Guilt and Empowerment
On influencing bias towards female leaders
Written by Kim Schuiten 10191089
MSc Student Political Theory
Graduate School of Social Sciences University of Amsterdam
Supervision by dr. G. Schumacher
Assistant Professor in Political Science
Social and Behavioural Sciences University of Amsterdam
Second assessor by dr. L. Mügge
Associate Professor in Political Science
Social and Behavioural Sciences University of Amsterdam
June 22, 2018 Abstract: Although women’s participation in the work and political sphere has greatly risen over the years, there is still a serious lack of women in national parliaments, corporate boards and CEO positions. In this thesis I will explore the existing stereotyping attitudes towards women, and how these relate to the misrepresentation of women in leadership positions. I have conducted an experiment in which I tried to manipulate both explicit and implicit stereotypes about women through the use of counter-stereotypic images and biographies. While I found that the counter-stereotypic images of women did not have any effect on implicit bias towards women, it did have a positive effect on women’s explicit attitude towards women. At the same time the treatments and the questions about the participants’ attitude towards women seemed to trigger a feeling of guilt among male participants which resulted in lower implicit bias towards women. However, too much awareness worked counter-effective for the male participants, it seemed to result in a ‘rebound effect’. It made them so aware of their prejudice towards women, which they wanted to suppress, that the prejudiced thoughts about women became hyperaccessible and they therefore showed more implicit bias towards women.
CONTENTS Introduction ... 3 Theoretical framework ... 5 ATTITUDES TOWARDS FEMALE LEADERS ... 5 IMPLICIT AND EXPLICIT ATTITUDES ... 8 Methods ... 12 PARTICIPANTS ... 13 MATERIALS ... 13 1. Treatment materials ... 13 2. IAT stimulus materials ... 14 PROCEDURE ... 15 INDEPENDENT VARIABLES ... 16 1. Demographics ... 16 2. Self-report measure. ... 16 3. Ideology ... 17 4. Previous done IAT’s ... 17 DATA PROCESSING ... 17 Analysis ... 18 REGRESSION MODELS ... 18 DISCUSSION ... 25 LIMITATIONS ... 29 Conclusion ... 30 References ... 32
3 “Despite gains in every profession, women remain underrepresented at all
levels of leadership. In Congress, on corporate boards, and in our nation’s colleges and universities, male leaders outnumber female leaders by considerable margins.” (AAUW, 2016)
INTRODUCTION
In this thesis I will explore possible ways to affect gender bias towards women in leadership roles through the use of priming images. Due to, amongst others, implicit attitudes about gender roles, women are less represented in political and business leadership roles (AAUW, 2016). Although women’s participation in the work and political sphere has greatly risen over the years, there is still a serious lack of women in national parliaments, corporate boards and CEO positions (Cataylst, 2018; United Nations, 2017). The amount of women in senior organizational ranks has even stagnated, and in some cases even declined (Ryan, Barreto and Schmitt, 2009; Davidson, Marilyn and Burke, Ronald, 2012; Vinnicombe Obe, Doldor and Turner, 2014). Not only is this an unfair distribution of high status jobs, it also leads to the devaluation of the daily life of all women, because when women are denied powerful leadership roles, they are denied the power to make a difference in the world (AAUW, 2016), and leaving important decisions about women and women’s bodies up to men.
Due to widely shared hegemonic cultural beliefs about gender roles, women are often connected with qualities like nurturing and caretaking, while men are connected with leadership qualities and tasks (Eagly and Mladinic, 1994). These beliefs are shared by both men and women and thereby have the self-fulfilling tendency to affect the behaviour of both sexes, leading to an unquestioned status quo (Ridgeway and Correll, 2004). Therefore it will not be enough to only change people’s explicit attitude towards the position of women, as these are often influenced by social standards and recent experiences. To have a full effect, there needs to be a change in the implicit bias that we have towards women as well, as implicit attitude is often an important predictor of people’s behaviour.
Moreover, a change in implicit attitude will be especially important for those who make up the leadership elite (mostly white highly-educated men; Celis and Erzeel, 2017) as they are an important aspect in deciding who becomes part of this elite. Due to quotas and the need to be more representative of society, women might have gained more high status jobs, or have become part of the political sphere. Notwithstanding, there remains a difference between the representation of women and actually being able to express the interest of women (Mügge and Erzeel, 2016; Celis and Mügge, 2018). As long as the current elite cannot see women as
having agentic qualities, they will include women, but they will not let them actually express their interests.
There are several ways to influence implicit attitude. You can, for example, motivate people to change their attitude; people are more likely to correct their implicit bias when they want to maintain a positive self-image or positive relation with others. Or you can use strategic measures such as the promotion of counter-stereotypes, or prime them with different contextual cues (Blair, Ma and Lenton, 2001; Leicht, Randsley de Moura and Crisp, 2014). In this thesis I have tried to influence both explicit and implicit attitudes towards women through an image-based experiment. I repeatedly placed images, combined with biographies about their proffesional succeses, of women in a leadership context to make participants connect agentic qualities (such as independence and assertiveness) as much to women as they do to men. Using images to affect people’s attidude is based on the idea of evaluative conditioning, in which you use priming materials (such as images, smells or words) to change someone’s attitude towards something (Houwer, Thomas and Baeyens, 2001; Olson and Fazio, 2001, 2002; Walther and Langer, 2008). By repeatedly paring women with a leadership context, I expected that participants would afterwards more easily connect women with words relating to leadership.
This thesis will start with an overview of the research on gender roles and how these affect the way we look at women and have caused a misrepresentation of women in leadership positions. I will explore the different ways of changing someone’s explicit and implicit attitude and how my experiment makes use of counterstereotypic images of women to apply the technique of evaluative conditioning. Thereafter I will continue with the experiment that I have conducted and the results that I have found. While the images of female leaders did not affect the implicit bias, it did affect the explicit attitude of female participants. Although the different treatments did not affect everyone’s implicit bias, I did find an interesting relationship between some of the priming materials (such as the treatments and the explicit attitude questions that were asked) and the implicit attitude of men, but not with women. It seems that men trigger a feeling of guilt when they are being primed with stereotypic images of women, or by questions about the position of women on the work floor. Other men seemed to suffer from a rebound effect after they had been primed too much. They were so aware of their prejudice towards women, which they wanted to suppress, that the prejudiced thoughts about women became hyperaccessible and therefore their implicit bias actually became worse.
5
THEORETICAL FRAMEWORK
Attitudes towards female leaders
There are several reasons why implicit attitudes about female leadership have led to the small amount of women among our leadership elites. Many of these reasons are related to implicit gender roles. Widely shared, hegemonic cultural beliefs about gender play an important role in our daily social relational contexts (Ridgeway and Correll, 2004). These social relational contexts are the contexts in which we define ourselves in relation to others, in order to comprehend the situation in which we find ourselves and form certain expectations.
The first and strongest classification that we make is the automatic sex categorization (Eagly and Carli, 2007: 88), which is based on hegemonic cultural beliefs about gender roles. These are shaped by the societal distribution (based on society’s division of labour and the psychological process of inferring traits from observations, to a perception of an entire group) of women into domestic roles and men into employee and leadership roles. The context in which we place a person or a group include expectations about what a group or a person is like (descriptive beliefs) and what a person or group should be like (prescriptive beliefs) (Eagly and Carli, 2007). This means that based on our experiences and observations (we have seen women working as nurses and teachers and men as CEO’s and fire-fighters) we develop expectations about men and women as a group. We connect the traits that were part of these observations (women we observed were caring and nurturing, while men were aggressive and dominant) to these groups as a whole, meaning that when we encounter someone from one of these groups, we expect them to have similar traits.
These expectations have led to automatic beliefs (beliefs that immediately come to mind when we encounter a man or a woman), such as that men are more status worthy and competent in agentic traits (such as assertiveness and aggression), while women are seen as less competent in such supposedly masculine domains, but they are nicer and better at nurturing and communal tasks (Eagly and Mladinic, 1994). Moreover, these expectations have the self-fulfilling tendency to affect people’s behaviour and evaluations. Such expectations can for example decrease the likelihood that a woman speaks up with confidence instead of waiting for someone else to act (Ridgeway and Correll, 2004) and have led to a more general confidence-gap between men and women (Simon and Hoyt, 2013; AAUW, 2016) because women have learned to see themselves according to those prescriptive beliefs; they are not supposed to be assertive and dominant.
Both the members of the privileged1 (men) and the unprivileged groups (women) have
come to share the thoughts, feelings and behaviours that justify the status quo, to the point that the legitimacy of these social arrangements is barely questioned (Devos, 2008). This explains, for example, why it took so long before women started to effectively stand for their right to vote, and it explains why women even today still do not feel that they belong in top positions.
Men are socialized to be more self-promoting and assertive because these are qualities that other men before them also had (they fit in with our prescriptive beliefs about men), while it is socially undesirable for a woman to have similar traits as these kind of traits go against our prescriptive beliefs about women, which delegitimizes the women that are reaching for, or are already holding, authoritative positions (Rudman and Kilianski, 2000; Heilman, 2001; Vial, Napier and Brescoll, 2016). Eagly and Karau have called this “role congruity”. This theory proposes that
“perceived incongruity between the female gender role and leadership roles leads to 2 forms of prejudice: (a) perceiving women less favorably than men as potential occupants of leadership roles and (b) evaluating behavior that fulfills the prescriptions of a leader role less favorably when it is enacted by a woman” (Eagly and Karau, 2002).
Positions that require a self-promoting and assertive attitude, such as authoritative positions, are therefore often deemed to be the best fit for men as they are the ones who we presume have these kinds of qualities. We perceive a ‘lack of fit’ (Heilman, 2001: 660) between the requirements of such an authoritative job and the stereotypic attributes that we ascribe to women. Thus, the greater the degree of stereotyping or the more ‘masculine’ we perceive a job to be, the worse the perceived fit will be and the higher the expectations of failure.
This, amongst others, leads to a tendency of women to structurally diminish and undervalue their professional skills and achievements, while men tend to overestimate their skills (AAUW, 2016). At the same time, others will view women as ill equipped to perform well on certain jobs and therefore diminish the successes of women by, for example, attributing the woman’s success to a male peer instead (Heilman, 2001).
7 This leads me to the following hypothesis:
Hypothesis 1: Men are automatically linked to agentic treats, while women are automatically linked to communal traits.
Not only are women viewed as incapable of having the qualities that are needed for leadership positions, and not only are women in leadership positions viewed as illegitimate or threatening to the position of men, these gender role stereotypes have also led to the structural devaluing of women’s self-confidence and decreasing aspirations, as I have explained above. Because men have held most of the leadership positions over time, we have come to automatically link the concept of leadership with masculine traits (Heilman, 2001). Moreover, media images often portray men and women in their traditional gender roles, confirming stereotypic expectations about women and therefore amplifying the lack of fit for women as political candidates, and influencing the way women perceive themselves. So when media depict female political candidates in stereotypical ways, not only voters’ perception of female candidates can be influenced, it also discourages women to enter politics (Simon and Hoyt, 2013; AAUW, 2016).
Women will also have to deal with implicit in-group favouritisms of men, which means that members of an group show an explicit preference for other members of their in-group. Especially members of high-status or privileged groups have the tendency to exhibit strong implicit in-group favouritism (Niven, 1998). Men can be seen as part of a privileged group, they are privileged in the sense that being male already gives them an advantage in acquiring high status jobs, and it can therefore be expected that they will show an implicit favouritism to other males (who are members of their in-group) because they will expect that people who are more like them, will perform better in jobs similar to theirs.
Multiple ways of overcoming these stereotype “threats” (being aware of the stereotypes about women can make women anxious about confirming these doubts, causing a lack of confidence; Eagly and Carli, 2007; Hoyt and Murphy, 2016) to women’s self-perception and aspiration to leadership have been researched. Among them is the role of female role-models (McGlone, Aronson and Kobrynowicz, 2006; Marx, Ko and Friedman, 2009; Dasgupta, 2011; Latu et al., 2013; Marx and Roman, 2016) or encountering counter-stereotypical depictions of women (Dasgupta and Asgari, 2004; Simon and Hoyt, 2013); when women see other successful female leaders or women who do counter-stereotypical things, they are more likely to think that they are capable of doing something similar. Another
way is to create an identity-safe environment (Carnes et al., 2005; Davies, Steele and Spencer, 2005; Burgess et al., 2012) or emphasize positive stereotypes by explicitly valuing feminine traits (Kray, Thompson and Galinsky, 2001; Kray, Galinsky and Thompson, 2002). This is to ensure that identifying with a woman should not be cause for anxiety. For example, by removing masculine traits from selection criteria in applications you emphasize that feminine qualities are equally important. Furthermore, early political socialization like parental encouragement, participation in competitive activities and creating a sense of self-confidence (Fox and Lawless, 2014) can help in preventing the development of stereotype threats in young women.
While it is important to counteract stereotype threats and to empower women to reach their full potential, I will argue that it is equally important to not hold back women who are pursuing leadership roles with stereotypical judgements. Role models have already proven to have some effect on reducing the bias towards the capabilities of female leaders, although this did not necessarily reflect a change in the explicit preference in both men and women towards women (Beaman et al., 2009). In the next section I will explore more ways through which encountering counter-stereotypical depictions of women can not only influence stereotype threats, but also influence other people’s stereotypical attitudes towards women.
Implicit and explicit attitudes
When discussing attitudes, we generally make a distinction between implicit and explicit attitudes, and you can not talk about implicit attitudes without talking about the highly influential article of Greenwald and Banaji (1995). When we talk about implicit attitudes, we talk about our unconscious attitudes, as opposed to our conscious and controlled explicit attitudes. These two attitudes do not necessarily overlap. Someone can, for example, claim that they “do not see colour” (this is their explicit attitude, it is the attitude they communicate to others), while their implicit attitude may prove differently when they encounter a person of a different ethnicity then their own. They are the reactions that occur without our awareness, control or intention (Devos, 2008). These reactions happen, for example, when we have to make a decision under pressure or based on a gut feeling. Implicit attitudes are also an important factor when trying to predict someone’s political preferences and behaviour (Gawronski, Galdi and Arcuri, 2015). This makes affecting implicit behaviour important when we are trying to influence people’s preference for female leaders.
To be able to influence one’s implicit attitudes, it is important to understand how they come in to being. While our explicit attitudes come from recent experiences (Rudman, Phelan
9 and Heppen, 2007), our implicit attitudes often reflect our introspective traces of past experiences and are developed through our environment’s socializing agents; we mimic the attitudes of our parents or friends for example (Devos, 2008). Already by the age of six, we have developed some of our implicit attitudes towards social groups. Especially an implicit favouritism toward one’s in-group will remain an important implicit attitude from then onwards, while our explicit attitudes towards people from the out-group might become more favourable when we grow older (Baron and Banaji, 2006). According to Rudman et al (2007), those early experiences need an emotional component to form our implicit attitude, as these attitudes stem from an associative learning system (through which you learn by remembering the relationship between two objects and concepts), which is more influenced by emotion, while our reflective learning system (through which you learn by reflecting on your actions and from which our explicit attitudes stem) is influenced by facts and accuracy.
If our implicit attitudes are already shaped by traces of past experiences, is it then not impossible to change them at a later stage? Although implicit attitudes might be even harder to change than explicit attitudes, it is possible. There are five general ways to influence
someone’s implicit attitudes2. If a respondent is motivated to maintain a positive self-image,
or maintain positive relations with certain others, they are more likely to alter their automatic attitudes. For example when a white respondent does an implicit attitude test on attitudes towards black people, they are more likely to show less automatic negativity towards black people when there is a black experimenter in the room because they feel motivated to maintain a positive self-image or positive relation with this experimenter (Brian S. Lowery, Hardin and Sinclair, 2001). Secondly, if strategic measures are used (such as the promotion of counter-stereotypes), it is possible to reduce the respondent’s stereotypes. Studies (Blair, Ma and Lenton, 2001; Leicht, Randsley de Moura and Crisp, 2014) found, for example, that when respondents were asked to use counter-stereotypical mental imagery of strong women, they were less likely to show automatic stereotypes towards women. Furthermore, if respondents are less focussed and are paying less attention to the priming images, they are less likely to show automatic stereotypes. The fourth measure to influence implicit attitude is based on contextual cues. Respondents will respond faster to Chinese stereotypical traits if they have seen a Chinese woman using chopsticks, while they will respond faster to stereotypic traits of women if they see the same woman putting on lipstick (Macrae, Bodenhausen and Milne, 1995). Finally, it is important to remember that the automatic reactions to an individual of a
group do not necessarily reflect the automatic reactions to everyone in that group. People will for example react more positively towards generally well-liked black people and more negatively towards disliked black people. It should also be mentioned that emotions and emotional experiences play an important role in the modification of implicit attitudes (Rudman, 2004; Rudman, Phelan and Heppen, 2007). This has for example been illustrated by the work of Matthes and Schmuck (2017) and Arendt et al. (2015) on negative stereotypical political advertisement about immigrants. They found that the advertisement posters that portrayed immigrants in a negative way by using highly emotional images had a big effect on people’s implicit and explicit biases towards immigrants.
From this overview I have a taken a couple of things that will benefit my experiment. First of all, it seems to be possible to affect implicit bias with the use of counter-stereotypes and providing contextual cues. Although a lot of the research on reducing implicit bias has been done on racial attitudes, I expect that the techniques apply to gender attitudes as well because both attitudes concern society’s marginalized groups and both concern implicit attitudes that are often different from the explicit attitudes. I therefore argue that using (counter-stereotypic) images of female leaders (in which the outfit, stance and a complimentary biography are the contextual cues) will lead to a reduced implicit bias towards the agentic qualities of women.
The findings on using counter-stereotypical measures to affect implicit attitudes, connects with other findings on how implicit attitudes can be redeveloped through repeatedly paring the attitude object (for example female leaders) with positive (or negative) stimuli (Devos, 2008). The term classical conditioning, evaluative conditioning to be more precise, comes into play here, as the effects of evaluative conditioning are most apparent in the area of persuasion (Walther et al., 2012). Evaluative conditioning is the process through which something (for example a picture), the conditioned stimulus, is repeatedly paired with an item that is subjectively liked or disliked, the unconditioned stimulus. The result of this process is a shift of the previously neutral conditioned stimulus to a point where it acquires more of the evaluative quality of the unconditioned stimulus (Houwer, Thomas and Baeyens, 2001; Olson and Fazio, 2001, 2002; Walther and Langer, 2008).
Examples of the process of evaluative conditioning are when a person (the conditioned stimulus) is associated with a disliked person (the unconditioned stimulus). He or she will then be judged in the same manner as the disliked person. Or when a person belongs to an in-group, he or she will be evaluated more positive compared to when that person belongs to an out-group. Moreover, people will show more liking towards a person that their friends like,
11 and they will dislike someone that their friends dislike (Walther et al., 2012). The new person is evaluated based on the unconditioned stimulus; the associations that come with the context in which this new person is placed. If that context is negative (my friends dislike this person, or this person looks like someone that I do not like), I will probably have negative opinions about this person. The conditioned stimulus would in the case of this thesis be a woman. I want women to acquire more of the evaluative qualities of a leader, so I repeatedly pare women within a context of leadership to cause a shift in how women are evaluated.
Research on implicit prejudice towards black people found that prejudice was lower when participants were exposed to well-liked black Americans and disliked white Americans (Dasgupta et al., 2001), when participants were shown images of black people in positive contexts (a family barbeque instead of a gang incident; Wittenbrink, Judd and Park, 2001), or when participants were showed pictures of black people in positive social roles (e.g. a lawyer in a prison context instead of a prisoner; Barden et al., 2004; Maddux et al., 2005). This suggests that seeing a person in a particular context can have an effect on someone’s associative evaluation of this person in a positive or negative way (depending on the context and the social role this person is playing in this context). Furthermore, it is also proven that repetition can enhance positive attitudes (Zajonc, 1968), especially if the perceiver already has previous (positive) knowledge about the message that is repeated and even more if the message contains a strong argument (e.g. Cacioppo and Petty, 1979). As I have argued before, I expect these constructs to work in a similar way for beliefs about gender. The images of women in a leadership context will therefore have more impact if it is repeated and if the perceiver already has some knowledge about female leaders and their competencies.
This leads me to the following hypothesis:
Hypothesis 2: Repeated counter-stereotypical images of women will lead to a decrease in stereotypical attitudes about women.
What is also interesting, is that while Beaman et al. (2009) found that exposure to female leaders in Indian villages with a female leadership quota did not necessarily alter their preference for male leaders, it did weaken the negative stereotypes they held about gender roles in the public and domestic spheres. They found that after the first time women were in office in these villages, these women were rated as less competent than their male colleagues. However, after having women as leaders for five years, the villagers already rated the female leaders at par with the male leaders. Moreover, research has also found that if someone
encounters more women in strong or leadership roles, the connection of women to leader becomes more accessible in their memory (Holland, Verplanken and van Knippenberg, 2003). This leads me to believe that exposure to counter-stereotypical women will not only lead to a decrease in stereotypical attitudes, this change in stereotype will eventually also lead to a more general shift in attitude towards female leaders.
METHODS
To explore whether counter-stereotypic images of women can affect the implicit and explicit attitudes towards women as leaders, I have conducted an experiment in which participants were shown images of female leaders, accompanied with a biography about their professional successes. I expected that participants who had seen the images of female leaders, accompanied with their biographies, would more easily connect agentic qualities to women (instead of communal qualities) compared to people who had not seen these images.
The experiment that I designed consisted out of four treatments with several manipulated factors (men or women, leadership or neutral roles) and participants were assigned randomly to one of these treatments. After the treatment, participants had to complete an implicit attitude test. Explicit attitude was measured through a series of questions, which they either had to answer before or after the treatment. Respondents also answered a couple of general questions about themselves.
I have conducted an experiment that included not only female participants, but also male participants to see whether the experiment had a different effect on men and women. Furthermore, because the experiment could be done online, the experiment was more accessible and therefore participants did not all have the same background. Moreover, I have chosen to use real images instead of mental images of hypothetical female leaders because I believe that an actual image is more likely to have an effect on people. Previous research has been done on the effects of images (or mental images; Blair, Ma and Lenton, 2001; Leicht, Randsley de Moura and Crisp, 2014) on implicit bias towards the agentic qualities of women. However, most of this research was done on the effects that these images have on women and how this influences their self-esteem or leadership aspirations (Dasgupta and Asgari, 2004; Davies, Steele and Spencer, 2005; Simon and Hoyt, 2013), or consisted only of participants of a psychology introductory course (Rudman and Kilianski, 2000). Therefore I expect to be able to say more about how people from different backgrounds react to this kind of stimuli
13 and whether it cause a change in attitude towards women instead of a change in attitude of women about themselves.
Participants
Participants were collected through Facebook and Snowballing the experiment through friends and family. Participants were offered to participate in a raffle for a €30 voucher for Bol.com. 182 participants completed the survey. Of these 182 respondents, 72 were male and 110 were female, 76% were highly educated (University or higher), 79% were younger than 30, and 81 participants grew up in an urban, 61 in a suburban, and 43 in a rural area. While the difference in age and education was not as expected (I expected that due to the accessible nature of the online experiment, I would have a more varied sample), there was a good spread in male and female participants and a good representation from urban and less urban locations. The experiment has been viewed and approved by the ethics commission of the University of Amsterdam.
Materials
1. Treatment materials
The first two treatments consisted of eight images of either male leaders or female leaders. All the images were complimented with a biography about this person’s
professional successes. The
biographies were the same for both the male and the female treatment, only the names were changed. The
other two treatments were the control treatments. In these treatments the respondents were also shown images of either men or women, but this time the images were more neutral and accompanied with a less successful biography. All the images were pre-tested by a group consisting of seven people who judged the people in the images on possessing either agentic or communal qualities.
The images of the priming treatments consisted of both real, but relatively unknown, people in leadership roles (such as campaign photos of the candidates of the 2015 Canadian
Figure 1: Example image and biography from the female leadership
federal election) and models depicting a man or a woman in a leadership situation. The images of the control treatments consisted of portraits of people in more neutral clothing in more neutral settings. I have opted to only use images of white men and women, instead of men and women from different ethnicities. I have tried to isolate the female aspect of prejudice, and adding ethnicity would have complicated the question as to what had the most influence on participants; the gender or the ethnicity of the people depicted in the pictures.
2. IAT stimulus materials
Respondents were then asked to complete an Implicit Attitude Test (IAT; A. G. Greenwald, McGhee, and Schwartz, 1998). I used the standard seven-block procedure for the design of the IAT. In the IAT, respondents would first see a window with the word “Men” in one corner and the word “Women” in the other. They would then have to assign the male names (e.g. Steven, Paul or Max) that appeared in the middle of the window to the category Men and the female names (e.g. Josephine, Maria or Kate) to the Women category.
In the next step the categories were changed to “Communal” and “Competitive” with the corresponding words individualistic, competitive, independent, challenging, self-sufficient, autonomous and hierarchical for Competitive and communal, connected, commitment, together, kinship, supportive and interdependent for Communal. These stimuli have been used in other studies and were therefore already tested (website Project Implicit; website Measure my Mindbug; Rudman and Kilianski, 2000). I have chosen the word Competitive instead of Agentic as one of the categories because I found during my own pre-testing that to many people it was unclear what the word agentic exactly meant. In the final stage the two categories were combined (e.g. Men and Communal in one corner and Women and Competitive in the other) and respondents were again asked to assign the words in the middle to the right category.
The respondents’ speed of assigning the words to the right categories was measured in milliseconds. It is expected that respondents with a higher stereotypical bias towards women will answer slower when asked to connect the Competitive words to Women, because this combination is not as mentally accessible as the stereotypical combination between Communal words and Women (Eagly and Mladinic, 1994; Rudman and Kilianski, 2000). The IAT was implemented into Qualtrics using iatgen (Carpenter et al., 2017).
15 Procedure
The experiment could be completed by everyone with a laptop or computer, through an anonymous link. Through this method it was easier to reach a wider audience, but it made it impossible to keep an account of who took the test, whether they did the test twice or not and the conditions (such as the amount of distraction in their environment) in which the experiment was taken could also not be influenced.
Respondents were informed before the experiment started, through a consent form, that they participated in an experiment that researched how leadership preferences are formed and that they would see a couple of images. Before they started the experiment, they were asked a few questions concerning their age, their highest achieved education, their current country of residence and whether they grew up in an urban, suburban or rural area.
They were then instructed to pay close attention to the series of images they were about to see, and to remember as much of the image and biography as possible. It was not possible to go to the next image before a minimum time of 8 seconds had passed, so people were forced to spend at least 8 seconds to look at the image and read the biography.
After this task, respondents were asked to complete the IAT. The IAT consisted of seven blocks of twenty or forty trials per block (a trial is each time the participant had to connect a word to one of the
categories). The first three blocks are practice blocks consisting of twenty trials each, then there is a critical block of forty trials, followed by another practice block of forty trials. Finally, the participant completes another practice block of twenty trials, but this time with the categories switched up, followed by a critical block of forty trials.
After the completion of the
IAT and the questions on explicit attitudes towards women in leadership roles, respondents were asked about their political identity on social and economic issues and their gender before they finished the experiment.
Figure 2: Screenshot of the IAT of the fourth block, where the
Independent variables
Beside the image treatment as the main independent variable to measure whether certain depicted gender roles have an effect on implicit attitude, I used a few other dependent variables that might affect response time on the IAT.
1. Demographics
To assess whether the type of environment in which someone grows up has effect on implicit attitude towards women, participants were asked a couple of demographic questions. I wanted to know whether the participant had grown up in an urban, suburban or rural area to find out whether growing up in a rural area has the same effect on prejudice towards women as it has on prejudice towards black people (Middleton, 1976; Tuch, 1987). Furthermore I have asked about the respondents’ age and highest achieved education and I expected older and lower educated participants to be more prejudiced towards women (Meertens and Pettigrew, 1997; Garcia-retamero and López-zafra, 2006; Gonsalkorale, Sherman and Christoph, 2009; Hodson and Busseri, 2012). Finally, I asked the participants whether they were male or female, and I expected women to be less prejudiced towards women (because they are part of their in-group) than men.
2. Self-report measure.
The respondents were also asked a couple of questions regarding their attitude towards women as (potential) leaders. The questions were based on the IAT of the Measure My Mindbug website (http://www.measuremymindbug.nl/) and the research of Islam and Nasira on attitude towards female leadership (Islam and Nasira, 2016). Respondents were randomly assigned to a test that contained these questions before or after the treatment. This was to see whether the treatment also had an effect on the explicit attitudes. Respondents could express their agreement on 16 statements on a scale ranging from 1 (strongly agree) to 7 (strongly disagree) on statements such as “Men and women are equally able to be a successful leader”. The respondents’ agreement to the items was averaged after reverse-scoring the appropriate items. A higher average score on the explicit attitude questions indicate a higher explicit bias towards women as leaders.
17 3. Ideology
As someone’s political ideology might have an effect on how they view the position of women (Meertens and Pettigrew, 1997; Duckitt, 2001; Sibley, Wilson and Duckitt, 2007; Duckitt and Sibley, 2010; Hodson and Busseri, 2012; Douglas and Sutton, 2014; Hodson, Macinnis and Busseri, 2017), I have also checked for political ideology on a 10-point scale. I have divided political ideology over political views on cultural issues and political views on economic issues because political identity is often not merely a simple left – right scale. Political preferences are usually based on both a cultural and economic level. Both questions could be answered by 0 being extremely liberal to 10 being extremely conservative.
4. Previous done IAT’s
To check for the possibility that someone becomes “better” at Implicit Attitude Tests, I asked whether the respondent had done 0, 1, 2 or 3 or more IAT’s before in their life (Greenwald and Nosek, 2001; Brian A. Nosek, Anothony G. Greenwald and Mahzarin R. Banaji, 2007). This is based on the assumption that someone might have a faster response time if they have done multiple IAT’s before and understand how the test works. Their reaction time might be shorter as they get “the hang of it”, like someone becoming better and faster in a video game.
Data processing
For the data cleaning procedures of the IAT results I followed the data analysis script provided by iatgen (Carpenter et al., 2017) for data analysis in R. Each trial, returned three results: a number to identify which stimulus was displayed, whether the participant responded correctly (“C”) or incorrectly (“X”) and their reaction time in ms. Individual trials over 10,000 ms are deleted, as well as data of which more than 10% of the responses is faster than 300 ms because this would indicate that participants did not complete the IAT sincerely; they were either being too cautious in their response or were simply ‘button smashing’ to get through the IAT quickly.
The D-score of the IAT of an individual is then calculated by taking the response time means of the different blocks (one for the practice blocks and one for the critical blocks). These two means are divided by inclusive standard deviations (i.e., SD of “block 3 merged with 6” and SD of “block 4 merged with 7”) and are then averaged. This average is the individual’s D-score and it indicates in which condition (women with communal words or women with competitive words) the participant was faster. To calculate the explicit attitude score, the mean of the scores on each explicit attitude question was taken per participant.
These means were rounded off to the closest whole number for some of the data visualizations.
ANALYSIS
To assess whether the treatment had a significant effect on the respondent’s score on the Implicit Attitude Test, I compared their D-scores. The D-score of the IAT of an individual is calculated by taking the response time means of the different blocks (one for the practice blocks and one for the critical blocks). These two means are divided by inclusive standard deviations (i.e., SD of “block 3 merged with 6” and SD of “block 4 merged with 7”) and are then averaged. This average is the individual’s D-score (Carpenter et al., 2017) and it indicates in which condition he or she was faster. A positive D-score means that the individual was faster in the blocks where the pairings were compatible with his or her mental association and he or she would therefore have a stronger implicit bias towards women, and thus a negative D-score means that the individual has a less strong implicit bias towards women (and a D-score of 0 indicates that there is no association at all). I therefore expected the respondents with the female leadership treatment to have lower D-scores than the other respondents.
Regression models
When looking at the means of the IAT D-scores after the different treatments we can already see an interesting difference (see also figure 1). While the differences in mean for the whole sample do not differ much (Mfemale leader = 0.294 with an sd of 0.325; Mfemale neutral = 0.246 with an
sd of 0.277; Mmale leader = 0.217 with an sd of 0.324 and Mmale neutral = 0.295 with an sd of 0.349),
the differences in mean for just the males in the sample do seem to vary (Mfemale leader = 0.294
with an sd of 0.259,; Mfemale neutral = 0.159 with an sd of 0.264; Mmale leader = 0.184 with an sd of
0.310 and Mmale neutral = 0.348 with an sd of 0.386). Furthermore, participants scored on average
a 3.2 out of 7 (sd = 0.62) on explicit bias, an average of 3.5 out of 10 (sd = 2.3) on cultural ideology and an average of 4.1 out of 10 (sd = 2.2) on economic ideology.
19 A multiple linear regression was calculated to predict the IAT D-scores based on whether the respondent had the treatment with the female leader, female neutral or male leader pictures and biographies (see table 1, model 1). Based on the F-test I rejected the model and I therefore calculated the same model per gender. Based on these F-tests both models were again rejected, however, the female neutral treatment was significant for the model with just male participants (see table 1, model 3), with p < 0.1 and a coefficient of
-0.189. The whole model has an R2
of 0.063 (F(3, 68) = 1.516, p > 0.1). Male participans who received the female neutral treatment therefore had a predicted D-score equal to 0.348 − 0.189 (TREATMENT), where treatment is coded as 1 = received the female neutral treatment and 0 = did not receive the female neutral treatment.
Table 1: Regression Results
Dependent variable: IAT Score
(1) (2) (3) (4)
Female Leader Treatment 0.037 0.028 0.054 0.080 (0.066) (0.069) (0.105) (0.108) Female Neutral Treatment 0.049 0.054 0.189⇤ 0.234⇤⇤
(0.070) (0.070) (0.104) (0.105) Male Leader Treatment 0.078 0.066 0.165 0.247⇤⇤
(0.068) (0.070) (0.104) (0.110) Age 0.049 0.028 (0.031) (0.041) Education 0.001 0.016 (0.033) (0.046) Gender 0.006 (0.052) Area of Descent 0.007 0.017 (0.032) (0.052)
Previous number of IAT’s 0.015 0.027
(0.022) (0.034) Explicit Attitude 0.042⇤⇤ 0.022 (0.020) (0.032) Cultural Ideology 0.018 0.052⇤⇤ (0.015) (0.021) Economic Ideology 0.005 0.028 (0.015) (0.021)
Explicit Attitude Treatment 0.004 0.045
(0.049) (0.079) Constant 0.295⇤⇤⇤ 0.262 0.348⇤⇤⇤ 0.352 (0.049) (0.231) (0.076) (0.341) Observations 182 182 72 72 R2 0.007 0.065 0.063 0.199 Adjusted R2 -0.009 -0.002 0.021 0.052 Note: ⇤p<0.1;⇤⇤p<0.05;⇤⇤⇤p<0.01 1
I then calculated a linear regression including all the other independent variables. The model that included the whole sample (model 2) was rejected based on the F-test (F(12, 169) = 0.632, p > 0.01), and it had an R2
of 0.065. However, the explicit attitude score is significant (p < 0.05) in this model, with a coefficient of -0.042. The effect between Explicit Attitude and the IAT D-score was negative, meaning that participants who had a higher explicit bias towards women, had a lower D-score and were therefore less implicitly biased. Although the variable Age was not significant as a whole, the age group of 46 to 60 years old was significant (p < 0.05) with a coefficient of 0.202. This means that participants between the age of 46 and 60 had a higher implicit bias compared to participants younger than 46.
When applying the same variables from model 2 to the male population of the sample there was again no significance (F(11, 60) = 1.318, p > 0.1) but it had an R2
of 0.199. Moreover, the female neutral treatment, the male leader treatment and Cultural Ideology were all siginificant (p < 0.05) with coeffecients of −0.234, −0.247 and −0.052 resepctively. All three variables have a negative effect and Cultural Ideology is based on a scale of 1 to 10 where 1 is extremely liberal and 10 is extremely conservative on cultural issues. Therefore respondents who received the female neutral or male leader treatment and the respondents who were more conservative had lower D-scores, and therefore have less of an implicit bias. I applied the significant variables in a separate model (see table 2), omitting all the unimportant
Table 2: Regression Results
Dependent variable: IAT Score
Female Leader Treatment 0.029
(0.102) Female Neutral Treatment 0.228⇤⇤
(0.101)
Male Leader Treatment 0.212⇤⇤
(0.102) Cultural Ideology 0.039⇤⇤ (0.015) Constant 0.527⇤⇤⇤ (0.102) Observations 72 R2 0.145 Adjusted R2 0.094 Note: ⇤p<0.1; ⇤⇤p<0.05; ⇤⇤⇤p<0.01 2
21 variables and also omitting the female participants. This time, a significant regression
equation was found (F(4, 67) = 2.848, p < 0.05) with an R2
of 0.145.
Based on the significant variables from the regression models I visualized some of the different mean scores on the IAT, again based on the IAT D-scores. As you can see in figure 3, the variety in D-scores is much bigger for the men, compared to the women. The men had a mean of 0.348 (sd = 0.386) after viewing the male neutral treatment and a mean of 0.159 (sd = 0.264) after viewing the female neutral treatment. This means that in the male control group men were quicker in attributing agentic qualities to men and communal qualities to women then when they had seen neutral images of women. It thefore seems that the first null hypothesis (men and women are equally linked to agentic and communal traits) can be rejected for the male participants in the male control group, but not in the female control group. Furthermore it seems that the female leader treatment (M = 0.294, sd = 0.259) had little to no positive effect compared to the control groups (and it also had no significance in the regression models) and therefore the second hypothesis (repeated counter-stereotypical images of women will lead to a decrease in stereotypical attitudes about women) seems to be rejected. 0.0 0.1 0.2 0.3 0.4 0.5
Female leader Female neutral Male leader Male neutral
Treatment D -s cor e Gender Male Female
Figure 3: D-score after each treatment, per gender
Moreover, it is interesing to see that the male leader treatment seemed to have a big postitive effect on the men (M = 0.348, sd = 0.386), while it did not do a lot for the female participants (M = 0.262, sd = 0.328) compared to the control group. The difference in effect of the female neutral treatment between men and women is also interesting. While the female neutral treatment had a rather positive effect on the male participants (M = 0.159, sd = 0.264), for the female participants it actually caused a bit more implicit bias towards women (M = 0.318, sd = 0.272).
I was also curious whether having to answer the explicit attitude questions before doing the IAT would have an effect on the IAT results of the men3 as well. Interestingly
enough, this actually had a rather big influence (see figure 4). As you can see it did not really matter whether male participants had to complete the questions on their explicit attitude towards women as leaders before or after the treatment and the IAT, when they were shown images of male or female leaders. It did, however, have a big impact when they were shown the control groups; the neutral female treatment (M = 0.259; sd = 0.225 versus M = 0.048; sd = 0.271) or the neutral male treatment (M = 0.198; sd = 0.446 versus M = 0.499; sd = 0.260). This means that male participants in the female neutral treatment group had almost no implicit
3 The difference in mean was not significant enough when checking for only females or for the whole sample.
0.0 0.2 0.4 0.6
Female leader Female neutral Male leader Male neutral
Treatment D -s cor e Explicit attitude before IAT Explicit attitude after IAT
Figure 4: D-score after explicit attitude was asked before or after the IAT (male participants)
23 bias towards women when they had to complete the explicit attitude questions after the IAT, while the male particpants in the male neutral treatment group showed substantially more implicit bias towards women when they had to complete the explicit attitude questions after the IAT.
I then compared the scores on the explicit attitude with the IAT scores (see figure 5), where 0 meant participants had no explicit bias towards women at all, and 7 signalled a lot of explicit bias towards women. I did this for the whole sample because the sample with only males was to small to say anything sensible about it (and also because the explicit attitude variable was only significant in model 2 of the regression analyses, which included the whole sample). Contrary to expectation, people who scored very low on the explicit attitude (0 out of 7), scored relatively high on implicit atttide (M = 0.230, sd = 0.326). However, there was a significant difference in this group between people who completed the explicit attitude questions before the IAT (M = 0.405, sd = 0.135) and the people who completed the explicit attitude questions after the IAT (M = 0.110, sd = 0.366). Furthermore, the negative relation between explicit attitude and implicit attitude can clearly be seen in this figure. It is remarkable how participants who scored a 4 on the explicit attitude, scored very low on implicit bias (M = 0.009, sd = 0.430), with participants who completed the explicit attitude questions before the IAT scoring significantly lower (M = -0.077, sd = 0.478) than their
-0.2 0.0 0.2 0.4 0.6 0 1 2 3 4 5
Explicit attitude score
D -s cor e Explicit attitude before IAT Explicit attitude after IAT
Figure 5: Implicit attitude versus explicit attitude score, after explicit attitude was asked before or after the IAT
24 counterparts who completed the explicit attitude questions after the IAT (M = 0.241, sd = 0.117). Participants who had relatively high explicit attitude scores (5 out of 7), in the line of expectancy, also had higher implicit attitude scores (M = 0.239, sd = 0.381).
Finally, I wanted to see whether the treatments had an effect on the explicit attitude (see figure 6). As you can see, both the female leader (M = 2.794, sd = 1.520 versus M = 2.219, sd = 1.148) and the male leader (M = 1.972, sd = 1.140 versus M = 2.571, sd = 1.259) had a significant effect on the whole sample. What is also interesting, is how the female leader treatment had a rather large effect on female participants (M = 2.790, sd = 1.638 versus M = 1.985, sd = 1.132), while the male leader treatment barely had an effect on the female participants (M = 1.979, sd = 1.108 versus M = 2.200, sd = 1.240). However, for the male participants it was the other way around. The female leader treatment barely had any effect on their explicit attitude (M = 2.804, sd = 1.306 versus M = 2.579, sd = 1.127), while the male leader treatment had a rather large effect (M = 1.958, sd = 1.313 versus M = 3.000, sd = 1.185) for the male participants4.
4 These numbers are not shown in figure 4. For further clarification see table 6.1 and 6.2 in appendix 6
0 1 2 3
Female leader Female neutral Male leader Male neutral
Treatment E x p li ci t sc or e Explicit attitude before IAT Explicit attitude after IAT
25 Discussion
There are a couple of interesting effects that have come to the surface in this experiment; some of them are contradicting each other. To start with the insignificant variables. It is possible that the variables Age and Education were not significant as a whole because there was not enough variety among participants. Gender was also not significant in the whole model but, as described below, it was important in some of the isolated effects. Contradictory to expectation, Are of Descent was not significant at all and this means that, in this study, whether you grew up in a rural or urban environment has no significant effect on your implicit bias towards women. The insignificance of the Previous number of IAT’s tells me that participants who have done multiple IAT’s before in their life are not necessarily faster in responding to IAT’s than their counterparts who have never done an IAT before. This could, however, also depend on how much time there was between the IAT they did in my experiment and the last IAT they had done before that. I have not checked for this. Finally, Economic Ideology was not significant, while Cultural Ideology was. This would mean that prejudice relates to one’s position towards cultural issues and not so much to one’s position on economic issues.
Now, on to the significant variables. First of all, men who saw the images of neutral women were more likely to show less of an implicit bias towards women as agentic people, compared to the men who saw the images of neutral men. This means that the first null hypothesis (men and women are equally assigned to agentic and communal qualities) is rejected, albeit only for men, and only in the male control group (neutral male treatment), as these participants showed a stronger implicit bias compared to the participants in the other treatments. It also shows that merely encountering women can already have an effect on men’s implicit bias. It might be that the effect of maintaining a positive self-image comes into play here. Like the white respondents who had to do an implicit attitude test with a black experimenter in the room (Brian S Lowery, Hardin and Sinclair, 2001), the men who were confronted with images of women might have felt an internal obligation to maintain a positive image towards women during the test. This does not, however, explain why there was not a similar effect with the images of female leaders.
As can be seen in figure 3, the different treatments had less of an effect on women than it did on men. One explanation could be that women are harder to convince about a change in the status quo. By being constantly reminded about the need to have communal traits and by society emphasizing these communal traits in them, women might have
internalized these gender roles, which in turn shows in their implicit attitude results (also see Garcia-retamero and López-zafra, 2006, who found a similar result). Thereby, images of neutral women will only confirm these gender roles for them, which might explain why the female neutral treatment had such a negative effect on the female participants, compared to the other treatments. This relates to the role congruity theory (Eagly and Karau, 2002) and the research on stereotype threat (Eagly and Carli, 2007; Hoyt and Murphy, 2016); when women encounter other women that are stereotype confirming, they are more likely to think that women are less favorable as leadership candidates.
The male participants possibly felt guilt toward their out-group (women) after they had seen the male leaders. As they are men themselves, the images of male leaders might have highlighted the fact that their in-group is priviliged, and this could have evoked guilt in the male participants. This is in line with existing research on the mediating effect of guilt and the presence of cues that create awareness of the possibility of biased responding (Dovidio, Kawakami and Gaertner, 2000; Devine et al., 2002; Monteith et al., 2002; Hofmann et al., 2005). The mediating effect of cues on gender bias are probably also the reason why male participants in the male neutral treatment who completed the explicit attitude questions before the IAT were much less implicitly biased towards women than the male participants who completed the explicit questions afterwards (see figure 4). Guilt as a mediating factor in implicit bias also confirms the research on how emotion mediates implicit bias (Rudman, 2004; Rudman, Phelan and Heppen, 2007).
The fact that male participants who were in the female neutral treatment actually showed less implicit bias when they did the explicit attitude questions after the IAT completely contradicts the previous story. The male participants in the female neutral treatment who had already done the explicit attitude questions before the IAT were maybe aware of their bias but the neutral images of the women might have given them a new reason to justify their sexist attitude. This might hold especially if said participants were aversive
sexists5, who will try not to appear prejudiced to others, but are more likely to show prejudice
when they can justify or rationalize their prejudiced responses (Dovidio, Kawakami and Gaertner, 2000).
However, this still does not explain why the men in the female neutral showed so little implicit bias when they were asked the explicit attitude questions after the IAT. It is therefore
5 Aversive sexists are people who claim not to be sexist, due to social pressure, but when they are tested on their
implicit bias they show a stereotypical prejudice towards women (Dovidio, Kawakami and Gaertner, 2000; Son Hing et al., 2008).
27 also possible that the men in the female neutral treatment who answered the explicit attidue question before the IAT, were suffering from a “rebound effect” (Dovidio, Kawakami and Gaertner, 2000). This effect occurs when people attemt to suppress stereotypes (for example because they were just made aware of their explicit prejudice toward women and they then also saw neutral images of women), but through that process cause a hyperaccessibility of the prejudiced thoughts is cause for higher level of implicit bias.
Another interesting find is the negative effect between explicit attitude and implicit bias for the whole sample, and cultural ideology and implicit bias for the male participants, which goes against the research that is done on the relation between explicit attitude and implicit attitude (Brauer, Wasel and Niedenthal, 2000; Cunningham, Nezlek and Banaji, 2004; Hofmann et al., 2005; Son Hing et al., 2008). Although it is known that aversive sexists show a low explicit bias while they still have a relatively high implicit bias, research has not found that aversive sexists (or aversive racists) actually scored lower on the implicit bias test compared to the blatant sexists or racists6 (Devine et al., 2002; Hofmann et al., 2005; Son Hing et al., 2008).
I looked back at figure 5 to try to make sense of this negative relationship. The first thing that was noticable, was the difference in D-scores between participants who had done the explicit attitude questions before the IAT, and those who did those questions after the IAT. It is possible that those who started with the explicit attitude questions were more inclined to answer the questions according to social standards. They therefore might have scored very low on explicit attitude, but were not able to respond according to social standards during the IAT and therefore had relatively high implicit scores. This would be in line with research on aversive sexism (Son Hing et al., 2008). This is confirmed by the data represented in figure 6. Participants (especially men) who started with the explicit attitude questions before they had the male leader treatment scored relatively low, while participants who completed the explicit attitude questions after this treatment showed more of an explicit bias towards women. It is interesting how this effect was mostly detected among the male participants and was almost nonexistent among the female participants. This might mean that men are initially more inclined to answer according to social standards, but will answer more according to their implicit attitude when they are able to justify their bias (because they have just seen images that confirm the idea that men are better leaders than women).
6 Blatant sexists should score high on both explicit as implicit bias (Cunningham, Nezlek and Banaji, 2004; Son
The negative trend of the group who had the explicit attitude question first (see figure 5) might be explained by the factor of guilt and a reaction of compensation. When participants scored higher (up until 4 out of 7) on the explicit attitude questions, they realized how much bias they actually have towards women and they might have felt guilty about that, resulting in lower IAT scores. Meanwhile, participants who scored the highest on the explicit attitude
questions (5 out of 7) might represent the modern sexists7 who would score relatively high on
the implicit attitude, especially when their female stereotypical ideas were triggered (Son Hing et al., 2008).
Particpants who started with the IAT might be more representful of people who are truly unbiased towards women because they have not been been primed with the explicit attitude questions yet. As you can see in figure 5, of the group who completed the explicit attitude questions after the IAT, those who scored the lowest on implicit bias were also the ones who scored the lowest on the explicit attitude (0 out of 7). This would be in line with research on the regulation of explicit and implicit bias (Devine et al., 2002). People who are genuinely unbiased towards women should also score low on the implicit atittude test (Son Hing et al., 2008). Meanwhile, aversive sexists might score relatively low on the explicit attitude (1 or 2 out of 7), but still have relatively high IAT scores. This can be explained by the multistep process of prejudice reduction (Devine et al., 2002). To reduce prejudice, you first have to consciously decide that responding in a biased way is inappropriate for this day and age, then you have to adopt the non-prejudiced beliefs and standard before these standards can be internalized and become part of your identity. Thereby, some of the participants who scored relatively high on the IAT might have realized that they were not performing well on the IAT and tried to compensate on the explicit attitude questions afterwards.
The negative relation between cultural ideology and implicit bias (if participants were more liberal, they were more implicitly biased) among the male population in the sample might have to do with the D-score outliers from participants who scored on the extreme sides of the cultural ideology scale. As there were only 72 male participants, these outliers may
7 Borrowed from Son Hing et al’s work on a two-dimensional model that employs explicit and implicit attitudes
to characterize prejudice, modern racists “believe that discrimination no longer exists and that Blacks are making too many demands that upset the status quo (2008: 972).” I expect that modern sexists think in a similar way: women are making too many demands that upset the status quo.
29 have skewed the result without actually saying something about the general effect between the two variables.
Then there is the question why the images of the female leaders did not have the predicted effect on implicit attitudes. The fact that the counter-stereotypical images of female leaders did not have any effect on both the male and the female participants, goes directly against the research that did find a positive correlation between counter-stereotypic images and performance on the implicit attitude test (see for example Blair, Ma, and Lenton 2001; Leicht, Randsley de Moura, and Crisp 2014). The images of the female leaders even had the smallest effect of all the treatments. It is possible that the images of female leaders went against the presecritpive beliefs about what a woman should look like or act like and therefore invoked feelings of disaprovement, which would also be in line with incongruity theory (Eagly and Karau, 2002). Nevertheless, the female leader treatment did have an effect on the explicit attitudes of women. While the women had a relatively high score in explicit bias if they completed the explicit attitude questions before the female leader treatment, there explicit bias was much lower when they completed the explicit attitude questions after the female leader treatment. This actually confirms that explicit attitudes come from recent experiences (Rudman, Phelan and Heppen, 2007) while implicit attitudes take longer to form.
Limitations
This study was in a way rather limited, mostly due to time and budget constrains. It would have been interesting to see whether the images have an effect over a longer period of time, which could be done by having participants complete the same IAT a few weeks after the first time. Furthermore, I do not know under which circumstances the tests were taken, as they were done online. The experiment required attention and not too much distraction and this was not guaranteed. It is possible that participants did the test while being in a space where women were present (for example in a study hall or in a café), which could influence their bias the same way the bias of an aversive racist is influenced when a black experimenter is in the room. However, I do think it is important to remember that social research is done to replicate effects in everyday life. As implicit bias is never formed in an isolated space, but while surrounded by different distractions and other influencing factors, I think the results from this study are still very relevant. Beside these constraints, it is important to remember that the automatic reactions to an individual of a group do not necessarily reflect the automatic reactions to everyone in that group, and it is hard to say what kind of reaction the
