Sexual Orientation and Race Intersectionally Reduce the Perceived Gendered Nature of Normative Stereotypes in the United States

Cognitive Biases in Gender Normative Stereotypes

Below we review research on the prevalence of gender-, sexuality-, and race-based cognitive biases. More specifically, cognitive biases such as androcentrism (i.e., the assumption that an unspecified person is a man), heterocentrism (i.e., the assumption that an unspecified person is straight), and Eurocentrism (i.e., the assumption that an unspecified person in the [U.S.] is White) often influence the representations generated when calling to mind a “person,” “woman,” or “man” (Alt et al., 2020^; Bailey et al., 2019^, 2020^; Devos & Banaji, 2005^; Ghavami & Peplau, 2013^; Hamilton, 1991^; Lick & Johnson, 2016). As an example, words that purportedly activate representations that are inclusive of both men and women (e.g., chairmen) implicitly activate only masculine concepts (Banaji & Hardin, 1996). Relatedly, people show a heterocentric bias in perception, assuming heterosexuality even when the base rates suggest otherwise (Lick & Johnson, 2016). Finally, perceptions of people as White permeate many domains of cognitive representations, including leadership (Gündemir et al., 2014^; Rosette et al., 2008) and the color of God's skin (Roberts et al., 2020).

These biases also influence stereotypes of groups. For example, descriptive stereotypes of racial and sexual minorities do not always conform to the descriptive stereotypes of the relevant majority group, leading to unique descriptive stereotypes for minorities (Hall et al., 2019). As an example, using a free-response paradigm, Ghavami and Peplau (2013) showed that descriptive stereotypes of racial groups exhibited androcentric biases, such that the descriptive stereotypes spontaneously generated for broad racial categories (e.g., Black people or White people) most closely matched the descriptive stereotypes generated for men in that racial category (e.g., Black and White men) than the women (e.g., Black and White women). This pattern was mirrored for Eurocentric biases, as the descriptive stereotypes generated for the category of “men” and “women” most closely matched those generated for White men and women.

Descriptive stereotypes of lesbian women and gay men also do not conform to the descriptive stereotypes of their straight counterparts. The literature suggests that most people hold assumptions of gender inversion or the assumption that gay men and straight women are similar in traits, while lesbian women and straight men are similar because the target of their sexual attraction is the same (Blashill & Powlishta, 2009a^, 2009b^, 2012^; Kite & Deaux, 1987^; Lehavot & Lambert, 2007). Thus, gay men are seen as possessing feminine and lesbian women are seen as possessing masculine characteristics and preferences, suggesting that traditional gender descriptive stereotypes might be limited to heterosexual men and women.

Finally, biases influence stereotypes and perceptions in an intersectional way, as certain sub-groups are foregrounded to a greater extent in the social milieu (of the U.S.) than others along multiple dimensions (Hall et al., 2019). For example, Black women do not activate the category “women” as strongly as White women nor the category “Black” as strongly as Black men (Carpinella et al., 2015^; Goff et al., 2008^; Johnson et al., 2012) in various categorization tasks. Follow-up research has corroborated the basic finding that “Black” is seen as masculine in the U.S. society (Johnson et al., 2012^; Thomas et al., 2014) and extended it to include U.S. Asian-Americans (Schug et al., 2015^, 2017) who are presumed feminine, suggesting that gender can influence perceptions of race beyond androcentrism.

Normative Stereotypes at the Intersections

In this paper, we extend past work by asking whether normative stereotypes are also influenced by androcentric, heterocentric, and Eurocentric biases. As stated earlier, prescriptive and proscriptive stereotypes are uniquely associated with backlash discrimination when people violate the normative expectations placed on them. Although there is often an equivalence between descriptive and prescriptive expectations—how we anticipate people to be is also how we believe they should be (Prentice & Carranza, 2002)—almost all of the work done to date on intersectional stereotypes has been on descriptive stereotypes.

There is reason to believe that in contrast to majoritized groups, normative stereotypes for men and women at the intersection of minoritized identities will not reflect their descriptive counterparts. For example, gay men and lesbian women can face discrimination precisely because they violate expectations placed upon their gender (Lehavot & Lambert, 2007). If true, that suggests that a gay man who acts femininely will be in line with descriptive expectations based on his sexual orientation but violate prescriptive expectations of masculinity based on his gender. Thus, femininity for gay men would be proscribed. However, a feminine gay man might be acting exactly as expected because he is gay, keeping descriptive expectations in line with normative ones. From this perspective, femininity for gay men would be prescribed and not punished. It is currently unclear whether femininity is prescribed or proscribed in gay men, underlying the main point that it is necessary to directly assess normative stereotypes when both gender and sexual orientation are foregrounded to address these competing deductive conclusions.

Similarly, while Black men descriptively are seen as hyper-masculine (Galinsky et al., 2013), the backlash they face based on that masculinity (Plant et al., 2011) suggests that normatively in some contexts, they should not be masculine at all. The descriptive stereotype of masculinity might not be prescribed for Black men but instead proscribed, although masculinity is prescribed for the broader “man” category (Prentice & Carranza, 2002). Indeed, Black men who are perceived as less masculinely, by removing their beards (Livingston & Pearce, 2009), acting communally (Livingston et al., 2012), or even being gay (Pedulla, 2014) can either face less discrimination than Black men who are perceived more masculinely, or qualitatively different forms of discrimination that vary by kind rather than degree. As another example, Asian men are stereotyped as effeminate, small, and feminine (Ghavami & Peplau, 2013) but often react in hegemonically masculine ways to counteract that stereotype (Lu & Wong, 2013). Such behaviors are in line with a normative expectation that while their group is stereotyped as feminine, they are not supposed to act that way.

These examples highlight the importance of directly investigating normative stereotypes for intersectional targets, as descriptive and normative stereotypes might not be congruent for minoritized groups. Further, as minoritized identities are not always foregrounded during perception, assessing how normative pressures change when various aspects of intersectional identity are made salient is critical. Finally, backlash can occur not only when one violates normative expectations that get activated in a given situation (i.e., a woman running for political office versus applying to be an elementary school teacher) but also when one violates expectations in comparison to other groups. Thus, knowing the relative-normative stereotypes for intersectional groups is necessary to predict when, for example, a Black man might be punished for being masculine. A Black man might be punished for being masculine in general or perhaps only when directly compared to a less masculine White man. Indeed, White men's understanding of their masculinity can be directly tied to Black men's masculinity, as when a White man's masculinity is threatened, shooting and killing a Black man can increase it (Richardson & Goff, 2014).

From a practical application standpoint, understanding the unique normative pressures of individuals at the intersection of identity might help elucidate the boundaries of recommendations to mitigate unequal gendered treatment. For example, recommendations for women to “lean in” and assert themselves (Sandberg, 2013) have been criticized for being tone-deaf to women of color (Harts, 2020) and a potentially hazardous “one size fits all” strategy. Work on descriptive stereotypes suggests that the assumption that women are not agentic is nuanced by race, as only White women are seen as non-agentic from both a competence and a dominance standpoint (Rosette et al., 2016). In contrast, Black women are seen to possess traits in line with agentic dominance but not agentic competence, implying that “leaning in” might lead to different repercussions for Black women compared to White women. However, to fully understand those repercussions, normative expectations are also needed. Extending this example, if Black women are normatively perceived to be assertive and dominant when both race and gender are salient, that would suggest that in these circumstances, Black women could be rewarded rather than punished for dominant behavior (Livingston et al., 2012). Considering both descriptive and normative expectations in tandem can give a starting place to help contextualize when “leaning in” would be beneficial or detrimental to women of color.

The Current Investigation

In light of the aforementioned cognitive biases, we should not assume that gendered normative stereotypes are equally normative for men and women of other sexual orientations and races beyond straight and White. Given normative stereotypes’ role in engendering backlash, it is important to understand how people, and society at large, desire men and women of different social identities to behave (Hall et al., 2015b). This paper adds to the discussion of gender dynamics by providing an unfiltered view of the landscape of gender normative stereotypes for intersectional targets by sexual orientation (Study 1) and race (Study 2). We do so by soliciting “average American” perceptions of the normativity of 70 stereotypes for straight and sexually minoritized targets in Study 1 and for Asian, Black, Latino, Middle-Eastern, and White targets in Study 2.

This design allows us to not only directly compare specific normative stereotypes across a range of groups, but to also be in conversation with previous work that assessed stereotypes on the “average American” level (see Fiske et al., 2002^; Prentice & Carranza, 2002). It is important to note, however, that assessing stereotypes from the point of view of the “average American” is still limiting because of the prototypicality biases mentioned earlier. As “average” U.S. Americans are likely seen to be White, cis-gender men, we are assessing perceptions of intersectional groups that have been colored by this influence. That being said, we hope this paper will serve as a useful reference for future work on intersectionality as well as a place to generate hypotheses that extend and nuance existing research on gender.

Participants

We used a rule-of-thumb strategy to recruit roughly 100 participants per condition (Ghavami & Peplau, 2013). We recruited participants from the Harvard Digital Lab for the Social Sciences (DLABSS; www.dlabss.harvard.edu; Strange et al., 2019), a growing pool of over 16,000 people who agree to participate in research on a volunteer basis. Volunteers complete a master survey and complete as many studies available on the website as they wish. Participants in the pool are primarily based in the U.S. with the northeast region overrepresented. We only recruited participants who lived in the U.S. We did not include direct attention checks to assess the quality of the data, which is a limitation. DLABSS participants volunteered their time and were not paid for their participation. Of the 560 recruited participants, 309 finished the entire study with an average completion time of 24 min. The only exclusion criterion we employed was the failure to finish the study. Participants in Study 1 predominantly self-identified as White (n = 244), with the rest of the population identifying as Black (n = 12), Asian (n = 9), American Indian (n = 2), and “Other” (n = 14). Twenty-eight people did not indicate their racial group. Participants also predominantly self-identified as heterosexual (n = 256) while 19 identified as gay or lesbian, 15 as bisexual, four as pansexual, four as asexual, three as queer, two as questioning, three as “other”, and three preferring not to answer. Finally, 129 participants self-identified as male, 177 as female, two as “other” and one preferring not to answer (In both studies, we asked participants for their gender but mistakenly included labels for sex; for further breakdown of participant demographics, see the online Supplemental materials Tables S1 and S2).

Materials and Procedure

We asked participants to indicate the desirability of three targets displaying a series of 70 traits such as being “Assertive,” “Warm and kind,” and “Honest” (see online Supplemental materials for the full list of traits). These traits were amalgamated from several sources to encompass traits that are historically considered masculine, feminine, and gender-neutral (Bem, 1974^; Hall et al., 2015b^; Prentice & Carranza, 2002), as well as traits that are often used to describe sexually minoritized groups (e.g., “Theatrical” for gay people) or racially minoritized groups (e.g., “Musical” for U.S. Black Americans). Participants first familiarized themselves with the traits before moving to the three trait rating tasks.

All participants completed three rating tasks—one for a man, one for a woman, and one for a person—for a total of 210 traits rated on desirability. What differed between participants was the sexual orientation of the man, woman, and person targets they rated. We randomly assigned participants to rate trait desirability for straight targets (i.e., a straight man, straight woman, and straight person), gay targets (i.e., a gay man, lesbian woman, and homosexual person), or label-less targets (i.e., a man, a woman, and person). The label-less condition represented the standard way of assessing gender normative stereotypes (subsequently referred to as the “Control condition”) and constituted a direct replication of Prentice and Carranza (2002) for 60 of the 70 traits used. We note that the use of “homosexual” to denote the gender-neutral person term in the gay condition is now considered offensive and does not conform to American Psychological Association's (2015a^; 2015b) guidelines for inclusive language. We recommend future work use more inclusive language such as a sexually minoritized person.

Participants rated the traits on a 9-point scale from 1 (extremely undesirable) to 9 (extremely desirable) while answering, “How desirable is it in American society for a [TARGET] to possess each of the following characteristics?” We purposefully asked participants to indicate how they thought the “average American” would respond to help reduce demand characteristics and social desirability concerns (Fiske et al., 2002). We recognize that the use of “American” to refer to people from the U.S. was unnecessarily U.S.-centric. We do believe that participants understood what geographic area we were limiting their responses to, as all participants were currently living in the U.S.

The targets were presented sequentially such that participants rated trait desirability for all 70 traits for one target before moving on to the next. The order of the man and woman targets was displayed in a counterbalanced order such that the man target was first or second compared to the woman target, while the person target was always rated last. After the three rating tasks, participants described what they thought a typical straight and gay man as well as a typical straight or lesbian woman looked and acted like in a free-response task. Finally, participants filled out a series of exploratory questionnaires, including a sexual prototypicality scale and the Internal and External Motivation to appear Non-Prejudiced scale (IMS-EMS; Plant & Devine, 1998) adapted for LGBTQ people. We do not discuss the free responses or the exploratory questionnaires in this paper.

Data Analysis Strategy

We analyzed the data using a multilevel regression model in which Target Gender (3 Levels: Man, Woman, Person), Target Sexual Orientation (3 Levels: Straight, Gay/Lesbian, Control), and Traits (70 Levels: Aggressive, Ambitious, Approval Seeking, etc.) interacted to predict desirability ratings, as well as included a participant-level random intercept to account for the within-subject Trait and Target Gender factors. Target Gender and Trait were within-subjects while Target Sexual Orientation was between-subjects. We effects-coded each of our variables with “Person,” the Control condition, and the trait “Yielding” as reference categories. We also accounted for the impact of Participant Gender on desirability ratings as done in previous research (Prentice & Carranza, 2002) by including the two-way interactions between Participant Gender (effects coded) and each of our variables of interest. This was important as women overall reported more extreme stereotype prescriptions and proscriptions than men.

There were two alternative data analytic approaches we considered, namely running individual models for each trait as well as reducing the number of traits considered through some form of aggregation. We found these alternative approaches to be less optimal than the one described above. First, sequentially running 70 (for each trait) with in-between analyses of variance (ANOVAs) could not take into consideration the fact that each trait rating within Target Gender (210 trait ratings in total) was related. Thus, using multilevel models, while complex, allows for the greatest accounting of within-subject variance. Second, we did not reduce the number of trait categories for two reasons. The first reason was that these studies are meant to showcase the landscape of gender normative stereotypes at the intersection of sexual orientation and gender, allowing the current work to serve as a useful point of reference for future work on specific attributes that center intersectionality. Leaving the traits disaggregated fully showcases the richness of normative stereotypes.

The second reason was methodological in that reducing the data in either a bottom-up or top-down approach had problems. Creating superordinate categories from theory (i.e., top–down) requires us to make assumptions about the nature of gender stereotypes that the current work is meant to challenge. For example, what traits should be labeled masculine or feminine is likely group-specific, making any aggregation a poorer fit for one group relative to another. Similarly, creating superordinate categories in a bottom-up approach (e.g., exploratory factor analysis) makes it impossible to compare specific stereotypes across groups, one of the main contributions to the literature of the current design in conjunction with others, such as open-ended free responses (Ghavami & Peplau, 2013). In the spirit of serving as a reference, all data, data analysis scripts, descriptive statistics, and Supplementary analyses for this study as well as Study 2, can be found on the Center for Open Science website (https://osf.io/nu8v2/).

We used the resulting models as the basis for planned comparisons across and within Target Gender, Target Sexual Orientation, and Traits. To account for Type 1 error, we used the “emmeans” package in R (Lenth, 2022), controlling for multiple comparisons for all contrasts using the more stringent Bonferroni method. For large samples and complicated models, “emmeans” uses z-statistics for hypothesis testing instead of t-statistics, as it is much easier to calculate computationally. In this instance, using a z-test is the equivalent of using a t-test with infinite degrees of freedom. As the t-distribution approximates the z-distribution (i.e., the standard normal distribution) above 30 degrees of freedom, this is a reasonable substitution given our sample size in each factor-combination. At each stage, the more complicated model better fit the data as the marginal pseudo-R² (Bartoń, 2022) was .50.

Using this analytic strategy, we calculated the study's sensitivity to detect an effect as small as d = 0.04 (f² = 0.02), assuming the ANOVA repeated measures within-between interaction setting in G*Power (Erdfelder et al., 1996), an alpha level of 0.95, 80% power, three between-subject groups, and 210 repeated measures. Given our posthoc tests, a more relevant sensitivity analysis showed we could detect a t-test effect as small as 0.13–0.41, which includes the range for the potential paired and between-subject t-tests.

Hypothesis 1: Gender Normative Stereotypes Still Exist

We hypothesized that the gender normative stereotypes uncovered by Prentice and Carranza in 2002 would still exist in the U.S. We investigated this question by examining gender differences in trait desirability between “Man” and “Woman” in the Control condition. As hypothesized, women and men were held to different normative standards on a variety of traits (see online Supplemental materials, Table S5 for comparisons with findings from Prentice and Carranza, 2002). There were 55 traits out of 70 (78.6%) for which participants’ desirability for a man versus a woman to display those traits significantly differed. The trait differences in desirability for a man versus a woman were robust, with some trait differences (e.g., feminine and masculine) as large as six units on a nine-point scale. Thus, Hypothesis 1 was supported.

The top five prescriptive stereotypes for men included being self-reliant (M= 8.17), displaying leadership ability (M= 8.09), being ambitious (M= 8.06), being masculine (M= 8.03), and being decisive (M= 8.03), and the top five prescriptive stereotypes for women included being feminine (M= 8.08), being warm and kind (M= 8.01), being patient (M= 7.86), paying attention to appearances (M= 7.77), and being polite (M= 7.74). The top five proscriptive stereotypes for men included being weak (M= 1.77), being feminine (M= 2.16), being naïve (M= 2.16), being gullible (M= 2.17), and being child-like (M= 2.45), and the top five for women were proscribed the most from being arrogant, (M= 2.16), being masculine (M= 2.28), being controlling (M= 2.68), being promiscuous (M= 3.04), and being aggressive (M= 3.15; see Table S6 in the online Supplemental Materials for the full listing of trait desirabilities by target. Table S7 contains the prescriptive, proscriptive, and neutral designations of traits in general while Tables S8–S13 contain these labels by target).

Hypothesis 2: Androcentric and Heterocentric Biases Impact Perceptions of Normative Stereotypes

Next, we examined whether people's normative stereotypes were influenced by androcentrism and heterocentrism.

Androcentrism

We empirically defined androcentrism to mean that trait desirabilities would be more similar, or have greater conceptual overlap, between “Person” and “Man” targets compared to “Person” and “Woman” targets. We examined androcentrism by looking at the interaction between Target Gender and Trait, finding a significant interaction, F(138, 62299.76) = 40.06, p < .001. Collapsing across sexual orientation, the data supported an androcentric bias in gender normative stereotypes (Table 2), supporting Hypothesis 2 (analyses that also included Target Sexual Orientation were explored in Hypothesis 3). Overall, there were 30 traits for which the Person categories differed from the “Man” categories. In contrast, there were 42 traits for which the “Person” categories differed from the “Woman” categories. Of the 30 traits for which the Man categories differed from the Person categories, only four of them were traits for which the “Woman” categories did not also differ from the “Person” categories (i.e., Cooperative, Cynical, Expresses emotion, and Stubborn).

Open multimedia modal

Heterocentrism

We empirically defined heterocentrism to mean that the trait desirabilities for targets without a specified sexual orientation—“Man,” “Woman,” and “Person”—would be more similar to “Straight Man,” “Straight Woman,” and “Straight Person,” respectively, than “Gay Man,” “Lesbian Woman,” and “Homosexual Person.” We examined heterocentrism by comparing Target Sexual Orientation across traits, collapsing across Target Gender (see Hypothesis 3 for the three-way interaction). In support of Hypothesis 2, there was a significant interaction between Target Sexual Orientation and Trait, F(138, 62299.68) = 34.91, p < .001 (Figure 1). There was only a single trait for which the “Control” targets differed from the “Straight” targets (i.e., Materialistic). In contrast, there were 63 traits for which the “Control” targets significantly differed from the “Gay” targets (see Table 2).

Open multimedia modal

Hypothesis 3: Sexual Orientation and Race Altered the Prescriptive Stereotypes of Men and Women

We hypothesized that we would see evidence of gender inversion in normative stereotypes such that the normative stereotypes of gay men would be similar to straight women, while the normative stereotypes of lesbian women would be similar to straight men. We labeled traits as showing evidence of gender inversion if there were significant gender differences between straight men and straight women, as well as between gay men and lesbian women. Furthermore, the direction of the gender difference needed to be reversed from one sexual orientation to another—for example, if it was more desirable for a straight man to be aggressive compared to a straight woman, aggressiveness needed to be less desirable for a gay man compared to a lesbian woman. As expected, the three-way interaction was significant, F(276, 62299.72) = 17.69, p < .001. All pairwise comparisons for each trait by Target Gender and Target Sexual Orientation can be found in the online Supplemental materials (Table S14), as well as graphs of each trait plotted separately (see online Supplemental Materials in Figure S4).

Gender Inversion

There were 15 out of 70 traits for which there was gender inversion (Figure 2 and Figure S1). A subset of these traits (i.e., Attention to appearances, Cheerful, Excitable, Expresses emotion, Friendly, Stubborn, Theatrical, and Warm and kind) showed a unique type of gender inversion in which the desirability ratings for either the man or the woman target was the same for all sexual orientations but diverged for the opposite gender. What made this pattern of gender inversion noteworthy was that the normative pressure differed by sexual orientation for only one gender. For example, the desirability of paying attention to appearances showed gender inversion such that straight women and gay men faced a stronger prescription to pay attention to their appearances compared to straight men and lesbian women, respectively. However, the desirability of paying attention to appearances was the same for men of all sexual orientations but diverged for women, with people stating a lowered desirability for lesbian women to pay attention to their appearances compared to straight women. “Excitable,” “Expresses emotion,” and “Theatrical,” were the three traits for which there was normative pressure only on men of different sexual orientations compared to women.

Open multimedia modal

Other Interaction Patterns

Gender inversion did not describe the pattern of prescription and proscription for the bulk of the traits, however. Instead, for 41 traits, we found a pattern we are labeling as “Sexual Orientation Asymmetry,” such that there were differences in the normative expectations between a man and a woman, but only for one of the sexual orientations (see Figures S2 and S3). For 37 traits, there were significant gender differences between the straight targets but not for the gay/lesbian targets (i.e., Ambitious, Arrogant, Athletic, Business sense, Career oriented, Child-like, Competitive, Cooperative, Cynical, Decisive/able to make decisions, Defends own beliefs, Emotional, Feminine, Gullible, Happy, High self-esteem, Impressionable, Intelligent, Leadership ability, Loves children, Masculine, Melodramatic, Naïve, Patient, Polite, Promiscuous, Rational, Rebellious, Self-reliant, Shy, Spiritual, Strong personality, Weak, Well-educated, Wholesome, Willing to take risks, and Yielding). There were only four traits for which there were significant differences between gay men and lesbian women but not between straight men and women (i.e., Flirtatious, Materialistic, Musical, and Well-dressed). Thus, people had distinct gendered norms for how straight men and straight women should and should not act but did not have as strong gendered expectations for gay men and lesbian women. This conclusion is further supported by the fact that there were 52 traits for which trait desirability for a straight man differed from a straight woman but only 19 traits for which trait desirability differed between a gay man and a lesbian woman.

The last 14 traits did not show any gender differentiation in either straight or gay/lesbian targets. Of these traits, 13 of them were traits that showed a main effect of Target Sexual Orientation. More specifically, trait desirability was higher for straight targets than gay/lesbian targets for 10 traits (i.e., Clean, Conservative, Consistent, Dependable, Disciplined, Enthusiastic, Honest, Likeable, Loyal, and Protects others) while trait desirability was higher for gay targets than straight targets for three traits (i.e., Moody, Nosy, and Superstitious).

Participants

We again recruited roughly 100 participants per condition. Participants in this study came from Amazon Mechanical Turk (MTurk; www.mturk.com) recruited by TurkPrime (now CloudResearch, www.cloudresearch.com). MTurk is an online platform where people can complete tasks as a function of their skill sets and demographic characteristics for pay. MTurk is convenient and allows for easy recruitment of large sample sizes within a short amount of time. Research has found MTurk participants to be more diverse than participants from standard online pools and U.S. colleges (Berinsky et al., 2012^; Buhrmester et al., 2011), although there can be concerns about low data quality (Chmielewski & Kucker, 2020). Unfortunately, we again did not include direct attention checks.

We restricted the pool of participants to those local to the U. S. We paid our MTurk participants $1.20 for their time on a survey that took an average of 23 min to complete. Six hundred and nineteen out of 694 participants finished the study in full and were included in the analyses. Participants predominantly self-identified as White (n = 429), with the rest of the participants self-identifying as Black (n = 71), East Asian (n = 35), South Asian (n = 32), multiracial (n = 31), Latino/a (n = 8), Middle-Eastern (n = 6), and “Other” (n = 4). Three people declined to indicate their racial group. More people identified as female(n = 326) than as male(n = 284), while three participants indicated they were transgender. Six participants declined to give any information about their sex (for further information on sample demographic characteristics, see online Supplemental materials, Table S15). Due to a coding error, we did not collect sexual orientation information in this study.

Materials and Procedure

The procedure for Study 2 was identical to Study 1 with the sole change of Target Race replacing Target Sexual Orientation. Participants rated the desirability of the same 70 traits for targets that differed by gender (within-subjects) and race (between-subjects). We randomly assigned participants to one of six conditions: rating trait desirability for a man, a woman, and a person that was either Asian, Black, Latino, Middle-Eastern, White, or label-less (the “Control condition”). After the three rating tasks, participants described what they thought a typical man or woman of different races looked and acted like in a free-response task. Finally, participants filled out a series of exploratory questionnaires, including a racial prototypicality scale and the IMS-EMS scale (Plant & Devine, 1998) adapted for ethnic minorities. We do not discuss the free-responses or the exploratory questionnaires in this paper.

Data Analysis Strategy

We analyzed the data in Study 2 similarly to Study 1, but instead of Target Sexual Orientation, we interacted Target Race (six levels: White, Black, Asian, Latino, Middle-Eastern, Control) with Target Gender and Traits predicting desirability ratings using a multilevel model, including participant random intercept and the two-way interactions with Participant Gender. Target Gender and Trait were again within-subjects while Target Race was between-subjects. At each stage, the more complicated model better fit the data, as the marginal pseudo-R² (Bartoń, 2022) in Study 2 was .33. Table 3 (as well as Table S18 in the online Supplementary materials) contains the average desirability ratings for each trait by Target Gender and Target Race, classified as either prescriptive, proscriptive, or neutral (the ANOVA Tables for all interactions for desirability ratings and desirability difference scores can be found in the online Supplemental materials, Tables S16 and S17, respectively).

Open multimedia modal

Hypothesis 1: Gender Normative Stereotypes Still Exist

Men and women were also held to different normative standards in the Control condition within Study 2. There were 48 traits out of 70 (68.6%) for which participants’ desirability for a man versus a woman to display those traits significantly differed. To place the continuity of gendered prescriptions in context, there were 60 traits that both we and Prentice and Carranza (2002) studied. Of those 60, Prentice and Carranza found significant gender differences in 55 of them. We found significant differences for 47 of the 60 in Study 1 and 41 in Study 2, supporting Hypothesis 1 (see online Supplemental Table S5). The continuity of gender normative stereotypes extended to the more nuanced categorization of traits as intensified and relaxed prescriptive and proscriptive as outlined in Prentice and Carranza (2002). Those analyses can be found in the online Supplemental materials.

The most prescribed and proscribed stereotypes for men and women showed remarkable consistency across Study 1 and Study 2, as the top five prescriptive stereotypes for men again included being self-reliant (M = 7.85), displaying leadership ability (M = 7.93), being ambitious (M = 7.77), being masculine (M = 7.99), and being decisive (M = 7.78). The top five prescriptive stereotypes for women included being feminine (M = 8.00), being warm and kind (M = 7.73), being patient (M = 7.59), paying attention to appearances (M = 7.83), and being polite (M = 7.86). The top five proscriptive stereotypes for men included being weak (M = 2.15), being feminine (M = 2.34), naïve (M = 2.66), being gullible (M = 2.44), and being child-like (M = 2.78), and women were proscribed the most from being arrogant, (M = 2.71), being masculine (M = 2.36), controlling (M = 3.05), promiscuous (M = 3.54), and aggressive (M = 3.11; see Table S18 in the online Supplemental materials for the full listing of traits by target. Tables S19–S20 contain the traits labeled as prescriptive, proscriptive, and neutral overall, while Tables S21–S32 contain the traits labeled as intensified and relaxed by target.).

Hypothesis 2: Cognitive biases Impact People's Normative Stereotypes

Hypothesis 3: Race Altered the Perceived Normativity of Stereotypes Between Men and Women

Although there was not strong evidence for Eurocentrism in stereotypes in isolation, gendered stereotypes as a dynamic between men and women showed evidence of Eurocentrism (for additional details, see online Supplemental materials Figures S5–S10). As expected, the three-way interaction was significant, F(690, 127728.01) = 3.18, p < .001. We hypothesized normative stereotypes would conform to expectations set by the gendered race hypothesis: gender normative stereotypes for Black men and women would be masculinized, while normative stereotypes for Asian men and women would be feminized (Galinsky et al., 2013). We also explored whether we would find similar patterns for Latino and Middle Eastern men and women. We tested our hypotheses around the gendered race hypothesis by first examining racial differences in the desirability to be masculine and feminine (See online Supplemental materials Tables S33–S35 for all pairwise comparisons and Figure S15 for each trait graphed individually.).

Implications

These findings have several implications for the work on gender normative stereotypes specifically, and gender research more broadly. We highlight two. The first is that our current understanding of gender normative stereotypes (limited as it may be to the perspectives of dominant group members) might only apply to minoritized groups under restricted situations. Across both sexual orientation and race, being part of a minoritized group had a greater influence on people's normative expectations than being a particular gender. People did not desire gay/lesbian, Black, Asian, Latino, and Middle-Eastern men and women to act as differently from one another as they desired straight and White men and women to act. This implies that the normative pressures on minoritized groups fluctuate depending upon which aspect of their identity is foregrounded. Research on representations of transgender men and women has found a similar lack of gender differentiation within minoritized targets (Gallagher & Bodenhausen, 2021), a phenomenon they label as “de-gendering.” Other work has also found that global stereotypes around competence and warmth for Black men and women show more similarity than the same stereotypes between White men and women (Coles & Pasek, 2020), further supporting the conclusion that gender differentiation is reduced within minoritized groups.

Interestingly, within racially minoritized groups, we found that people differentiated Middle-Eastern men from Middle-Eastern women the most. We can only speculate, but it could be due to beliefs that Middle-Eastern people are particularly religious and conservative (Ghavami & Peplau, 2013), and thus this group retains normative expectations in line with traditional gender roles. It could also be due to weaker group stereotypes, and thus participants defaulted to more prototypical gender stereotypes, which reflect majoritized groups. Finally, there could be an effect of status, as perceptions of race and status are related (Dupree et al., 2021^; Penner & Saperstein, 2008^, 2013). For example, U.S. Middle-Eastern and Asian Americans are stereotyped as rich while U.S. Black and Latino Americans are stereotyped as poor (Ghavami & Peplau, 2013); we found the greatest gender differences in the targets that are stereotyped as wealthier. Indeed, social role theory posits that the stereotypes of men and women come from the social roles they occupy (Eagly & Steffen, 1984). Gender is not the only social identity that has correlations with expected social roles along class and status dimensions, supporting an intersectional approach to normative stereotypes. More research is needed on the normative expectations of minoritized groups more broadly to understand when individuals default to stereotypes about one identity over another (Petsko & Bodenhausen, 2020).

Second, while there is often concordance between descriptive and normative stereotypes (Prentice & Carranza, 2002) for men and women, that concordance became untethered in minoritized groups. In other words, for majoritized straight and White individuals, descriptive stereotypes match normative stereotypes. How people assume society expects straight men and women to be matches how straight men and women ought to be and ought not to be as well. However, there seems to be a mismatch for minoritized targets. One glaring example is the masculinity associated with White and Black men. Descriptively Black men are seen as more masculine than White men (Hall et al., 2015a^; Johnson et al., 2012). If the descriptive stereotypes associated with Black men in terms of masculinity matched normative stereotypes, we would expect participants to also indicate that it is most desirable for Black men (out of other races) to be masculine. Instead, we found that people desired Black men to be less masculine than White men.

This finding suggests that if a Black man displays the same level of masculinity as a White man, he is already too masculine and thus might face penalties for such behavior when compared to White men. Given the large amount of research on discrimination faced by Black men because of stereotypes around dangerousness and threat (Ghavami & Peplau, 2013^; Livingston & Pearce, 2009^; Pager et al., 2009), comparing descriptive and normative expectations of Black men gives us another account of how these disparities can arise. Indeed, the discrepancy between descriptive and normative stereotypes also can explain why Black men who act communally or appear non-threatening are rewarded (Livingston et al., 2012^; Livingston & Pearce, 2009).

Normative stereotypes associated with lesbian women also show this discrepancy between descriptive and normative stereotypes. Work on descriptive stereotypes broadly shows that people assume lesbian women are like straight men across a multitude of traits and characteristics. Normatively, however, there seems to be a distinct split in the masculine traits that lesbian women are desired to embody more so than gay men or straight women compared to the ones where they are not. For example, although it was more desirable for lesbians to be “Aggressive,” “Assertive,” and “Forceful” compared to gay men or straight women, there were no differences in the desirability for traits like possessing “Business sense,” being “Career oriented,” or having “Leadership abilities.” This suggests that although it might be acceptable for lesbian women to possess more masculine traits from a personality perspective, it is not always acceptable for them to possess masculine traits in ways that would disturb the gender status quo. This split in masculine traits is mirrored in work on descriptive stereotypes of women of different races. Black women, who are also seen as more masculine, are seen as agentic from a dominance perspective but not from a competence perspective (Rosette et al., 2016).

Limitations

While this study expands our understanding of gender normative stereotypes, there are limitations to the research presented here. The most obvious limitation is the lack of diversity within our samples, especially regarding sexual orientation and race. We extended work on gender normative stereotypes by disaggregating the categories of “men,” “women,” and “people,” by sexual orientation and race; we did not want to further complicate the design by including participant sexual orientation and race into the analyses as well. Given that we asked participants to indicate the beliefs of “an average American,” the same cognitive biases studied here likely influenced participants’ answers, making it even more probable that these findings represent the beliefs of White and straight members of the U.S. Thus, we want to be very explicit and say that these findings represent the normative beliefs of majoritized U.S. White, straight group members, and should not be assumed to generalize beyond those groups. That being said, understanding the normative pressures majoritized groups place on others is important as these individuals disproportionately hold political, economic, and social power. Consequently, their normative beliefs are more likely to influence real-world outcomes like backlash and discrimination. Phenomena such as third-party prejudice are predicated upon the pluralistic beliefs embedded within society (Vial et al., 2019). Future studies should investigate the normative pressures minoritized groups place on themselves rather than expectations of what the broader society desires.

Another limitation is our simplified understanding of gender and sexual orientation. We assumed a gender binary in our design, asking individuals about their stereotypes of men and women. This design cannot further our understanding of genderfluid, genderqueer, or non-binary individuals; future studies should examine normative stereotypes for individuals along the gender spectrum. Similarly, we only assessed stereotypes for gay men and lesbian women without acknowledging the diversity within those categories as well as other sexual and gender minorities like bisexual or transgender men and women, respectively. Participants likely assumed a more feminine gay man and a more masculine lesbian woman, but we do not have evidence to speak to this. The fact that multiple subtypes of gay men (e.g., body-conscious gay men or flamboyant gay men) and lesbian women (e.g., butch lesbian, feminine lesbian) exist (Brambilla et al., 2011^; Clausell & Fiske, 2005) might be driving some of the patterns observed. We did not find a strong norm for gay men or lesbian women to be masculine or feminine, which could be due to participants aggregating across the subtypes. Additional intersectional work is needed to look at subtypes within these social categories.

A third limitation is that we assessed stereotypes of people of different races without specifying other social identities that likely were assumed. For example, being a Muslim is often associated with individuals identifying as Middle-Eastern, thus, stereotypes towards U.S. Middle-Eastern Americans likely reflect, to some extent, beliefs about U.S. Muslim Americans as well. As another example, we asked about U.S. Asian Americans, but the prototype for U.S. Asian Americans is East Asian rather than South Asian (Goh & McCue, 2021^; Kibria, 1998^; Kuo et al., 2020). Finally, there has been a recent surge of stereotype and perception research at the intersection of race and sexual orientation (Johnson & Ghavami, 2011^; Pedulla, 2014^; Petsko & Bodenhausen, 2019^; Preddie & Biernat, 2021), suggesting that these categories also influence one another in ways we do not examine here. Thus, the lack of effects by gender and race could be due to divergent representations of subgroups that we did not account for.

A fourth limitation is in the unidimensional way we approach the selected traits. These dynamics can become further complicated because how one is compelled to perform “warmth” or “intelligence” might change at the intersection. For example, while we found that Black women faced reduced prescriptions to be “feminine” compared to White women, how femininity is understood and reinforced is likely qualitatively different for White and Black women. We did not give definitions of any of our traits, meaning that more precise definitions could yield different results. As another example, both White women and women of color have stereotypes related to being sexual, but only Black and Latina women are described as “promiscuous” (Ghavami & Peplau, 2013). White women are “sexually liberal,” suggesting that people might represent the trait “promiscuous” with different connotations that matter for backlash and discrimination.

Finally, our determinations of gendered patterns relied on statistical breaks. We struck a balance between statistical rigor and not letting a focus on p-values overshadow the exploratory nature of the work. However, we let the data tell us where the important cleavages in patterns of gendered normative stereotypes lie, and there were some ambiguous cases. For example, while the desirability for straight and gay targets to be clean resulted in a main effect of sexual orientation and not an interaction, the pattern, by eye, is more closely aligned to a gender asymmetry effect. It was more desirable for a straight woman to be cleaner than a lesbian woman but equally as desirable for a straight and gay man to be clean. While this is indeed a gender asymmetry pattern, the fact that there were no significant gender differences within both straight and gay targets precluded it from the category. Thus, we see gender normative stereotypes as more amorphous than what we present, and this is, in part, why it was important to include all means and comparisons in Supplemental materials. We still believe there is substantive utility in our studies, as these studies are but the first of, we hope, many that will examine normative stereotypes intersectionally.

Future Directions

Despite these limitations, the results of these two studies invite inquiry into many avenues of research. We highlight four here. The first is to expand our understanding of intersectional gender normative stereotypes. We believe the work presented is just the beginning of a rich landscape that can parse when certain normative pressures are heightened or mitigated at the intersection of multiple groups. The use of big data might be particularly fruitful in this endeavor. Work using natural language processing has found gendered associations within the broader societal written corpus (Charlesworth et al., 2021). These tools can also be used to assess associations for gender nuanced by race in a more naturalistic setting that is closer to people's lived experiences. What types of words or phrases follow “Black woman” in our written corpus compared to just “woman”? Having answers to this and other questions will further allow for nuanced policy recommendations and perhaps identify areas where the expectations of women differ by subgroups (Williams, 2014).

Second, this work encourages further exploration into gendered backlash. The current findings from lesbian and racially minoritized women suggest that how they will experience role violations will be distinct from straight and White women. Although some recent research shows that Black women do not always experience backlash when acting in agentic ways compared to White women and women in general (Brescoll & Uhlmann, 2008^; Livingston et al., 2012), much of the work blurs over racial distinctions. Third and relatedly, how do people experience contextualized normative pressures? Our work suggests that the normative pressures Black women are under, for example, depend on whether they are being perceived primarily by their race, their gender, or a combination of the two. As people feel compelled to conform to expectations and mitigate negative impressions (Lu & Wong, 2013^; Neel et al., 2013), there is an outstanding question as to whether intersectional group members accurately assess the norms they are expected to adhere to.

Fourth, these findings also highlight the importance of studying intra-racial and intra-sexual gender dynamics, as the very nature of gender normative stereotypes seems predicated upon the social identities at play. What does it mean for our gendered theories that there is a smaller distinction between the normative stereotypes of gay, Black, Asian, Latino, and Middle Eastern men and women compared to straight and White men and women? How should we understand perceptions of, and engagement in, gender dynamics intra-sexually and intra-racially? We are excited about the new research that will explore these questions.

Practice Implications

Our results have implications for people researching stereotypes at the intersection of identities as well as for those studying disparate treatment based on stereotypes. This work is meant to be a reference for scholars to reflect on how normative expectations might impact intersectional groups in a variety of settings, providing rich soil for hypothesis generation and result analysis. For example, it might be important for a scholar studying the treatment of lesbian women in the workplace to know that lesbian women might be under intensified prescription to be agentic compared to straight women but not necessarily more business oriented. These juxtaposed stereotypes might give their study important context and shape research design. To the extent that the given workplace environment would lead lesbian women to be perceived at the intersection of their gender and sexual orientation, our findings can help foster unique hypotheses that would not be generated with the current understanding utilizing descriptive stereotypes alone. Our work further highlights the importance of explicitly labeling relevant categories that might be tacitly assumed to be a particular subgroup within a category. For example, we have learned to do so ourselves in our misuse of “American” to denote people from the U.S. Such an assumption introduces noise in our data to the extent that the label “American” is unclear to participants and is unnecessarily U.S.-centric. We join the call for work to address the hidden assumptions in researchers’ stimuli and writing.