Introduction

Imagine that you are a woman who works in an office that employs mostly men. One morning, as you walk past the office of a male colleague, you overhear him mention your name to another male coworker. Slowing down to get a better listen, you distinctly hear your two male colleagues describe you as an “aggressive bitch.”

Now imagine, as above, that you are a woman who works in a predominantly male office. One morning, while walking past the office of a male colleague, you overhear him mention your name to another male coworker. Slowing down to get a better listen, you hear your two male colleagues describe you as a “sweet girl.”

Which of these two scenarios would be more upsetting to the woman involved? We suspect that, in answer to this question, most people will choose the first, hostile sexism scenario. Because hostile sexism involves overtly angry attitudes and behavior toward women, it appears on the surface to be especially hurtful to its targets. In contrast, the second scenario depicts benevolent sexism, which consists of well-intentioned and paternalistic attitudes and behaviors toward women (Glick and Fiske 1996). Because of its protective, paternal nature, benevolent sexism may come across as relatively benign to observers. Given that people base their predictions about the emotional impact of future events on their beliefs about the relative severity of those events (Gilbert et al. 2004), angry insults will most likely earn higher emotional damage estimates than well-intentioned infantilization.

On this point, however, we suspect that most people are incorrect. A growing body of literature indicates that people are woefully bad at predicting the affective impact of future events (for reviews, see Wilson and Gilbert 2003, 2005). Building on this literature, we propose that most people commit a forecasting error when estimating the hurtfulness of ambivalent (hostile and benevolent) sexism. More specifically, we propose that people overestimate the hurtfulness of hostile sexism (because of its “dramatic” nature), and underestimate the hurtfulness of benevolent sexism (because of its subtlety and prosocial patina). The result is a tendency to assume that hostile sexism, relative to benevolent sexism, produces more extreme negative emotions in the short-term, and requires a longer recovery period in the long-term.

If so, then such errors in understanding are likely to intensify the difficulty women encounter when trying to communicate their discomfort with sexism to male friends, colleagues, or employers (Kaiser and Miller 2004; Swim and Hyers 1999). Given that targets of stereotypes expect others to evaluate them negatively for “overreacting” to episodes of discrimination (e.g., Stangor et al. 2002) recipients of benevolent sexism in particular might refrain from discussing their experiences with others. In turn, their failure to disclose their emotions may prolong their negative affective reactions to the sexism (e.g., Pennebaker 1997). Thus, in light of the frequency with which women encounter both hostile and benevolent sexism in their daily lives (Swim et al. 2001) it is important to investigate the accuracy of people’s beliefs about the emotional impact of ambivalent sexism. Doing so will add to the literature on ambivalent sexism by clarifying the impact that each form of sexism (hostile and benevolent) has on its targets.

Note that our research also has significance across many cultures. In work on ambivalent sexism, investigators have documented a seemingly universal tendency for people to hold ambivalent attitudes toward women (e.g., Glick 2006; Glick et al. 2000). Thus, although we collected data on an American sample, the experience of ambivalent sexism is not purely an American phenomenon.

The goals of the current study are threefold. First, we examine whether American undergraduates indeed commit a forecasting error when estimating the emotional impact of hostile and benevolent sexism. To do this, we compare people’s forecasts of the intensity and duration of victims’ negative affect against women’s real reactions to both types of sexism. Second, to understand why people mispredict women’s reactions to sexism, we test a model in which immediate affective reactions mediate the link between sexism type and recovery duration. That is, we ask whether people mispredict the duration of recovery to hostile and benevolent sexism because they mispredict how upset each type of sexism initially makes its recipients. Finally, we consider the roles of relevant individual difference variables in shaping women’s affective reactions to sexism. We begin by summarizing briefly the ambivalent sexism and affective forecasting literatures.

Ambivalent Sexism

A basic assumption behind work on ambivalent sexism is that sexist beliefs fall along two dimensions, one of which reflects antipathy toward women who challenge conventional gender roles, and the other of which reflects chivalrous attitudes toward women who embrace such roles (Glick and Fiske 1996, 2001). For example, the belief that “women are untrustworthy and manipulative” reflects hostile sexism, whereas the belief that “women are pure and should be cherished” reflects benevolent sexism. Furthermore, although these two dimensions are independent in theory, they are moderately correlated in practice. In fact, there is a tendency for people across a wide range of cultures to hold, simultaneously, both hostile and benevolent attitudes toward women (hence the “ambivalence”; see Glick et al. 2000).

Much of the research on ambivalent sexism uses the Ambivalent Sexism Inventory (Glick and Fiske 1996) — a measure of stable individual differences in hostile and benevolent beliefs about women — to predict various outcomes. This work reveals that ambivalent sexism predicts polarized evaluations of women. Among college samples, hostile sexism consistently predicts negative evaluations and stereotypes of women, whereas benevolent sexism consistently predicts positive evaluations and stereotypes. For instance, men’s hostile sexism scores are associated with less favorable evaluations of nontraditional women (e.g., career women), whereas their benevolent sexism scores are associated with more favorable evaluations of traditional women (e.g., homemakers; Glick et al. 1997). Individual differences in ambivalent sexism also predict tendencies toward wife abuse, tolerance for sexual harassment, attitudes about rape, and body dissatisfaction (Forbes et al. 2004; Glick et al. 2002; Russell and Trigg 2004; Viki and Abrams 2002), among other things. These findings thus focus on ambivalent sexism from the perspective of the attitude holder.

In comparison, relatively little work examines ambivalent sexism from the perspective of its recipients. Moreover, the research that does investigate women’s reactions to benevolent and hostile sexism tends to use one of two different methodological approaches, yielding two different pictures of the underlying phenomenon. One approach involves assessing participants’ responses to hypothetical stimulus materials that epitomize hostile and/or benevolent sexism. For example, Killianski and Rudman (1998) asked U.S. female undergraduates to form impressions of male targets who espoused either hostile or benevolent sexist beliefs. On average, women rated the benevolently sexist man more favorably than the hostilely sexist man. Similarly, Barreto and Ellemers (2005) presented participants with the results of a bogus study in which samples of respondents ostensibly endorsed statements that communicated either hostile or benevolent sexism. As in Kilianski and Rudman’s study, women participants in Barreto and Ellemers’ study formed a more favorable impression of the (bogus) respondents who held benevolent sexist beliefs as compared to those who held hostile sexist beliefs. Moreover, women viewed the holders of benevolent attitudes as less sexist than holders of hostile attitudes, and they felt less angry after reading the benevolent than hostile attitudes. Taken together, the results of these studies indicate that when most people, including women, consider hypothetical cases of hostile and benevolent sexism, they perceive hostile sexism as worse than benevolent sexism.

The other approach to understanding recipients’ views of ambivalent sexism involves exposing research participants to either hostile or benevolent sexism that is directed at them personally. For example, Dardenne et al. (2007) exposed unemployed adult and undergraduate French women to either benevolently sexist, hostilely sexist, or non-sexist (control) messages as part of an ostensible job hiring paradigm, and examined the effects of these messages on women’s subsequent performance on working memory tasks. Women in the hostile sexism condition heard a speech by an employment recruiter in which he complained about legal quotas requiring the industry to hire “people of the weaker sex,” chided women for being “so easily upset,” and accused feminists of “exaggerat[ing] women’s situation in industry simply to get more favors.” In contrast, the recruiter in the benevolent sexism condition assured women that their male coworkers “will cooperate and help you get used to the job” because “they know that the new employee could be a woman.” The results of this research revealed that the benevolent message impaired women’s cognitive performance relative to the control message, whereas women exposed to the hostile message did just as well as control participants. Moreover, the reason that women suffered impaired performance in the benevolent sexism condition is that they experienced mental intrusions related to self-doubts about their own abilities. Thus, when women are exposed to hostile and benevolent sexism that is directed at them, they appear to be more hurt by benevolent than hostile sexism.

To summarize, women seem less bothered by benevolent than hostile sexism when considering these attitudes hypothetically, but they seem more negatively impacted by benevolent than hostile sexism when these attitudes are directed at them personally. Given this pattern of findings, we suspect that ambivalent sexism is just the type of emotional event that falls prey to forecasting errors.

Affective Forecasting

People are often wrong when predicting how good or bad a future event will make them or others feel. Not only do people routinely overestimate how strongly some events will make them feel (e.g., Dunn et al. 2003; Gilbert et al. 1998; Wilson et al. 2000), they also underestimate the emotional impact of some events (e.g., Gilbert et al. 2002; Gilbert et al. 2004). Such fallacious forecasts have been demonstrated with a wide range of emotional events including being denied tenure (Gilbert et al. 1998), undergoing a romantic break-up (Eastwick et al. 2008), receiving money unexpectedly (Wilson et al. 2005), and learning that a favorite football team has won the game (Wilson et al. 2000). To our knowledge, however, no past work has investigated people’s tendency to commit forecasting errors with regard to ambivalent sexism.

Note that forecasting errors can assume several different forms. People may mispredict the intensity of their initial emotional reactions to future events (an initial intensity bias), they may mispredict the rate at which an emotional reaction will decay (a decay bias; see Eastwick et al. 2008; Wilson and Gilbert 2003), and they may mispredict how long it will take for their emotions to fade (referred to here as a duration bias, to distinguish it from Gilbert et al.’s 1998 durability bias, which refers specifically to the tendency to overestimate the duration of affective reactions). Here, we examine how the initial intensity bias and the duration bias work in tandem to drive people’s forecasting errors about ambivalent sexism. Specifically, we propose that the initial intensity bias fuels the duration bias, such that mispredictions about the initial intensity of people’s emotional reactions to a given event should lead to mispredictions about the length of time it takes for these emotions to fade.

We base our logic on work by Gilbert et al. (2004), in which undergraduate participants estimated both the intensity of their initial emotional reactions, and the duration of their affective responses, to nine different hypothetical events. For example, participants estimated how they would feel both during, and one week after, being turned down for a date, catching someone breaking into their gym locker, and being stood up for a study session. Across participants, these two estimates were strongly correlated (r = .88), indicating that people generally assume a very strong link between the intensity and duration of emotional reactions. Thus, we expected participants to over– or underestimate women’s recovery length following exposure to ambivalent sexism because they over– or underestimate women’s immediate affective reactions to the sexism.

Overview of Study

We tested for the presence of two types of forecasting errors with regard to women’s experiences with ambivalent sexism. Specifically, we asked whether people display an initial intensity bias (by overestimating women’s immediate affective reactions to hostile sexism and underestimating their affective reactions to benevolent sexism), and a duration bias (by overestimating the duration of women’s recovery from hostile sexism and underestimating the duration of their recovery from benevolent sexism). We further tested whether people’s mispredictions about initial intensity drive their mispredictions about the duration of recovery following both types of sexism.

Intuitively, it makes sense to assume that forecasting errors should be the unique province of the uninitiated, i.e., those who have no prior personal experience with the event about which they make forecasts. However, research suggests instead that people do not typically adjust their forecasts to match their experiences (Gilbert et al. 1998). Nonetheless, to ensure that people’s beliefs about the impact of hostile versus benevolent sexism do not merely reflect personal inexperience, we collected estimates of initial affective reactions and recovery length from three different classes of forecasters who varied in their levels of prior exposure to sexism against women: (a) men (who have never been victims of sexism against women), (b) naïve women (who claimed to have no personal experience with sexism), and (c) experienced women (who claimed personal experience with sexism, but considered a hypothetical rather than real instance of sexism when providing responses). To determine whether forecasters displayed initial intensity and duration biases, we compared their estimates of how hostile and benevolent sexism would make people feel to actual reports made by women experiencers of benevolent and hostile sexism. Thus, experiencers’ reports about real incidents of sexism served as the criterion for determining forecasters’ accuracy.

To examine the presence of an initial intensity bias in forecasters’ estimates, we assessed a range of negative affective responses to hostile or benevolent sexism. Research suggests that women who are exposed to sexism display heightened feelings of anger, disgust (e.g., LaFrance and Woodzicka 1998; Vescio et al. 2005), and sadness (Schneider et al. 2001). Moreover, targets of prejudice often experience self-consciousness (e.g., Frable et al. 1990; Pinel 1999), particularly when attention is drawn to their stigmatized status (Cioffi 2000; Saenz 1994), as it is during sexist incidents. Therefore, we asked both forecasters and experiencers to rate the intensity of women’s immediate affective reactions to sexism along several dimensions related to anger, disgust, sadness, shame, and embarrassment. To measure the duration bias, we asked forecasters and experiencers to either estimate or recall the number of days required to recover, emotionally, from sexism.

Finally, we explored whether certain individual differences played a role in experiencers’ tendency to recall specific episodes of sexism, or in their affective reactions to the incident that they recalled. Specifically, experiencers completed scales that assessed their chronic tendencies to perceive sexism in men’s behavior (Pinel 1999), and the strength of their identification with their gender (Luhtanen and Crocker 1992). We selected these individual differences based on research suggesting that women who are higher in stigma consciousness, and those who identify more strongly with their gender ingroup, are particularly likely to interpret ambiguous negative feedback or maltreatment as sexism (Major et al. 2003; Pinel 2004).

In summary, we manipulated both participants’ status (forecasters vs. experiencers) and the type of sexism (hostile vs. benevolent) we asked them about, and measured the effects of these variables on their estimates of initial affective reactions and recovery duration. In doing so, we tested the following hypotheses. Hypothesis 1: Forecasters will overestimate experiencers’ initial affective reactions to hostile sexism and underestimate their initial affective reactions to benevolent sexism. Hypothesis 2: Forecasters will overestimate the duration of experiencers’ recovery from hostile sexism and underestimate the duration of their recovery from benevolent sexism. Hypothesis 3: Mispredictions about initial intensity will mediate the link between participants’ status and their mispredictions about recovery duration.

Method

Participants

A total of 201 U.S. undergraduates (150 women and 51 men) participated, in groups of up to 20, in exchange for credit toward a course requirement. We deleted data from 13 women whose responses disqualified them (see below), leaving a final sample of 188 participants.

Procedure

Participants completed a questionnaire packet that began by defining sexism as “any comment or behavior directed toward a woman that suggests that she is unequal to a man, or is incapable of doing things that men can do.” Instructions next explained that “sexism may come in different forms: Sometimes it takes on a protective, caregiving tone, such as when a woman is told ‘not to worry her pretty head.’ Other times it involves more hostile behaviors, such as using a rude word to refer to a woman.” After this opening statement, the questionnaires differed depending on version.

Manipulating/Assigning Role

Men (n = 51) were all assigned to the role of male forecasters, and received a version of the questionnaire that did not query them about their personal experiences with sexism. Women, through random assignment, received a questionnaire that asked them about their personal experiences with sexism either before or after they completed the dependent variables. Of the women who indicated their personal sexism experiences at the beginning of the survey, those who claimed no personal experience (n = 44) became naïve forecasters, and those who reported at least one prior experience (n = 75) became experiencers. Of the women who indicated their personal sexism experiences at the end of the survey, those who indicated prior personal experience with both types of sexism (n = 18) became experienced forecasters, whereas those who indicated prior personal experience with neither or only one type of sexism (n = 13) were excluded from analyses (although retaining them did not change any of the significant effects or patterns reported below). Experienced forecasters indicated their personal experiences with sexism at the end of the survey because we did not want to prime thoughts of a specific sexism incident before they made their estimates.

Manipulating/Determining Sexism Type

The next section of the questionnaire contained detailed definitions and examples of benevolent and hostile sexism based on Glick and Fiske (1996). We counterbalanced the order in which we described benevolent and hostile sexism, but because order did not moderate any effects, we do not mention it further. Male forecasters, naïve forecasters, and experienced forecasters were instructed to imagine that they were a woman who had just experienced an episode of either benevolent or hostile sexism (determined by random assignment). In contrast, experiencers were prompted to think of “one specific occasion” when they experienced sexism, to describe the incident in detail, and to classify it as either benevolent (n = 35) or hostile (n = 40).

Affective Reactions

Forecasters then estimated the intensity of 12 different negative affective reactions that a woman would feel “immediately after experiencing the sexism” on scales of 1 (not at all) to 9 (very strongly). Experiencers instead recalled the intensity of their immediate affective reactions to the sexism episode that they had described.

We submitted affect ratings to a principal axis factor analysis with varimax rotation. Two factors emerged, with eigenvalues greater than two, that together accounted for 58% of the total variance. The first factor, which we labeled depression/fear (based on the two highest loading items) consisted of depressed, fearful, ashamed, guilty, embarrassed, doubtful of myself, and sad (all factor loadings > .60). The second factor, which we labeled anger/disgust (again based on the two highest loading items) consisted of angry, disgusted, hostile, resentful, and surprised (all factor loadings > .40). Both of these sets of items were internally consistent (αs > .87), so we averaged them to create composites.

Recovery Duration

Forecasters then estimated how many years, months, weeks, days, and/or hours they thought it would take to recover from the sexism (with “recover” defined as “no longer feeling intense, negative emotions; feeling resolved about things”). Using this same definition of “recover,” experiencers reported how many years, months, weeks, days, and/or hours it took them to recover from the sexist incident they identified. For both forecasters and experiencers, we created an index of recovery length by converting all estimates to days.

Individual Difference Scales

Experiencers then completed Pinel’s (1999) Stigma Consciousness Questionnaire for Women (SCQW), and four items modified from the identity subscale of Luhtanen and Crocker’s (1992) Collective Self-Esteem Scale (CSES). The SCQW is a 10-item scale that assesses women’s beliefs that men view and treat them in a sexist manner (e.g., “Most men have a problem viewing women as equals,” “My being female does not influence how men act with me” [reversed]). These items are rated on scales of 1 (strongly disagree) to 7 (strongly agree), and we averaged them to create an internally consistent (α = .77) index. The four items that we modified from the CSES assess the centrality of gender to women’s self-concept (“Being a woman is an important part of my self-image,” “Being a woman is not important to my sense of what kind of person I am” [reversed], “Being a woman is an important reflection of who I am,” and “Being a woman has very little to do with how I feel about myself” [reversed]). Experiencers rated these statements on the same 7-point scale that they used for the SCQW, and we averaged them (α = .65).

At this point, women forecasters who had not already described their personal experiences with sexism proceeded to do so. All participants then provided some demographic data and were thanked, offered information about free counseling services, and excused.

Results

Initial Intensity Bias

Our first hypothesis was that forecasters would overestimate experiencers’ initial affective reactions to hostile sexism and underestimate their initial affective reactions to benevolent sexism. To test this, we submitted the indices of anger/disgust and depression/fear to separate 4 (Role: male forecasters vs. naïve forecasters vs. experienced forecasters vs. experiencers) x 2 (Sexism type: benevolent vs. hostile) analyses of variance (ANOVAs).

The analysis on anger/disgust yielded the expected role-by-sexism interaction, F(3, 180) = 14.01, p < .001, but because inspection of the forecaster means revealed that all three groups of forecasters displayed identical patterns (see Table 1, columns 1–3), we collapsed across forecaster groups and re-ran the ANOVA as a 2 (Role: forecasters vs. experiencers) x 2 (Sexism type: benevolent vs. hostile) model. This analysis produced a role-by-sexism interaction, F(1, 184) = 40.25, p < .001, that qualified a main effect of sexism type, F(1, 184) = 13.98, p < .001. No main effect of role emerged, F < 1. As shown in Table 1 (columns 4–5), forecasters overestimated the intensity of experiencers’ angry and disgusted emotions after hostile sexism, F(1, 184) = 18.51, p < .001, whereas they underestimated how angry and disgusted experiencers felt after benevolent sexism, F(1, 184) = 21.83, p < .001. Looking at these data in another way, forecasters estimated much less anger in response to benevolent than hostile sexism, F(1, 184) = 63.20, p < .001, but experiencers of benevolent sexism reported marginally significantly more anger than experiencers of hostile sexism, F(1, 184) = 2.84, p = .09. Thus, forecasters’ responses revealed an intensity bias in their predictions about experiencers’ initial anger and disgust reactions to both types of sexism.

Table 1 Mean forecasted and recalled emotional impact of hostile and benevolent sexism.
Full size table

In the 4 x 2 ANOVA on depression/fear, the expected role-by-sexism interaction emerged again, F(3, 180) = 3.80, p < .02. However, all three groups of forecasters displayed nearly identical patterns (see Table 1, columns 1–3), so we combined them into a single group and conducted a 2 × 2 ANOVA. This yielded a role-by-sexism interaction, F(1, 184) = 10.43, p < .01, that qualified a main effect of sexism type, F(1, 184) = 14.14, p < .001. The effect of role was not significant, F(1, 184) = 2.57, p = .11. Consistent with our hypothesis, forecasters overestimated how depressed and fearful experiencers felt after hostile sexism, F(1, 184) = 11.52, p < .001 (see Table 1, columns 4–5). However, forecasters were fairly accurate in estimating experiencers’ depression and fear after benevolent sexism, F(1, 184) = 1.34, p = .25. Also, forecasters estimated that experiencers of benevolent sexism would feel much less depressed and fearful than experiencers of hostile sexism, F(1, 184) = 30.38, p < .001, but in fact, experiencers of both types of sexism reported very similar levels of these emotions, F < 1. In sum, when it came to depression and fear, forecasters committed an initial intensity bias regarding hostile, but not benevolent, sexism.

To examine directly whether prior experience with sexism increases people’s accuracy in predicting affective reactions to sexism, we also ran follow-up analyses that focused exclusively on experienced forecasters (n = 18) vs. experiencers (n = 75). The 2 x 2 ANOVA on anger/disgust yielded a role-by-sexism interaction, F(1, 89) = 18.85, p < .001, that qualified a main effect of sexism type, F(1, 89) = 8.91, p < .01, and no main effect of role, F < 1. Inspection of the interaction means (see Table 1, columns 3 and 5) revealed that experienced forecasters overestimated experiencers’ anger/disgust reactions to hostile sexism, F(1, 89) = 10.89, p < .01, and they underestimated experiencers’ anger/disgust reactions to benevolent sexism, F(1, 89) = 8.08, p < .01. Similarly, the ANOVA on depression/fear revealed a role-by-sexism interaction, F(1, 89) = 4.63, p < .04, a main effect of sexism type, F(1, 89) = 5.99, p < .02, and no effect of role, F < 1. This time, however, although the means were in the predicted directions, the simple effect of role did not reach significance in either the hostile or benevolent sexism conditions, Fs < 2.08, ps > .11. Thus, despite having personally experienced both types of sexism, our experienced forecasters still exhibited a clear initial intensity bias with regard to women’s anger/disgust reactions to ambivalent sexism. They also showed a non-significant trend toward an initial intensity bias with regard to women’s depressed and fearful reactions.

Duration Bias

Our second hypothesis was that forecasters would overestimate the duration of experiencers’ recovery from hostile sexism and underestimate the duration of their recovery from benevolent sexism. To test this, we first transformed the recovery length index, which ranged from .00 to 933.21 days with a standard deviation of 122.32 and a skew of 4.92. A square root transformation reduced the standard deviation to 5.20 and the skew to 3.23. We used the transformed variable in analyses, but display the non-transformed data in the text and table for presentation purposes.

We submitted the index of recovery length to the same 4 × 2 ANOVA described above. This analysis yielded a role-by-sexism interaction, F(3, 180) = 4.13, p < .01, but as shown in Table 1 (columns 1–3), all forecaster groups displayed a similar pattern. Therefore, we collapsed across forecaster groups and conducted a 2 × 2 ANOVA. In this model, neither role (F < 1) nor sexism type (F = 2.43, p = .12) reached significance, but the role-by-sexism interaction was significant, F(1, 184) = 10.92, p < .01. As predicted, forecasters overestimated how many days it would take experiencers to recover from hostile sexism, F(1, 184) = 6.24, p < .02, whereas they underestimated how many days it would take experiencers to recover from benevolent sexism, F(1, 184) = 4.72, p < .04 (see Table 1, columns 4–5). Moreover, forecasters predicted a significantly shorter recovery for victims of benevolent than hostile sexism, F(1, 184) = 14.70, p < .001, whereas experiencers of both types of sexism reported taking similarly long to recover from the incident, F(1, 184) = 1.27, p = .26. Thus, forecasters exhibited a duration bias, in that their estimates of recovery length were either inflated or deflated, depending on sexism type.

Next, we focused on the responses of only experienced forecasters and experiencers. The 2 × 2 ANOVA on recovery length yielded a role-by-sexism interaction, F(1, 89) = 4.88, p < .04, and no other effects, Fs < 1.68, ps > .19. Inspection of the interaction means (see Table 1, columns 3 and 5) revealed non-significant trends for experienced forecasters to overestimate women’s recovery duration following hostile sexism, F(1, 89) = 2.71, p = .10, and to underestimate women’s recovery duration following benevolent sexism, F(1, 89) = 2.19, p = .14. Again, prior experience with both types of sexism did not seem to increase women’s accuracy when making estimates about hypothetical cases of ambivalent sexism.

Mediation Analyses

Our third hypothesis was that mispredictions about initial intensity should mediate the link between participants’ status and their mispredictions about recovery duration. We tested this prediction using Preacher and Hayes’ (2004) bootstrapping method, which provides a non-parametric test of mediation that does not depend on assumptions of normality (which frequently are violated when using the more traditional Baron and Kenny (1986) approach to mediation). When the 95% confidence interval around a mediating variable’s point estimate does not contain zero, it means that the indirect path from the predictor variable to the outcome variable through the mediator is significant.

The independent variable in this model was a contrast code that represented the role-by-sexism interaction (forecaster/benevolent = -1, forecaster/hostile = + 1, experiencer/benevolent = + 1, experiencer/hostile = -1). Because we found stronger evidence of an initial intensity bias with regard to anger/disgust than depression/fear emotions, we treated the anger/disgust index as the only mediator in our model. The results of the bootstrapping method, with 5,000 resamples, confirmed the results of the ANOVAs reported above: The paths from the role-by-sexism interaction to both the mediator (anger/disgust) and the outcome variable (recovery length) were significant, ts > 3.60, ps < .001, as was the path from anger/disgust to recovery length, t = 5.21, p < .001. However, when anger/disgust was controlled for, the direct effect of the interaction term on recovery length dropped to non-significance, t = 1.11, p = .27. Finally, the indirect path from the interaction term to recovery length through anger/disgust was significant, as indicated by the fact that the 95% confidence interval (.5818 to 1.5334) around the point estimate (.9444) for anger/disgust did not contain zero. These results, which are summarized in Fig. 1, indicate full mediation by anger/disgust emotions, and suggest that forecasters demonstrated a duration bias regarding both hostile and benevolent sexism because they demonstrated an initial intensity bias with regard to experiencers’ anger– and disgust-related emotions following each type of sexism.

Fig. 1
figure 1

Results of analyses showing that estimates of angry, disgusted emotions mediate the association between the role-by-sexism interaction and estimates of recovery length. **p < .001

Full size image

Individual Differences

Given that experiencers self-generated the episode of sexism about which they answered all subsequent questions, it is possible that there were individual differences in the types of sexism that its victims noticed and/or remembered. Similarly, it is possible that individual differences predict the initial intensity of affective responses to, and the duration of recovery from, sexism. If so, this would limit the generalizability of our experiencer data to certain subsets of women. To explore this possibility, we conducted several regression analyses on the data of our 75 experiencers.

First, we conducted two simple regression analyses, predicting experiencers’ classification of their sexism episode as either benevolent or hostile from their scores on the SCQW and then the CSES. These analyses revealed no association between stigma consciousness or gender identity and tendency to recall and/or classify a personal sexism episode as either benevolent or hostile, ts < 1. Second, we conducted a series of six multiple regression analyses in which we predicted anger/disgust, depression/fear, and recovery length, separately, from (a) scores on the SCQW or CSES (after centering these on their means; Aiken and West 1991), (b) sexism type, and (c) the SCQW-by-sexism or CSES-by-sexism interaction term. In none of these analyses did the individual difference measure interact with sexism type to predict affective reactions or recovery length, all ts < 1.25, ps > .21. Main effects of stigma consciousness emerged in the analyses predicting anger/disgust (t = 1.98, p = .05) and depression/fear (t = 2.25, p < .03), suggesting that women higher in stigma consciousness felt stronger negative emotions following both types of sexism. However, there was no tendency for either stigma consciousness or gender identity to dictate the type of sexism that experiencers recalled, or to moderate the links between sexism type and affection reactions.

Discussion

When it comes to the emotional impact of ambivalent sexism, forecasters tend to assume that hostility is much worse than benevolence. According to experiencers of ambivalent sexism, however, forecasters’ intuitions are off-the-mark. Specifically, women who recalled real, personal incidents of hostile or benevolent sexism claimed to experience similar levels of anger– and depression-related emotions in response to the sexism, and they reported taking similarly long to recover (i.e., return to baseline affective states) following the incident. Thus, if we take experiencers’ accounts as an index of reality, it appears that most people — even those who have personally experienced both benevolent and hostile sexism — commit a forecasting error with regard to ambivalent sexism.

Moreover, we found evidence of two distinct but highly related types of forecasting errors. People committed an initial intensity bias by mispredicting the intensity of negative affect that women experience immediately following exposure to ambivalent sexism, and they committed a duration bias by mispredicting the length of time that it takes women to recover from sexism. The results of our mediation analyses established, further, that the initial intensity bias fuels the duration bias. It appears that people first consider the intensity of affect that experiencers feel upon exposure to sexism, and then use this estimate as a basis for gauging the duration of experiencers’ negative affect (e.g., Gilbert et al. 2004).

Why are people so wrong about the emotional impact of hostile and benevolent sexism? Several possible answers to this question are suggested in the affective forecasting literature. Consider first the tendency to overestimate the emotional impact of hostile sexism. One explanation for this error is that people are generally quite good at minimizing the emotional impact of unexpected negative events, through cognitive strategies such as rationalization, downward social comparisons, self-serving attributions, and the like. When estimating how bad a future negative event will make them or others feel, however, people fail to adjust their estimates to account for humans’ impressive cognitive coping abilities. The result is a chronic tendency to overestimate the intensity and duration of negative affective responses to unpleasant events (Gilbert et al. 1998). Thus, given how nasty hostile sexism seems, people may overestimate its impact because they do not realize how readily its victims will be able to minimize its implications for their mood and self-esteem.

Of course, this explanation cannot account for people’s tendency to underestimate the emotional impact of benevolent sexism. To explain why people’s negative affective reactions are sometimes worse than expected, Gilbert et al. (2004) proposed that experiencers of unpleasant events sometimes fail to engage their cognitive coping strategies when they should. If a given negative event does not cause much pain initially, then it may not trigger active coping efforts because people generally do not devote cognitive resources to coping unless they perceive a need to do so. Failure to cope, in turn, prolongs the duration of negative affect because no steps are taken to diminish one’s pain, however dull that pain may be. Thus, given that benevolent sexism seems relatively benign, people may underestimate its impact because they do not realize that its victims will fail to cope adequately with it.

We believe that experiencers of benevolent sexism may indeed have failed to cope adequately with their experience, but not for the same reasons that Gilbert et al. (2004) articulated. These researchers suspect that experiencers of negative events fail to cope when they do not feel enough psychological distress to trigger their psychological immune system. Clearly that was not the case with our experiencers of benevolent sexism; after all, they experienced just as much fear and depression and slightly more anger than did our experiencers of hostile sexism. Thus, if our experiencers of benevolent sexism failed to cope, it was not likely for lack of pain. Instead, it may have been because they hesitated to discuss their experiences out of an awareness that others might not understand the distress brought on by benevolent sexism. In short, their failure to cope may have been driven more by external, social factors than by internal factors.

We also suspect that part of the reason for the forecasting error concerning benevolent sexism lies in people’s failure to understand why sexism is so upsetting to its recipients in the first place. Regardless of whether sexism assumes a hostile or benevolent form, it always conveys (at least) two messages to its recipients. The first message is communicated by the emotional tone of the sexism, and it corresponds directly to sexism type: whereas hostile sexism conveys an angry, rude message, benevolent sexism conveys a kind-hearted, patronizing message. The second message, however, remains constant across sexism type — this is the message that, as a woman, the victim is a “lesser” sort of being than a man. We believe that it is this latter message that most likely explains women’s angry reactions to both hostile and benevolent sexism. However, when forecasting the emotional impact of sexism, people may focus instead on the first message that sexism conveys, and neglect the second message. That is, forecasters might wonder “How does it feel to be treated with anger (or with polite condescension)?” rather than “How does it feel to be treated as less than fully human?” If so, then it is no wonder that forecasters fail to appreciate the emotional impact of benevolent sexism. This possibility, of course, remains purely speculative given that our data do not allow us to test it directly. Thus, we believe that an important direction for future work involves uncovering the mechanism(s) that drive the forecasting errors documented here.

This brings us to some of the implications of this work for women’s well-being. As documented previously and illustrated here, exposure to sexism elicits negative affect among its recipients (e.g., LaFrance and Woodzicka 1998; Vescio et al. 2005). Moreover, some of our experiencers reported that their negative emotions took almost two months to dissipate following the sexist incident. For such women, active efforts to cope — by seeking social support or utilizing other emotion regulation strategies — would most likely yield benefits for well-being. Unfortunately, the prevalence of forecasting errors regarding ambivalent sexism suggests that experiencers may find it difficult to locate support providers who can empathize with their feelings. Depending on the particulars of the sexist incident, friends and family of the experiencer will likely expect her to feel either more or less upset, and to recover either more or less quickly, than she actually does. This state of affairs may diminish the quality of social support that experiencers of ambivalent sexism receive, as support providers tend to offer more useful assistance to the extent that they accurately perceive the emotional impact of a painful event and can empathize appropriately with the sufferer’s feelings (Lehman and Hemphill 1990; Trobst et al. 1994).

More specifically, if experiencers’ and support providers’ perceptions of the emotional impact of sexism do not match, then two types of unhelpful support may result. First, following hostile sexism, others may engage in maximization whereby they catastrophize the incident, express too much worry and concern, and/or behave in an overly protective manner toward the victim (e.g., Dakof and Taylor 1990; Lehman and Hemphill 1990). Second, following benevolent sexism, others may display minimization by responding as if the incident was “no big deal,” and/or conveying the belief that the victim is overreacting (e.g., Bosson et al. 2008; Dakof and Taylor 1990; Lehman et al. 1986; Lehman and Hemphill 1990; Pinel et al. 2009). From recipients’ perspective, both of these forms of support are unhelpful because they make people feel misunderstood. In turn, the receipt of unhelpful social support may discourage experiencers of sexism from seeking additional assistance, or otherwise airing their thoughts and feelings about the event.

Even experiencers who have not already received unhelpful social support may refrain from seeking others’ assistance in dealing with ambivalent sexism. Indeed, several theorists suggest that targets of stigma sometimes hesitate to seek support following episodes of discrimination because of the potential for being labeled “hypersensitive” or “a troublemaker” (e.g., Kaiser and Miller 2001; Stangor et al. 2002). This possibility is particularly relevant for sufferers of benevolent sexism. Given the prevalent tendency to assume that benevolent sexism is less upsetting than hostile sexism, experiencers of benevolent sexism may refrain from discussing their situation with others because they anticipate an unsympathetic response. As such, the forecasting errors that we documented here can potentially interfere, in several different ways, with women’s ability to cope effectively with sexism. We therefore view this as an important topic for future investigations.

Before closing, one final issue merits attention. Although we treated experiencers’ memories of their reactions to sexism as our index of “reality” in the current study, we acknowledge that this strategy may be problematic given that recall biases may distort people’s autobiographical memories for emotional experiences (e.g., Christianson and Safer 1996). To complicate matters, findings concerning the accuracy of people’s memory for their emotions is mixed, with some work indicating that people systematically overestimate the intensity of past negative emotions (e.g., Thomas and Diener 1990), and other work suggesting instead that people have fairly accurate memory for the intensity of past negative emotions (Levine 1997).

Despite our reliance on a less-than-ideal measurement strategy here, there are two factors that give us confidence in the validity of our experiencers’ memories. First, some research indicates that people are quite accurate at recalling the number of days on which they felt a given emotion (Brown et al. 2007), which at least lends credence to our duration data, if not our intensity data. Second, neither of the individual difference measures that we included moderated the link between sexism type and experiencers’ accounts of their emotional reactions. This suggests that people’s memories for hostile and benevolent sexism were not differentially affected by their chronic tendencies to assume sexist intentions on men’s part, or to view gender as a central component of the self. Nonetheless, our reliance on people’s memory constitutes a limitation of this study that, we hope, will be rectified in future research. One approach might involve manipulating both forecasters’ and experiencers’ exposure to an incident of hostile or benevolent sexism in the lab, and comparing forecasters’ predictions to experiencers’ affective reactions. Using a similar design, Kawakami et al. (2009) recently showed that White forecasters substantially overestimated White experiencers’ negative affective responses to racism toward a Black confederate. Applying this type of design to the topic of ambivalent sexism would be ideal, as it would reduce the problem of potential recall biases.

Summary and Conclusion

The findings presented here suggest that most people fail to appreciate the emotional impact of both hostile and benevolent sexism: Whereas women suffer less intensely and for a shorter period of time than others expect them to following hostile sexism, they suffer more intensely and for longer than others expect them to following benevolent sexism. This mismatch between women’s real experiences and forecasters’ assumptions may create problems for victims of sexism who wish to seek support and assistance from others. Important remaining questions include why people are so inaccurate when it comes to assessing the emotional impact of ambivalent sexism, and what, if anything, can be done to increase their accuracy.