The use of a single bipolar gender scale to evaluate the estimated gender breakdown of an occupational group has long been the standard measure for assessing the effects of linguistic-grammatical information and gender stereotypes on the mental representations of gender. However, this scale presupposes a reciprocal relationship between the gender poles, assuming that the estimated number of one gender group constrains the estimation of the other. This is questionable, as distinct linguistic modulations and gender stereotypes may exert asymmetrical influences on each gender group’s representation. The current study, therefore, used gender-separated scales and showed that an inclusive language manipulation on the presentation of occupations can increase the visibility of women when the occupations do not carry feminine stereotype information but can also decrease the visibility of men when the occupations do not carry masculine stereotype information. Using this scale may be a more appropriate approach for future studies to capture the influence of language forms on gender representation in a more nuanced way.
Introduction
The mental representation of gender is intricately shaped by the linguistic-grammatical information an individual has access to and one’s knowledge of gender stereotypes (e.g., P. Gygax et al., 2008, 2021; Sato & Athanasopoulos, 2018). For instance, in English, a word like nurses mentally activates a group composed predominantly of women in the representation as it has a strong stereotypical association with the feminine gender (Oakhill et al., 2005). In contrast, the grammatical masculine marking in the French equivalent infirmiers exerts a greater impact than the associated feminine stereotype, evoking a representation mostly of men as triggered by its masculine surface form (e.g., P. Gygax et al., 2008). Given the inherent grammatical differences between languages, research into the mental representation of gender has sought to explore how the various sources of gender information interact to produce distinct representations of women and men (e.g., P. Gygax et al., 2012; Sato et al., 2013; Schunack & Binanzer, 2022; Siyanova-Chanturia et al., 2012). One common approach employed by the studies that examine the influence of grammatical gender and gender stereotypical information is to present gender-stereotypical occupation nouns that are selected based on norming ratings of gender ratios (Gabriel et al., 2008, 2023; Kennison & Trofe, 2003; Misersky et al., 2014; Tibblin et al., 2023). Norming ratings of gender ratios are considered to provide an efficient measure of gender stereotypicality (Kim et al., 2024) and have thus been widely used as an approach for experimental stimuli selection. Studies collecting these norming ratings present occupational nouns manipulated for their grammatical form and/or gender stereotype and instruct their participants to estimate the proportion of women and men who occupy the occupational role (e.g., “Estimate the extent to which the presented social and occupational groups actually consist of women and men”, Misersky et al., 2014). Participants’ estimates are recorded on a bipolar scale, typically spanning 11 points, with one end representing [100% women and 0% men], and the other end representing [100% men and 0% women] (Braun et al., 2005; Gabriel et al., 2008, 2023; Hansen et al., 2016; Horvath et al., 2016; Misersky et al., 2014; Tibblin et al., 2023; Xiao et al., 2023).
Although this practice of estimating gender ratios on a bipolar scale is widespread, here, we question whether this scale format actually provides an ideal measurement that accurately captures the complexity of the perceived gender ratios of a given occupational group, particularly for studies that aim to use these gender ratios to examine the spontaneous activation and interaction of gender stereotype and grammatical information. This argument is based on the fact that bipolar scales assume a reciprocal and symmetrical relationship, such that a relative increase in the proportion of one gender group would unequivocally imply a decrease in the proportion of the other gender group (Yorke, 1983). However, as research on gender representation aims to evaluate the spontaneous modulations influenced by grammatical information and one’s knowledge of stereotypes, the estimates of the gender composition of a group of people may not accurately reflect the grading changes in gender ratios. Based on this reasoning, the present study aims to propose an alternative scale that evaluates how different sources of gender information impact the estimated distributions of women and men and to provide evidence for its efficacy in the context of gender representation research.
Gender Representation in French
The extensive research on gender representation is driven by the fact that there are asymmetries in the way gender is expressed in languages. This feature is particularly evident in grammatical gender languages such as French, where gender specification is explicit and obligatory (P. M. Gygax et al., 2019). The central argument in this line of research is that the grammatical masculine form, which is initially learnt as referring exclusively to men (e.g., mécaniciensgrammatically masculine [men mechanics]), can also function in a generic sense to address a group composed of both women and men (i.e., women and men mechanics) (P. Gygax et al., 2009). The multiple meanings of the masculine form can thus lead to an ambiguity in interpretation, with research showing that the interpretation referring exclusively to men is still largely favoured, obscuring the mental visibility of women (P. Gygax et al., 2008, 2021; Stahlberg et al., 2001, 2007). As a result, the use of the masculine form to convey a generic meaning is now widely discouraged (American Psychological Association, 2020) in terms of gender equality and political correctness (Maass et al., 2014). Although this practice of using the masculine form as a generic is still widespread, there is a burgeoning public interest in adopting alternative gender-fair language (GFL) forms that aim to mitigate this persistent masculine bias. For instance, the explicit use of both the grammatically feminine and masculine forms to refer to a group of individuals, known as the pair form (e.g., mécaniciennesgrammatically feminineet mécaniciensgrammatically masculine [women and men mechanics]), aims to cognitively emphasise that the group is made up of both genders rather than just men, thereby increasing the mental visibility of women.
In light of this emerging movement, there has been a surge of studies investigating the impact of different GFL forms compared to the masculine generic form (e.g., Hansen et al., 2016; Tibblin et al., 2023; Xiao et al., 2023). These studies often use the conventional bipolar scale or ask participants to estimate the proportional composition of the gender groups. The findings consistently demonstrate that using GFL instead of the masculine grammatical form leads to higher estimates of the proportion of women, highlighting the increased representation of women in participants’ mental representations. Despite the assumed consistency of the results, the methodology used in these studies does not allow us to determine whether GFL functionally amplifies the visibility of women while symmetrically diminishing the visibility of men or whether the representation of men remains unchanged in participants’ mental representation of gender. This issue also raises the question of whether gender representation is cognitively ipsative, in other words, whether the distribution of one gender category is dependent on the other category and can be inferred unquestionably.
Ipsativeness in Gender Representation
Logically, if gender representation is ipsative, we would argue that the representation of each gender category would be dependent on the estimated proportion of other gender categories. Therefore, an increase in the visibility of one gender group (e.g., women) will decrease the visibility of the other gender group (i.e., men). For example, thinking about nurses will activate images of not only women nurses but potentially also of the rather infrequent non-women nurses (i.e., men nurses) simultaneously. This may even be more pronounced when the pair form is used, as both gender categories are made linguistically salient. Alternatively, if gender representation is, in fact, not ipsative (i.e., non-reciprocal), we would argue that GFL may heighten the visibility of women while retaining the representation of men constant. In such a case, encountering a word like mechanics in GFL should activate a representation that is still associated with men (i.e., its prominent stereotype) but also with a heightened visibility of women. This view is supported by models of semantic memory networks, which suggest that when a category group is activated (i.e., mechanics), it spreads activations to connected nodes with strong associations (e.g., men), whereas modes with weaker connections (e.g., women) would receive a reduced level or no excitation (Anderson, 1983; Collins & Loftus, 1975). However, GFL might strengthen the latter association without weakening the former.
These opposing accounts raise questions about how linguistic interventions such as GFL impact the mental representation of gender distributions and how different research instruments or response scales that reflect these distributions can give rise to potential errors. For instance, if we assumed gender representation as non-ipsative, it would be methodologically questionable to create norming studies in which the two gender poles of the scale are interrelated, such that the estimate of one gender group constrains the estimate of the other gender group. That is, when estimating the gender breakdown of a word like mechanics, one might estimate that 90% of all workers in this occupation are men, as this occupation is stereotypically associated with the masculine gender. Without estimating the number of women mechanics, bipolar scales automatically set the proportion of women in this occupation at 10% due to their symmetrical nature. This means that the estimates of the number of women in an occupation are defined as gradations dependent on the opposite pole or, in other words, the estimated number of men workers in that occupation.
Another related aspect when considering potential errors arising from bipolar scales is that the associative strength for the opposing gender pole is not symmetrical. Specifically, ‘men’ generally represent the prototype for the category ‘human’, whereas ‘women’ do not (Bailey & LaFrance, 2017; Bem, 1993). In this respect, while ‘men’ (the default value) may be activated by the concept of ‘women’, ‘women’ may not necessarily be activated by the concept of ‘men’. This asymmetry could potentially lead to subtle association errors when relying on norms based on bipolar gender scales. Moreover, the existing literature on judgment errors suggests that when individuals are required to make quantitative estimates, they often rely on heuristics to weigh information, which in some cases can lead to inaccurate beliefs and systematic errors (Hertwig et al., 1999; von Helversen & Rieskamp, 2008; Zou & Bhatia, 2021).
In light of these issues, we argue that although estimations regarding the distributions of women and men in specific occupations are shown to be largely in line with true gender distribution ratios (Garnham et al., 2015), an evaluation of the response scales is vital as they may lead to critical errors. From a methodological view, and as suggested by Gannon and Ostrom (1996), explicitly labelling the two polar categories on a bipolar scale can lead to the activation of both mentioned categories, whereas employing unipolar scales will only activate the explicitly labelled category. This implies that when confronted with a bipolar scale, participants’ attention would be focused on both gender groups because the task requires them to make conscious decisions about both women and men (i.e., the salient categories on the scale). However, if gender-separated scales are employed, attention should be focused on only one of the gender groups made salient by the scale, essentially allowing participants to ignore the ratio of the other gender group. If we assume that certain GFL interventions may not have the same effects on the representation of women and men, then assessing the estimated number of women or men on a bipolar scale may not be sufficient to understand how different linguistic forms affect the estimated number of women and men.
The present study, therefore, evaluated the suitability of using gender-separated scales to independently assess the gender breakdown of occupational groups when assessing the effect of GFL on mental representations of gender. Specifically, we compared adults’ estimates of the distributions of women and men in gender-stereotypical occupations presented in either the masculine form or in the pair form in French. While previous studies have documented the efficiency of the pair form in enhancing the visibility of women, none have examined how these language forms affect estimates of the distributions of women independently of the estimates of the distributions of men and vice versa. Critically, dissociating the gender poles further allows us to address the potential reciprocity (or ipsativeness) of gender representation. While bipolar scales constrain the dependency of the two gender poles, using gender-separate scales may no longer constrain the gender groups, thereby providing a means to examine whether the estimates of one gender category are independent of the estimates of the other. Thus, we work under the premise that gender representation may not always be ipsative, as suggested by past studies. Finally, to prevent participants from using ipsative strategies when making their estimates (e.g., 70% means 30% from the other category), we deliberately avoided conceptualising the number of women and men as having an upper limit (i.e., totalling 100%; only women or only men). Thus, the poles of the scale ranged from “very few women” (très peu de femmes) or “very few men” (très peu d’hommes) to “many women” (beaucoup de femmes) or “many men” (beaucoup d’hommes), with the scale spanning on a continuous slider scale as opposed to a discrete Likert scale.
Methods
All materials, data and analysis codes related to this study are available on the Open Science Framework (OSF) project page: https://osf.io/avbp5/
Participants
We collected data from 90 students from the University of Fribourg (Switzerland) for course credit. An additional 54 native-French-speaking men were also recruited from the crowd sourcing platform Prolific to attain a gender balance. Two participants from the student sample were excluded as they were not native French speakers, resulting in a final sample of 145 native French-speaking participants (72 women, 72 men, 1 individual who preferred not to answer; Mage = 20.8 years, SDage = 3.66)1. The study was reviewed and approved by the Ethics Committee at the Department of Psychology of the University of Fribourg.
Materials
Based on Misersky et al.’s (2014) norming study, we selected six stereotypically feminine (e.g., infirmiers [nurses]), six stereotypically masculine (e.g., mécaniciens [mechanics]), and six gender-neutral (e.g., pharmaciens [pharmacists]) occupation nouns in French as stimuli. The average proportion of women perceived to be in the selected feminine, masculine, and gender-neutral occupations was 76%, 22%, and 51%, respectively. All the selected occupation nouns had distinct grammatically feminine and masculine forms. For instance, common nouns (Corbett, 1991) that have only one form to accommodate the grammatically feminine and masculine form, such as secrétaires (women and/or men secretaries), were not included in the stimuli. This criterion allowed us to construct items in each of the two language forms of interest: the grammatically masculine plural form theoretically and potentially representing a generic form (e.g., pharmaciensgrammatically masculine [women and/or men pharmacists]) and the combination of the grammatically feminine and the masculine form representing the pair form (e.g., pharmaciennesgrammatically feminineet pharmaciensgrammatically masculine [women and men pharmacists]). In addition, the labels of these occupations were slightly modified in line with the latest standards, taking into account the Swiss website for vocational guidance (orientation.ch, n.d.). For example, mécaniciennes et mécaniciens auto (women and men car mechanics) was revised to mécaniciennes et mécaniciens en maintenance automobile (women and men car maintenance mechanics). We then developed short job descriptions for each occupation noun referencing the official vocational website of the canton of Vaud in Switzerland (vd.ch/formation, n.d.). These descriptions made no reference to gender (see the Materials folder on the OSF project page for a full list of the occupations and their job descriptions).
Finally, to assess participants’ estimates of the distributions of women and men in each occupation, we devised separate scales, a women’s scale ranging from très peu de femmes to beaucoup de femmes (very few women to many women) and a men’s scale, ranging from très peu d’hommes to beaucoup d’hommes (very few men to many men). The scales were converted into a numerical value from 0 to 100 for analytical purposes, with the pole “very few women/men” reflecting “0” and “many women/men” reflecting “100”. The numerical representation remained hidden from the participants during data collection. To avoid any initial bias, the cursor used to indicate participant responses remained hidden at the start and became visible only upon participants indicated their responses with a mouse click on the scale.
Design
The experiment took a mixed factorial design with language form (2: masculine form vs. pair form) as a between-participant factor, and scale type (2: women vs. men) and occupation stereotype (3: feminine vs. masculine vs. gender-neutral) as within-participant factors. To ensure that each participant saw all items in each condition, we constructed a total of four lists using a Latin-square design, with participants randomly assigned to only one of the four lists. In each list, half of the items in each occupation stereotype condition were rated using the women’s scale and the remaining half using the men’s scale, with each list being fully presented either in the masculine or pair form. The gender-specific scales were presented in separate blocks, with block order counterbalanced across participants. The order of the presentation of the items within each block was randomised for each participant.
Procedure
The questionnaire was administered online using the Qualtrics platform (Provo, UT), where participants gave informed consent and responded to general demographic questions prior to completing the main questionnaire. In the main questionnaire, occupation nouns were presented one at a time, along with their brief descriptions in the respective language form condition. Participants were simply informed that there are a number of women (or men, depending on the experimental block of the gender scale type presented to them) who work in the given occupation and that their task was to indicate by positioning the cursor on the scale, whether they thought that there are very few or many women (or men) working in the given occupation. A screenshot of an item is shown in Figure 1.
Data Analysis
As the presentation order of the two gender scales was counterbalanced across participants, we ran an initial analysis to ascertain if there were differences in ratings between the two experimental blocks and whether being presented first with the women’s or men’s scale would influence gender distribution estimations. Participant ratings were thus subjected to a mixed ANOVA with scale order (women’s scale presented first vs. men’s scale presented first) as between-participant factor and experiment block (block 1 vs. block 2) as a within-participant factor. An unexpected main effect of scale order indicated that the ratings were higher in general when participants were presented with the women’s scale first (M = 55.5, SD = 28.2) than when they were presented with the men’s scale first (M = 52.7, SD = 28.7), F(1, 143) = 7.62, η2G = .03, p = .007. No other main or interaction effects were significant. To control for this effect of scale order, we included this predictor as a fixed effect in the main analyses assessing language form effects.
Next, to assess whether gender distribution estimates are influenced by their associated stereotype or by the language form in which they were described, ratings for the occupation nouns were subjected to linear mixed-effect models using the mixed() function in the afex package (Singmann et al., 2023). The model included fixed effects corresponding to our theoretical interests, consisting of language form (masculine generic form vs. pair form), occupation stereotype (feminine vs. masculine vs. gender-neutral), scale type (women vs. men), and their interactions. Additionally, to account for the effect of scale order that emerged in the preliminary analysis, this predictor was also included in the model. All predictors were sum coded as per default of the afex package. We fitted the maximal random effect structure justified by the experimental design (Barr et al., 2013), including both random intercepts for participants and items, as well as by-participant random slopes for scale type, occupation stereotype, and their interaction, and by-item random slopes for scale type, language form, and their interaction. In case of non-convergence or singular fit, we first suppressed correlations among random parameters. If the issues persisted, we proceeded to remove the random slope with the lowest estimated variance. Thus, the final random effect structure included random intercepts for participant and items, uncorrelated by-participant slopes for scale type, occupation stereotype, and their interaction, as well as an uncorrelated by-item slope for scale type. Follow-up analyses were conducted using the emmeans package (Lenth, 2023) and corrected with Holm’s method. Effect sizes were calculated with the effectsize package (Ben-Shachar et al., 2020).
Results
In addition to the scale order effect (already indicated in the preliminary analysis, F[1, 141.04] = 9.56, ηp2 = .06, 95% CI [.01, 0.15], p = .002)2 the analyses revealed a predicted Occupation Stereotype x Scale Type interaction, F(2, 16.46) = 87.13, ηp2 = .91, 95% CI [.80, .95], p < .001, as well as a Language Form x Scale Type interaction, F(1, 142.40) = 15.99, ηp2 = .10, 95% CI [.03, .20], p < .001. Critically for our study, both interactions were further qualified by a Language Form x Scale Type x Occupation Stereotype interaction, F(2, 135.88) = 3.41, ηp2 = .05, 95% CI [.00, .13], p = .036, showing that the use of our particular scale yielded different patterns of results depending on the gender stereotypes the occupation nouns were carrying (see Figure 2)3.
Specifically, the three-way interaction showed that presenting feminine stereotyped occupations in the masculine or pair form did not lead to differences in the estimates of women (MMasculine Form = 78.7 vs. MPair Form = 78.8; b = -0.34, SE = 2.00, t-ratio = - .17, p = .865, d = -.03) or men (MMasculine Form= 31.0 vs. MPair Form = 32.0; b = -1.21, SE = 2.00, t-ratio = - .60, p = 0.546, d = - .09). In contrast, presenting masculine stereotyped occupations in the pair form led to greater estimates of women (i.e., estimates on the women’s scale) than when they were presented in the masculine form (MMasculine Form = 19.9 vs. MPair Form = 24.2; b = -4.73, SE = 1.88, t-ratio = -2.51, p = .012, d = - .34). However, the estimates of men (i.e., estimates on the men’s scale) for masculine stereotyped occupations were unaffected by language form (MMasculine Form = 84.4 vs. MPair Form = 82.0; b = 2.03, SE = 1.88, t-ratio = 1.08, p = .281, d = .15). As for gender-neutral occupations, the pair form led to greater estimates of women than when the masculine form was used (MMasculine Form = 51.9 vs. MPair Form = 57.1; b = -5.54, SE = 2.21, t-ratio = -2.51, p = .012, d = - .40), yet this pattern was reversed for the estimates of men (MMasculine Form = 57.5 vs. MPair Form = 51.3; b = 5.90, SE = 2.21, t-ratio = 2.68, p = .008, d = .43). No other main effects or interactions were significant (ps > .09).
These results suggest that the responses were strongly driven by stereotypes and can therefore be considered ipsative overall (i.e., if an occupation is thought of as including more women, it will include fewer men). Crucially, however, the effect of GFL was not entirely ipsative across stereotype conditions. Essentially, for gender-neutral occupations, the effect was ipsative, as an increase in the estimated number of women was paralleled by a decrease in the estimated number of men. For stereotypically feminine occupations, GFL did not affect estimations. However, interestingly, for masculine stereotyped occupations, the effect of GFL was not ipsative, as an increase in the estimated number of women was not accompanied by a decrease in the estimated number of men.
Discussion
The current study investigated whether using the pair form rather than the masculine form leads to an increase in the visibility of women in participants’ mental representations while either decreasing the visibility of men or keeping the visibility of men constant. In doing so, we addressed the suitability of using gender-separated scales to overcome the potential ipsativeness of gender representations when studying gender-related linguistic interventions (i.e., GFL).
Language Form Effects
Our analyses revealed that the language form used to present occupation nouns indeed impacted the mental representation of gender, although to different degrees, depending on the stereotype of the occupation nouns. Specifically, using the pair form led to a higher estimated number of women than when the masculine form was used to present masculine stereotyped and gender-neutral occupations. However, although the estimated number of men simultaneously decreased for gender-neutral occupations in the pair form, this was not the case for masculine stereotyped role nouns, with the estimated number of men remaining constant for the latter. The estimated number of women for stereotypical feminine occupations remained high and constant regardless of language form, while the estimated number of men remained low and constant.
These results suggest that the pair form not only emphasises the feminine gender for occupations that do not have a strong association with women, but it can also diminish the visibility of men in the absence of strong masculine stereotype information. However, in the presence of masculine stereotype information, it does not exert the same effects. As for the masculine form, we have shown that when presented alone, the masculine bias it triggers is cognitively robust. The extent of this bias manifests differently for masculine-stereotyped and gender-neutral occupations, affecting the estimated number of women and men distinctly. Since past studies relied on single bipolar scales, those showing an increase in the estimated number of women have always assumed that such an increase necessitated a corresponding decrease in the estimated number of men. However, the use of gender-separated scales in our study showed that this was not the case and that the extent of the influence of the pair form on the representation of men can differ depending on the presence of different gender stereotype information.
The robust masculine bias, apparent in gender-neutral and stereotypically masculine occupations, is most likely brought about by the continued exposure to the use of the masculine form as a specific referent (i.e., masculine = men). These findings align well with the notion of the masculine default (Cheryan & Markus, 2020), wherein the masculine gender provides the social standard that dictates people’s beliefs, expectations, norms, and practices. The use of independent gender scales shows that the representations of men were essentially impervious to the effects of the language form in which they were presented (i.e., participants saw similar distributions of men in both forms). However, the representation of women remained flexible and open to subtle linguistic interventions. The fact that the women’s scale showed stronger effects in the language form in which the masculine stereotyped and gender-neutral occupations were presented may also be due to the fact that there is a greater variability in the occupations taken up by modern women. In other words, it is more common for women to enter occupations traditionally dominated by men than for men to enter occupations traditionally dominated by women (England, 2010). This is because there is a growing number of women obtaining higher education degrees (England, 2010), and women-dominated occupations tend to pay less (England, 1992), discouraging men from entering these positions. This would also explain why stereotypically feminine occupations showed no influences of language forms, given that the stereotype association with women was so robust.
To test this hypothesis further and disentangle the effects of language from those of societal mechanisms, one possibility for future studies would be to present jobs in the grammatically feminine form and ask participants to estimate the distribution of men in these occupations. This would be particularly interesting for feminine stereotyped occupations, which are often presented in the grammatically feminine form (e.g., “nurses” is often translated as “infirmièresgrammaticallyfeminine”) despite men also occupying these positions.
Regarding our recommendation of using gender-separated scales, it is also worth noting that measuring the estimations of the distributions for distinct gender groups may be more appropriate for the growing research interests investigating the representation of non-binary individuals or individuals who do not identify with the binary categorisation of gender groups (Decock et al., 2023; Richards et al., 2016; Stetie & Zunino, 2022; Zacharski & Ferstl, 2023).
Finally, it is important to address concerns that may arise regarding the lack of preregistration for this study. Although we adhered strictly to our initial study plan, we recognise that the absence of preregistration can raise valid questions about the reliability of our findings, which should be considered exploratory. Additionally, our sample size may not be large enough to establish our three-way interaction as a true effect definitively or to ensure that our effect size meets the appropriate threshold4. Therefore, future studies with adequate power are essential to replicate our results and provide robust evidence.
Ipsativeness in Gender Representation and Their Cognitive Mechanisms
A critical aspect that deserves discussion is considering the type of processes we are addressing. As has been mentioned, gender ratio ratings have been commonly used as a basis for stimuli selection in numerous psycholinguistic studies examining the online processing of gender stereotypes and/or grammatical gender information (Gabriel & Gygax, 2008; P. Gygax et al., 2008; Irmen & Kurovskaja, 2010; Oakhill et al., 2005; Sato et al., 2016). These studies have consistently shown that stereotype and grammatical information are immediately and spontaneously activated, leading to an intricate interaction emerging at early temporal time frames.
These rather spontaneous processes, however, need to be differentiated with more strategic processes, such as making decisions about gender ratios using bipolar scales, in which participants are forced to simultaneously activate both the less dominant gender category and the more dominant one. In other words, when explicitly instructed to estimate the gender distribution of women or men occupying certain occupations, a bipolar scale obliges them to answer by contrasting both gender categories. To overcome this issue, we have tried to present a separate scale for each gender category. However, the ipsative nature of the GFL effect was still apparent for gender-neutral stereotype occupations and feminine stereotyped occupations to some extent. Only for the masculine stereotyped occupations did our two scales show seemingly independent GFL effects. Researchers interested in fully separating the effects associated with the representation of women from those associated with the representation of men should use tasks that do not explicitly require participants to make assumptions about gender distributions or ratios as a whole (Carreiras et al., 1996; Oakhill et al., 2005; Reynolds et al., 2006). For example, through a series of experiments, Morehouse et al. (2022) showed that readers struggled to resolve a simple riddle in associating a stereotypical male occupation surgeon with a woman (i.e., mother). Had the participants simultaneously activated women when processing surgeon (as would be done when estimating ratios or the number of members of different gender categories), such difficulty would probably not have emerged. Future research may want to more consistently address ipsativeness in the ways gender representation data are collected and interpreted. For now, our study hints that subtle yet significant differences may appear when using non-ipsative scales.
Conclusion
In summary, our study showed for the first time that while the visibility of women is cognitively increased when using the pair form, changes in the representation of men vary according to the presence or absence of specific gender stereotypes. We showed that the use of gender-separated scales to make estimations about the gender breakdown of occupational groups was efficient in revealing the different extent of the influence of different language forms, given that it allows individuals to focus specifically on the aspect of a particular gender group occupying the position without necessarily making both gender groups salient. In this way, researchers can obtain a more fine-grained picture of how specific language forms and gender stereotype information may interact and contribute to the mental representations of women and men. These scales may be beneficial as gender perception may show ipsativeness in certain contexts. Future research may address the specific contexts in which ipsativeness is more or less pronounced.
Author Contributions
S.S. and P.M.G. conceived and planned the study. S.S. and P.M.G. developed the theoretical framework and collected the data. S.S. conducted the analysis, and P.M.G. supervised the project. S.S., P.M.G., U.G., J.O. and L.E. contributed to interpreting the results. S.S. took the lead in writing the draft of the manuscript, and all authors commented on previous versions of the manuscript. All authors read and approved the final manuscript.
Competing Interests
The authors declare no conflict of interest regarding the research reported in this article.
Funding
The research reported in this paper was funded by the Swiss National Science Foundation (SNSF; FN-501100001711-188841) awarded to Pascal Mark Gygax (PI), Ute Gabriel, and Jane Oakhill.
Data Accessibility Statements
All data, materials and analysis codes related to this study are available on the Open Science Framework (OSF) project page: https://osf.io/avbp5/
Footnotes
Participants’ self-identified gender was assessed using a question included in the questionnaire. Participants could respond between woman, man, other, or prefer not to answer.
Removing the scale order predictor from the model did not change other effects.
To further explore whether the current findings depend on the self-identified gender of the participant, we reran the analysis including participant gender in the model. We eliminated one participant who preferred not to answer to the question regarding their self-identified gender, resulting in a total of 72 women and 72 men. This analysis revealed that although the main effect of participant gender was significant F(1, 139.69) = 9.92, ηp2 = 0.06, 95% CI [.01, 1.00], p = .002, with women showing significantly higher ratings than men (Mwomen = 55.2 vs. Mmen = 52.9), none of the remaining effects differed from the main analyses.
We conducted a posthoc power analysis using the function powerSim() function in the simr package (Green & MacLeod, 2016) to evaluate the ability to detect our three-way interaction. The estimated power was 61.90%, CI (58.8, 64.92), which is below the commonly desired threshold of 80%. This suggests that while our findings are promising, further research with larger sample sizes is necessary to confirm the robustness of our results.