key: cord-1036127-8r40z1y9
authors: Mecit, Alican; Shrum, L. J.; Lowrey, Tina M.
title: COVID‐19 is Feminine: Grammatical Gender Influences Danger Perceptions and Precautionary Behavioral Intentions by Activating Gender Stereotypes
date: 2021-06-08
journal: J Consum Psychol
DOI: 10.1002/jcpy.1257
sha: 3dee43e60520e02decc05059dba5dc59c37e070f
doc_id: 1036127
cord_uid: 8r40z1y9

Gendered languages assign masculine and feminine grammatical gender to all nouns, including nonhuman entities. In French and Spanish, the name of the disease resulting from the virus (COVID‐19) is grammatically feminine, whereas the virus that causes the disease (coronavirus) is masculine. In this research, we test whether the grammatical gender mark affects judgments. In a series of experiments with French and Spanish speakers, we show that grammatical gender affects virus‐related judgments consistent with gender stereotypes: feminine‐ (vs. masculine‐) marked terms for the virus lead individuals to assign lower stereotypical masculine characteristics to the virus, which in turn reduces their danger perceptions. The effect generalizes to precautionary consumer behavior intentions (e.g., avoiding restaurants, movies, public transportation, etc.) as well as to other diseases, and is moderated by individual differences in chronic gender stereotyping. These effects occur even though the grammatical gender assignment is semantically arbitrary.

gender stereotypical perceptions (weaker, more passive, etc.), it may lead to perceptions that the virus or disease is less dangerous, as well as lower intentions to engage in precautionary behaviors to avoid contracting the disease. There is some indirect evidence consistent with this reasoning. Using archival data, Jung et al. (2014) showed that hurricanes with feminine names caused more deaths than hurricanes with masculine names. Although the underlying process was not tested, the authors proposed that the effects occurred because the femininenamed hurricanes were considered less risky because of gender-stereotypical associations, resulting in less preparedness for potential negative consequences. However, it is also important to note that the archival results have been strongly contested (Christensen & Christensen, 2014; Malter, 2014; Smith, 2016 ).

In the current research, we test the proposition that activating thoughts about the virus using the feminine (vs. masculine) grammatical gender will lead to gender-stereotypical perceptions of the virus (weak, passive, etc.), which in turn will lead to lower danger perceptions. We also expect that grammatical gender will influence precautionary consumer behavioral intentions, and that the effect of grammatical gender on individuals' stereotypical judgments about the virus will be stronger for individuals who hold strong (vs. weak) gender stereotypes.

Our research makes several contributions. First, we extend previous linguistic research (Konishi, 1993) showing that grammatical gender influences perceptions of masculinity and femininity in gender-stereotypical ways by showing that these perceptions influence downstream judgments (perceptions, behavioral intentions). To our knowledge, this is the first research to show such downstream consequences, suggesting that the effects of grammatical gender on gender-stereotypical perceptions happen spontaneously, rather than only when the perceptual judgments are explicitly elicited (cf. Konishi, 1993; Sera et al., 1994) . Second, we extend the findings of Jung et al. (2014) , showing the effects of name gender on risk perceptions by explicating the underlying process, demonstrating the generalizability of the effect, and with a more subtle activation of gender stereotyping (grammatical gender). Finally, we contribute to research on the intersection of language and consumer behaviour (Pogacar et al., 2018) by showing that in gendered languages, although Accepted Article grammatical gender of nonhuman nouns is an irrelevant contextual cue, it nevertheless affects consumer judgments by activating gender stereotypes.

We tested our propositions in a series of experiments with native French and Spanish speakers. Studies 1a-1c tested whether grammatical gender of the virus or disease affects danger perceptions and precautionary consumer behavioral intentions. Study 2 tested whether the findings generalize to diseases other than . Studies 3 and 4 tested the process and theoretically relevant boundary conditions. All participants provided informed consent, and we analyzed the data only after all measures had been collected. We only excluded participants based on a priori rules (see MDA, Part 2 for details). We measured mood and demographics in all studies, but their inclusion as covariates did not materially affect the results, and participant gender did not interact with grammatical gender (MDA, Part 2). All studies were conducted in the participant's native language. All raw data and stimuli are posted at https://osf.io/9437y (Mecit et al., 2021) .

Study 1 tested the hypothesis that activating thoughts about the virus using the feminine (vs. masculine) gender mark will lead to lower perceptions of danger and lower intentions of taking precautions to avoid contracting the virus in potential consumption situations. We tested this hypothesis in three separate studies (1a-1c) that were designed to address issues of generalizability and rule out alternative explanations. We began data collection in May 2020, when France and Spain were currently under their first prolonged lockdown (see MDA, Part 2 for dates). Thus, we measured future rather than current danger perceptions to avoid possible ceiling effects due to the overwhelming and devastating data coming in about the pandemic.

The procedure and design of the studies were identical except for sample composition and the manipulation of grammatical gender. Sample details, along with descriptive results, are shown in Table 1 for all studies (see also MDA, Part 2). For each language of administration, we restricted participants to native language speakers.

The experiments used a one-factor (grammatical gender: masculine, feminine) between-subjects design, with random assignment to conditions. Participants were told that

This article is protected by copyright. All rights reserved they would be participating in a short study about the public's reactions to the recent pandemic.

questions. For Study 1a (French) and Study 1b (Spanish), in the masculine grammatical gender condition, the instructions and the questions referred to le (Study 1a) or el (Study 1b) coronavirus, and in the feminine gender condition they referred to la COVID-19. These are the correct usages for the grammatical gender, and thus are the most ecologically valid. In Study 1c (French), to control for potential confounds related to the different words (coronavirus vs. COVID-19), we manipulated grammatical gender by whether the instructions and questions referred to le COVID-19 or la COVID-19. Although the masculine form for is grammatically incorrect, French speakers more often than not mistakenly use it (see MDA, Part 7 for examples). Thus, the manipulation also has ecological validity. In addition, we also conducted an additional study in English with native English speakers, to address the same confound issues (MDA, Part 3).

We measured precautionary consumer behavior intentions with six questions concerning future consumption behavior likely to be impacted by the coronavirus (likelihood of eating at a restaurant soon, traveling by plane, etc.), and measured future danger perceptions of the virus with five questions (how long will the virus remain dangerous, how likely it is that there will be a second wave, etc.). Factor analyses indicate that the perceptions and intention measures generally load on distinct factors, although the pattern structure varied across studies (MDA, Part 4). We created composite measures of behavioral intentions and danger perceptions (see MDA, Part 5 for alphas). Participants then answered an attention check question and provided demographic and mood information.

Our hypotheses were supported in all three studies (Table 1) . French participants (Study 1a) in the feminine condition (M feminine =4.71, SD feminine =1.20) thought that the virus would be less dangerous in the future compared to those in the masculine condition (M masculine =5.14, SD masculine =1.06; t(145) =2.35, p=.02, d=0.39) , and also intended to be less cautious in their future behaviors (M feminine =4.07, SD feminine =1.22 vs. M masculine =4.56, SD masculine =1.24; t(145) =2.42, p=.017, d=0.40 =2.18, p=.031, d=0.36) , and precautionary intentions (M feminine =4.08, SD feminine =1.12 vs. M masculine =4.51, SD masculine =1.05; t(148) =2.41, p=.017, d=0.39 ).

The results of Studies 1a-1c provide converging evidence that grammatical gender influences perceptions of danger and intentions to engage in precautionary consumer behavior. Study 1b with Spanish participants shows that the effects are not specific to French, and Study 1c provides further evidence in support of grammatical gender effects by demonstrating the effects holding the name constant (la vs. le .

Study 2 tested whether the results of the previous studies generalize to diseases other than COVID-19. To do so, we constructed a set of actual diseases, half of which take the masculine gender mark in French and half take the feminine gender mark, and had participants rate their severity and fatality. The two gender-marked subsets did not differ in terms of actual severity and fatality (MDA, Part 6). We expected that French participants would judge the feminine-gender-marked set to be less dangerous than the masculine-marked set, consistent with the previous studies. However, given that English does not grammatically mark gender, we expected that the danger judgments would not differ for the English participants.

Participants were either French and native French speakers (n=100) or English and native English speakers (n=100) who were randomly assigned to conditions in a 2 (grammatical gender: masculine, feminine)×2 (native language: French, English) mixed design, with language as a between-subjects factor and grammatical gender as a withinsubjects factor.

In a study ostensibly about judgments concerning different diseases, participants evaluated the severity and the fatality of 18 different diseases and health conditions, 9 of which took the feminine grammatical gender (e.g., tuberculosis, malaria) and 9 of which took Accepted Article the masculine grammatical gender in French (e.g., diabetes, tetanus; MDA, Part 1). We averaged the severity and fatality ratings for each disease in the respective set to create composite measures of danger for the feminine (α=.85) and masculine (α=.84) disease sets.

Participants then provided demographic and mood information.

A repeated measures ANOVA revealed that only the interaction was significant, F(1,198)=94.67, p<.001, η 2 =.32. As expected, French speakers judged the set of femininemarked diseases (M=4.74, SD=0.78) to be less dangerous than the masculine-marked set (M=5.15, SD=0.67; t(99) =9.26, p<.001) . However, English speakers unexpectedly judged the feminine set (M=4.75; SD=0.66) to be more dangerous than the masculine set (M=4.55, SD=.59; t(99)=-4.51, p<.001), even though there were no gender cues for English participants. Although the masculine and feminine disease sets did not differ in objective risk, this reversal may have occurred because risk judgments were based on factors other than objective risk (e.g., accessibility; Lichtenstein et al., 1978) . Regardless, the pattern of the interaction is consistent with our theorizing.

Study 3 tested whether stereotypical judgments about the virus mediate the effect of grammatical gender on danger perceptions. We also tested a theoretically relevant boundary condition. We expected that chronic gender stereotypes would moderate the mediation effect, such that the effects of grammatical gender on stereotypical judgements about COVID-19 would be stronger for those who hold stronger gender stereotypes (moderation at path a). To demonstrate generalizability, we measured current danger perceptions in Study 3, given that the study was conducted when many of the pandemic-related restrictions had been lifted (January 2021).

Participants were 305 native French speakers who were randomly assigned to conditions in a one-factor (grammatical gender: masculine, feminine) between-subjects design, with chronic gender stereotyping as a measured moderator.

The manipulation of grammatical gender was the same as in Study 1c (le vs. la . Participants first answered five questions concerning their current danger perceptions about COVID-19 (e.g., how threatened do you feel, how difficult is it to

This article is protected by copyright. All rights reserved eradicate). Next, following a filler task designed to clear working memory (solving 15 anagrams), we measured their stereotypical judgments about COVID-19 by having them rate COVID-19 on a set of four bipolar adjectives adapted from previous studies on gender stereotypes (e.g., weak/strong, passive/aggressive; Konishi, 1993; Rudman et al., 2001) . The danger perceptions (α=.71) and stereotypical judgments (α=.90) loaded on distinct factors (MDA, Part 4).

Participants then answered an attention check question, followed by a 24-item gender stereotypes questionnaire designed to assess individual differences in gender stereotyping.

The questionnaire asked how typical it would be for men and women to each possess 12 different characteristics that are typical and atypical of each gender (e.g., strong, gentle). We calculated typicality ratings by subtracting inconsistent gender stereotypes from consistent ones for each gender, and then created composite measures of gender stereotyping tendencies by averaging the relative typicality ratings for men (α=.86) and women (α=.89), with higher scores indicating greater traditional gender stereotypes (Hentschel et al., 2019; Ruble, 1983;  see MDA, Part 1 for calculation details). Finally, participants provided demographic and mood information.

As predicted, participants in the feminine condition (M=4.43, SD=1.06) perceived COVID-19 to be less dangerous than did those in the masculine condition (M=4.78, SD=0.94); t(300)=3. 03, p=.003, d=0.35) . Participants in the feminine condition (M=2.93, SD=1.26) also associated COVID-19 with more stereotypical feminine characteristics than did those in the masculine condition (M=2.45, SD=1.10; t(300) =-3.20, p=.002, d=0.37) .

Regression analyses further revealed that the predicted grammatical gender × chronic gender stereotyping interaction was significant, both for gender-stereotypical judgments (β=0.56, SE=0.16, p<.001; Figure 1a ) and danger perceptions (β=-0.39, SE=0.12, p<.001; Figure 1b ).

Finally, the manipulation of grammatical gender did not influence the chronic gender stereotyping measure (p = .36).

We tested for moderated mediation using Hayes' (2017) PROCESS Model 7 with 5,000 bootstrapping re-samples (see MDA, Part 6 for tests of additional models). The moderating effect of chronic gender stereotyping on the relation between grammatical gender

This article is protected by copyright. All rights reserved and gender stereotypes about the virus was significant (β=0.56, SE=0. 16, p<.001) . Gender stereotypes about the virus also significantly influenced current danger perceptions (β=-0.46, SE=0.04, p<.001). Controlling for gender stereotypical judgments about the virus, the direct effect of grammatical gender on danger perceptions is not significant (p=.17). Probing further, at the mean level of the moderator (chronic gender stereotyping), the effect of grammatical gender on danger perceptions is mediated by stereotypical judgments about COVID-19 (β=-0.21, SE=0.07, 95% CI=[-0.35, -0.08]), and as predicted, the effect is stronger for participants who hold stronger gender stereotypes (1 SD above the mean; β=-0.45, SE=0.11, 95% CI=[-0.67, -0.23]) compared to those who hold weaker stereotypes (1 SD below the mean; β=0.02, SE=0.09, 95% CI=[-0.15, 0.20]), the latter of which is not significant (Figure 1a) .

Study 4 tested whether stereotypical judgments about the virus explain the effect of grammatical gender by manipulating the process. If the effects occur because grammatical gender influences gender-stereotypic perceptions, then reducing gender stereotyping should attenuate the effect. We tested this hypothesis by priming a counter-stereotypical gender mindset (Blair et al., 2001) . We also used a new, expanded measure of gender stereotypes about the virus that included more items, and we changed the procedure slightly by asking them to consider COVID-19 as a person to make the use of the new items (e.g., gentle, kind) more plausible. Finally, we measured precautionary behavioral intentions with new measures that are more current.

Participants were 402 native French speakers who were randomly assigned to conditions in a 2 (grammatical gender: le COVID-19, la COVID-19)×2 (counter-stereotypic mental imagery: yes, no) between-subjects design.

Participants were told that they would be participating in two different studies, purportedly about social perception and their reactions to the recent pandemic. As part of the first study, participants first answered an open-ended question that served as the manipulation of a counter-stereotypic mental imagery. Participants in the counter-stereotypic condition were asked to describe what a strong woman is like, why she is strong, and what she is

This article is protected by copyright. All rights reserved capable of doing, whereas participants in the control condition were asked to describe what a vacation in Corsica is like, how the place looks, and what people do there (Blair et al., 2001) .

Next, as part of the second, unrelated study, participants were provided with a brief description of the French government's measures against COVID-19 for the next four weeks concerning the third wave, and were asked to answer a series of questions about their behaviors during this four-week period and perceptions concerning COVID-19. The manipulation of grammatical gender was the same as in the previous study: The instructions and the questions referred to either le COVID-19 (masculine) or la COVID-19 (feminine).

Participants first answered four questions concerning their precautionary behaviors related to COVID-19 (e.g., increase their online shopping to decrease face-to-face contact, buy masks that provide extra security, etc.), completed a filler task similar to Study 3, and then completed items that measured gender stereotypical judgments about COVID-19 by having them rate COVID-19 on a list of 12 adjectives (e.g., aggressive, mean, submissive, kind; see MDA, Part 1). The behavioral intentions (α=.74) and stereotypical judgements (α=.84) loaded on distinct factors, and we computed composite measures, with higher scores indicating greater precaution and greater stereotyping. Participants then answered an attention check question, and provided demographic and mood information.

A two-way ANOVA revealed a significant main effect of grammatical gender, F(1,387)=9.91, p=.002, η 2 =.025, with participants in the feminine (vs. masculine) condition intending to be less cautious in their behaviors (Table 1) .40; t(196)= 0.49, p=.63) . In addition, the pattern was consistent with our theorizing: Relative to the control condition, priming a counter-stereotypic mindset significantly increased precautionary intentions (p = .035) and reduced stereotypical judgments about the virus (p < .001) in the feminine condition, but did not significantly affect intentions and judgments in the masculine condition (ps > .15; Table 1 ).

This article is protected by copyright. All rights reserved 

In this research, we demonstrate how a simple linguistic cue-the grammatical genderaffects both perceptions of danger and intentions to engage in precautionary behaviors.

Across a series of experiments, we show that COVID-19 is considered less likely to be dangerous when the disease is marked with the feminine (vs. masculine) grammatical gender, and that the grammatical gender effect also generalizes to other diseases as well. To the best of our knowledge, this research is the first to causally demonstrate such downstream effects of grammatical gender on judgment and decision-making. Moreover, the findings have immediate relevance, as the success of several measures (e.g., social distancing and washing hands) is dependent upon the willingness of individuals to adopt the behaviors, and one determinant of compliance is the extent to which individuals perceive the virus to be dangerous.

We also provide a process explanation of the grammatical gender effects. Previous research has shown that grammatical gender can affect connotations of masculinity and femininity (Konishi, 1993; Sera et al., 1994 ; for a review, see Gentner & Goldin-Meadow, 2003) . We take this research one step further, showing that grammatical gender affects stereotypical judgments about the virus, which in turn are used in constructing judgments about the danger of the virus. Consistent with this mechanism, we further identify individual differences in gender stereotyping as a theoretically relevant boundary condition, such that the effect of grammatical gender on stereotypical judgments related to masculinity and femininity are stronger for people holding stronger gender stereotypes. Thus, our research provides a theoretical mechanism that has potentially important consequences.

This article is protected by copyright. All rights reserved Our research has implications for a number of research areas. First, the research contributes to the literature documenting the effects of seemingly irrelevant information on important consumer judgments. Assignment of grammatical gender is arbitrary, and thus should not logically influence judgments; speakers of gendered languages are well-aware that grammatical gender has no meaning for nonhuman entities. Second, our findings extend research on grammatical gender effects, and contribute to the larger debate as to whether language influences thought (Lucy, 1997; Whorf, 1952) . Our research further confirms and extends the findings on the implicit nature of grammatical gender effects (Boutonnet et al., 2012; Cubelli et al., 2011) by showing that grammatical gender can influence judgements and decision-making, even if such information is irrelevant and not explicitly elicited.

Our findings also suggest avenues for future research. For one, although our research focused on specific virus-related judgments, grammatical gender is likely to influence other types of consumer judgments. For example, to the extent that grammatical gender nonconsciously activates gender-related concepts, it may influence judgments of gendermarked brand names and products. Further, to the extent that grammatical gender imparts human-related information (either masculine or feminine), the gender mark of a product (or the absence of gender marks in genderless languages) may influence how consumers interact with products, such as the extent to which they anthropomorphize them.

An additional question is the extent to which our findings generalize to other gendered languages. Our experimental findings show that the effects hold for both French and Spanish. However, French and Spanish are both romance languages and have two grammatical genders. Given that grammatical gender effects are more likely to occur in languages with only two grammatical genders (Maciuszek & Świątkowska, 2019) , one avenue for future research is to test the generalizability of the effect in languages with more than two grammatical genders.

One limitation of the current research is that we were unable to show the effects on consequential choice, which was hampered by limitations of the pandemic restrictions. Future research would benefit from such tests, both experimentally and time-lagged studies.

Finally, the effects we show are novel, and the novelty is primarily driven by the novelty of the situation: Two new words are introduced into the lexicon, the words describe similar things but take different grammatical gender marks, and for one of the words (COVID-19), speakers often use the grammatical gender incorrectly. We show that a

This article is protected by copyright. All rights reserved seemingly irrelevant grammatical cue affects perceptions of danger and intentions to take precautionary measures. Thus, even though the motivation of Académie Française for urging proper grammar usage is surely well-intentioned, it may have had unfortunate unintended consequences. 

This article is protected by copyright. All rights reserved

Participants and design. One hundred fifty members of the Prolific Academic UK online research panel (79 women, 70 men, 1 not indicated; M age = 42.49, SD = 13.96) participated in the experiment in exchange for £0.30. We restricted participants to native English speakers, located in the United Kingdom, who had previous Prolific approval rates greater than 95%. The experiment was a one-factor (coronavirus vs. COVID-19) between-subjects design.

The experiment was conducted in English.

Procedure. The experimental procedure and the questions were identical to those of Study 2 and 3, except that the conditions did not include a grammatical gender because the study was conducted in English. We created composite measures of precautionary behavioral intentions (α = .64) and future danger perceptions (α = .65) by averaging all items, with higher scores indicating greater cautionary behavior and perceived danger. Finally, participants provided demographic and mood information. 

This article is protected by copyright. All rights reserved

Factor Analyses for Study 1 French Sample. We first examined the factorability of all 11 items for future danger perceptions and precautionary consumer behavioral intentions. The Kaiser-Meyer-Olkin measure of sampling adequacy was .82, above the commonly recommended value of .6, and Bartlett's test of sphericity was significant, χ2 (55) = 695.49, p < .001. The diagonals of the anti-image correlation matrix were also all over 0.5. All these suggested that it was acceptable to proceed with the factor analysis.

A factor analysis was performed using the Maximum Likelihood method of extraction. A Promax rotation was performed since factors were expected to be correlated. The obtained pattern matrix is displayed in Table 2 . Only items with factor loadings of above .30 are shown.

As the table shows, the danger perceptions and behavioral intentions load on distinct factors, although the behavioral intentions were composed of two factors in this sample. 

This article is protected by copyright. All rights reserved 

This article is protected by copyright. All rights reserved

We first examined the factorability of all 9 items for current danger perceptions and stereotypical judgments about COVID-19. The Kaiser-Meyer-Olkin measure of sampling adequacy was .88, above the commonly recommended value of .6, and Bartlett's test of sphericity was significant, χ2 (36) = 1193.49, p < .001. The diagonals of the anti-image correlation matrix were also all over 0.5.

A factor analysis was performed using the Maximum Likelihood method of extraction. A Promax rotation was performed since factors were expected to be correlated. The obtained pattern matrix is displayed in Table 5 . Only items with factor loadings of above .25 are shown. 

This article is protected by copyright. All rights reserved The moderating effect of chronic gender stereotypes on the relation between stereotypical judgments and danger perceptions (moderation on path b) was significant (β=-0.12, SE=0.03, p < .001) (Figure 2 ).

Probing further, at the mean level of the moderator (chronic gender stereotypes), the effect of grammatical gender on danger perceptions is mediated by stereotypical judgments about COVID-19 (β=-0.19, SE=0.06, 95% CI=[-0.32, -0.07]), and as predicted, the effect is stronger for participants who hold stronger gender stereotypes (1 SD above the mean; β=-0.52, SE=0.12, 95% CI=[-0.76, -0.27]) compared to those who hold weaker stereotypes (1 SD below the mean;

β=0.02, SE=0.06, 95% CI=[-0.10, 0.14]), the latter of which is not significant.

This article is protected by copyright. All rights reserved This article is protected by copyright. All rights reserved Accepted Article

The dynamics of masculine-agentic and feminine-communal traits: Findings from a prospective study

Imagining stereotypes away: The moderation of implicit stereotypes through mental imagery

Sex, syntax, and semantics

Language in mind: Advances in the study of language and thought

Unconscious effects of grammatical gender during object categorisation

Are female hurricanes really deadlier than male hurricanes?

The effect of grammatical gender on object categorization

Gender and aggressive behavior: A meta-analytic review of the social psychological literature

A model of (often mixed) stereotype content: Competence and warmth respectively follow from perceived status and competition

Language in mind: Advances in the study of language and thought

Introduction to mediation, moderation, and conditional process analysis: A regression-based approach

The multiple dimensions of gender stereotypes: A current look at men's and women's characterizations of others and themselves

Female hurricanes are deadlier than male hurricanes

Accepted Article This article is protected by copyright. All rights reserved

The semantics of grammatical gender: A cross-cultural study

Judged frequency of lethal events

Linguistic relativity

Grammatical gender influences semantic categorization and implicit cognition in Polish

Female hurricanes are not deadlier than male hurricanes

Grammatical gender & coronavirus [data set

The effects of linguistic devices on consumer information processing and persuasion: A language complexity × processing mode framework

Sex stereotypes: Issues of change in the 1970s

Implicit self-concept and evaluative implicit gender stereotypes: Self and ingroup share desirable traits

Grammatical and conceptual forces in the attribution of gender by English and Spanish speakers

When language affects cognition and when it does not: An analysis of grammatical gender and classification

Hurricane names: A bunch of hot air? Weather and Climate Extremes

Language, mind, and reality

Accepted Article