit associations. The MFRM has proved to be a valid tool for this purpose. The model represents a rigorous frame of reference in which estimating and comparing the speed of categorization of the stimuli. Moreover, by allowing the analysis of differential stimulus functioning, the model provides us with the contribution of each stimulus to the overall IAT measure. Such an analysis is advisable when distinct and opposing drives are involved, as it can be the case of people with different sexual orientations. Disentangling the contribution of these drives could have important consequences for a clearer understanding of sexual prejudice and discrimination, with the aim of developing informed social policies for the reduction of these antisocial phenomena. Such a result is particularly desirable if we consider that gay and bisexual individuals, being targets of prejudice and discrimination, often face minority stress phenomena, with detrimental consequences on well-being, quality of life, and even mental health [38].

Acknowledgments

The authors thank the principal investigators of Project Implicit for their permission to report data.