“Can someone shut her mouth?” – an experimental study of male politicians attacking female opponents on social media and the effects of these attacks

– an experimental study of male politicians attacking female opponents on social

media and the effects of these attacks

Nanna Bülow Buhl

Student ID: 12281360 Master’s Thesis

Graduate School of Communication

Political Communication

Supervisor: Sanne Kruikemeier

1 Abstract

This thesis sets out to research the effects of negative campaigning in a modern setting. Considering the recent increase in awareness of women’s treatment in the public eye, the thesis tests the effects of a male political candidate attacking a female opponent online, specifically to test if a backlash occurs for the sponsoring politician. This relationship between attack and backlash, in the form of decreased favorability, is tested for a moderation effect via the content of the attack (if it is gendered or not). The thesis finds inspiration in previous research on negative campaigning and the effects of negative campaigning in general as well as specific effects of gender in political

communication. While academia does not have a clear answer to whether or not it is effective for politicians to “go negative”, the recent years’ awareness of treatment of women after the #MeToo-movement leads the expectations of the tests. It is expected that a male candidate attacking a female opponent online will receive backlash in form of lower evaluations, and that this backlash will be greater if the attack is related to the woman’s gender.

The hypotheses are tested via a survey experiment. The tests showed that the sponsoring male politician did receive significantly lower evaluations by attacking the opponent (as opposed to a non-attacking comment about her), but this effect was not moderated by the gendered content of the attack. The findings of this research is limited by the nature of the experiment, that controls for external variables, a situation that is not found similar in real-life political campaigns, where partisanship, political interest, individual tolerance for negativity and the noise of other campaign activities can disturb the message.

2 Introduction

This thesis is inspired by an incident that happened in the Danish general election of spring 2019. Social democratic mayor of the town of Roskilde and parliament candidate, Joy Mogensen, had recently announced that she was pregnant with help from a sperm donor (Kristensen, 2019) and was planning to be a solo parent to the coming child (Vindum, 2019) as she was not in a

relationship at the time. Another social democrat, member of the city council of Copenhagen Simon Simonsen commented on her pregnancy in a Facebook post. He described it as “self-centered” and a “self-actualization project” to bring a child into the world knowing it would not have a father and a “tragedy of Shakespearean dimensions” when there simultaneously are about 500,000 wonderful childless men” (Nørgaard, 2019) in Denmark. Simonsen was immediately and widely criticized for this comment (Beck, 2019; Møller & Jørgensen, 2019), and debate about his Facebook post

continued for several days, accusing him of shaming women who chose to have children alone and suggesting that they are lazy (Hagemeister, 2019). The debate resulted in his local charter of the party removing their support for him as a candidate (Øllgaard, 2019). He continued as a candidate only because the deadline for removing his name from the ballots had passed (Dahlberg, 2019) but was ultimately not elected to Parliament, while Mogensen was elected and later appointed as minister of culture (Ertmann, 2019).

The incident shows how quickly and severely a candidate can get hurt after making an attack on an opponent. This effect where an attack made by a political candidate hurts their own evaluations is often called a “backlash” effect (Fridkin & Kenney, 2011, p. 316). This specific attack shows a male candidate attacking a female candidate for something related to her gender, and thus it speaks into the awareness of women’s experiences in society that has been prominent in most of the world since the #MeToo movement went viral in 2017 and even resulted in legal changes (North, 2019). Recent examples of critique due to this are Austrian politician Efgani Dönmez being

3 expelled because of a sexist tweet (Cuddy, 2018) and a Republican political candidate having donations from The Republican Party revoked after posting a sexist and suggestive tweet about former Democratic Presidential Candidate Kamala Harris (Crowley, 2020). Keeping this awareness in mind it sparks the question of whether the effects of the incident with Simonsen and Mogensen were so severe due to the gendered nature of his attack on her. Conversely, had Simonsen attacked her for something not related to her gender, but equally irrelevant to her political abilities, would he had received the same level of backlash for his comment?

The results of previous work on political attacks and negative campaigning are mixed, resulting in inconclusive answers of whether or not “going negative” is an effective strategy for a candidate to use (Craig & Rippere, 2016, p. 394). While there is no substantial evidence that negative campaigning helps a politician in winning votes, there is evidence that negativity is more memorable (Lau et al., 2007, p. 1183) and that visibility of political candidates affect voting

behavior (Van der Pas & Aaldering, 2020, p. 116). Negative campaigning has been shown to affect candidate evaluations, but has been shown to be moderated by type of attack (Craig et al., 2014), gender of the candidates (Craig & Rippere, 2016; Schultz & Pancer, 1997), candidate’s position in the election (Fridkin & Kenney, 2011) and gender of the voters seeing the attack (Fridkin &

Kenney, 2009; Phillips, 2019). However, attacks can be risky as well. Not only is it possible that an attack will result in lowered favorability for the sponsor, it might also not be effectful in lowering the favorability of the attacked candidate.

Building on the notion of negative campaigning and attack literature, this research will bring further knowledge of which attacks are acceptable to voters. Specifically, this thesis will test

whether gendered attacks and non-attacks are acceptable and investigate how making these affect the evaluations of the sponsoring politician. The thesis will thus tap into an awareness in society of

4 women in the public eye (North, 2019) and how they are treated and commented on by others. It will thus add to the literature by testing how a specific type of comment is received by voters.

Furthermore, this study focuses on attacks on social media because political campaigns are increasingly moving online (Phillips, 2019, p. 3). While most literature about negative campaigning is written and researched on campaign ads, more and more political campaigning is done online and specifically on social media (Fischer, 2018). Donald Trump and Hillary Clinton spent a combined 81 million dollars on ads on Facebook during the 2016 presidential election (Halpern, 2019), a figure that is representative of a greater movement of resources toward online advertising. Candidates not only present ads on social media, here they have a platform to debate other candidates and connect with the voters directly. Because everything is fast and accessible, many candidates put in a lot of effort and resources into online advertising.

The purpose of this study is to tap into the reactions male politicians receive from voters because of actions toward their female colleagues. This study investigates whether a male politician receives more backlash when commenting on something related to the woman being a woman, than he would, had he commented about something else related to her. The knowledge of aspects of backlash from negative campaigning is limited, and it is contested whether or not negative campaigning at all is beneficial for the sponsor (Craig et al., 2014, p. 647). By researching this topic, the paper will contribute to the literature by adding empirical knowledge in a specific niche within attacks and backlashes. This niche of gender-related attacks seems to be particularly

prevalent and important to voters; therefore, it is important to investigate. Therefore, this paper will answer the following research question:

Research Question: To what extent do attacks on politicians change voters’ perception of the source of the attack (the sponsoring politician), and what is the role of gender in this relationship?

5 Theory

While some literature on negative campaigning shows that attacking a political opponent can have a positive effect on the sponsor’s evaluations (Craig et al., 2014; Fridkin & Kenney, 2009; Schultz & Pancer, 1997), some research also shows an unintended backlash effect (Craig et al., 2014; Fridkin & Kenney, 2011), where the sponsor’s own evaluations were hurt after the attack. If an attack is executed particularly poorly, it might even raise favorability for the attacked candidate as well hurting evaluations of the sponsoring candidate.

While attacking can be risky, there is evidence that attack tweets give more retweets than non-attack tweets (Lee & Xu, 2018; Stromer-Galley et al., 2018), that negative elements make a campaign more memorable (Lau et al., 2007, p. 1183) and that higher awareness of a politician leads them to be viewed more viable amongst the electorates (Van der Pas & Aaldering, 2020, p. 116). This attention is due to the well-established “negativity bias”, which states that negative information is more likely to attract attention than positive information (Fridkin & Kenney, 2008, p. 697). Based in an evolutionary theory, this bias is explained by a human survival mechanism; it increases the chance of survival to keep an eye out for dangerous situations and avoid risky outcomes (Fridkin & Kenney, 2008, p. 697). Knowing this, politicians and consultants continue using attacks in order to get attention and get ahead in the race.

This theory section will first, discuss why backlashes happen after attacks. Thereafter, the impact of incivility and attacks on the basis of gender will be discussed, before reviewing on which media negative campaigning happen currently.

Lau and Rovner (2009, p. 295) describe the backlash effect as evaluations of an attacking candidate decreasing after an attack they sponsored. If an attack is not made properly, a backlash is a real risk for the sponsor. Craig and Rippere (2016, p. 393) pose that a successful attack is not just

6 negative, but also well-crafted, relevant, and specific about what the target has done wrong. If these criteria are not met, there is a risk that the attack will not have the intended effect on the opponent’s favorability or even worse result in the sponsor’s favorability decreasing.

A branch of the backlash literature explains backlashes to occur because voters cannot see the relevance of an attack for their voting decision. Voters are met with vast amounts of information during an election campaign and relevance is a way to decide which information is “relevant for governing” (Fridkin & Kenney, 2011, p. 308). Voters are interested in knowing how a candidate can affect their daily life while in office, as opposed to information on past “bad behavior” or opinions about policy that is not relevant in the current election (Craig et al, 2014, p. 650). These are not considered helpful for making a voting decision. This can also explain why for politicians often do not face serious consequences when an old photo of them dressed in blackface appears (Wootson Jr., 2019).

Another branch of backlash and negative campaigning literature in examines how gender and use of gender stereotypes affect the amount of backlash a sponsor receives. Stereotypes can affect political communication, when there is a mismatch between the perceived characteristics of a social group and the perceived requirements to fulfill a social role, according to role congruency theory (Van der Pas & Aaldering, 2020, pp. 117-118). Stereotypical male traits like leadership and strength are associated with agentic roles such as community leaders and economic providers (Bauer, 2015, p. 692), while stereotypical female traits like honesty and caring (Fridkin & Kenney, 2009, p. 304) make some voters perceive it as less likely that the women can perform an agentic role well (Simon & Hoyt, 2008). Furthermore, politicians who do not act in the way people expect them to are in the risk of having their message rejected (Craig & Rippere, 2016). This matches with the thesis of expectancy violation theory, which argues that people respond more strongly to

7 788). This can leave female politicians in a gender trap – if they are not perceived as tough and strong, they are not performing their role as politicians, but if they are, they do not perform their gender.

Gender may have an effect on perception of politicians because of different issues raised in attacks and how likely men and women are to go negative (Phillips, 2019). It is seen that women are more hurt by attacks and by making attacks (Craig & Rippere, 2016), although men take severe backlash from attacking women in an uncivil matter (Fridkin & Kenney, 2009). Fridkin and Kenney (2009) explain this as stereotypes interacting with the negative information. Because men are

stereotypically seen as more aggressive, it seems more okay that they are attacked. Meanwhile, women are seen as less aggressive and more compassionate (p. 304), therefore voters are more likely to experience an attack on them as being unfair. Fridkin and Kenney (2011) furthermore add that similarly to pain thresholds (p. 309), voters have different tolerance of negativity which affects the effect of both relevance and civility, in that people who are less tolerant will perceive attacks as more negative and consider it less acceptable. This can also be related to the gender of the voter, as women are more likely to be averse to aggression than men are (Phillips, 2019).

Fridkin and Kenney (2008) propose civility as an important mechanism, because respecting societal norms signal willingness to secure the well-being of others (p. 698) and is expected of politicians. The authors use the example of calling an opponent “dumb” which is perceived as vicious and less accepted than messages with the same overall message in civil tones such as “inexperienced” and “lacking knowledge”. The uncivil attacks do not play by the social rules of behavior in the public sphere and voters are therefore less likely to accept them. Because voters expect for politicians to uphold to social standards in public, the surprise of breach of the social standards causes the uncivil messages to receive much attention (p. 699), this was recently seen in the first period of Donald Trump’s presidency, where his tweets were often mainstream news.

8 The current American president and the campaign that lead him to this position is also a good example of how modern electoral campaigns are increasingly happening on social media. The following paragraph discusses why this has happened and why it is important that research into political communication follows this move.

Electoral campaigning is to a great degree moving away from television and instead moving toward social media platforms (Phillips, 2019). Academia has, until now, mostly focused on

negative campaigning as impact of campaign ads on television. As campaigns are increasingly moved to social media it is important to investigate the effects in this area. Furthermore, research into social media has shown that gender can moderate the effects on social media communication in general. Bode (2017) proposes that the genders engage differently with political communication on social media, with women being more likely to engage with non-offensive content while Evans and Clark (2016) showed that female politicians are more likely to go negative than male politicians on social media. Because the voters are present online more and more resources are spent trying to get voter’s online attention, spending on digital advertising was estimated to be increased by 260% from 2014 to 2018 in American elections (Fischer, 2018). Negativity bias has been shown to exist in social media use as well as on traditional media. Lee and Xu (2018) showed that about half of the tweets by Hilary Clinton and Donald Trump in the 2016 American presidential election were

negative, and that these were significantly more likely to be retweeted, a finding also supported by Stromer-Galley and others (2018). Perhaps for the same reason, there are more negative tweets in races with tight competition (Stromer-Galley et al., 2018).

Following the discussion above and research on negative campaigning, it is not clear whether or not attacking a political opponent is a good tactic. Even though there is much evidence that negative campaigning receives more attention than positive or neutral campaign actions (Lee & Xu, 2018; Stromer-Galley et al., 2018; Van der Pas & Aaldring, 2020), there is also evidence that it

9 can result in backlash for the sponsoring politician (Craig et al., 2014; Craig & Rippere, 2016; Fridkin & Kenney, 2011). Because of the awareness of women’s treatment in the recent years, it is possible that a male politician attacking a female politician is likely to backlash. This is believed both because there is a movement of support for women in the public, as well an awareness for especially men misbehaving toward women. Therefore, this thesis will fist investigate the following hypothesis:

H1: A male politician making an attacking comment on social media toward a female politician will have a lower favorability than a politician making a neutral comment on social media.

Figure 1 – Expected causal relationship of Hypothesis 1

In addition to reacting negatively when a male politician attacks a female politician, it can be expected that the content of this attack can moderate this effect. Specifically, it can be expected that if the attack is related to the female politician being a woman or using language that references to her gender is less accepted, because the mistreatment of women started the #MeToo-movement and several incidents with specifically gendered comments toward women have sparked much criticism for the sponsor (Crowley, 2020; Cuddy, 2018; Forgey, 2019; North, 2019). Therefore, the effect from H1 is expected to be moderated and the second hypothesis this thesis will test reads as follows:

Attack from male candidate on female candidate

Favorability of sponsoring candidate

10 H2: The effect of a male politician’s attack on a female politician is moderated by the gendered content of the message in such a way that when a male politician makes a gendered attack towards a female politician, the sponsor’s favorability will decrease more than it would with a non-gendered attack.

Figure 2 – Expected causal relationship of Hypothesis 2

The hypotheses will be tested through a survey experiment. The survey experiment allows a test of small differences in social media posts, which can measure whether the treatment groups react differently to the comments in a controlled environment. By using real politicians, the setting of the experiment is realistic and therefore is thought to show an effect close to a similar effect in a real-life setting. The authenticity of the experiment was further increased by using tweets

reproduced from real politician’s comments.

Respondents and design

Data for this study was collected from the end of April to beginning of May in 2020, through sharing an anonymous survey link on social media and in the author’s personal network. After being shared on social media, the author continually shared the survey in survey sharing

Attack from male candidate on female candidate

Favorability of sponsoring candidate

Gendered attack (compared to non-gendered attack)

11 groups on social media. Given that these groups mostly consist of students sharing surveys in order to get more respondents for school projects and university papers, participants are of a young age which is showed by the median age of participants is 22, but also secured an international

participant base with participants reporting being from 34 different countries. This also correlates with the fact that the median length of education was 3-4 years of higher education. Participants were mostly female, with 68.7 % of the participant base being women. See Appendix 1.1-1.4 for descriptive statistics on the participants.

In order to test the negative campaigning effect in a realistic setting, it was important that the treatment stimuli used in the experiment resembles messages that could be posted in a real-life political campaign. Several cases were researched wherein an either comment or attack related to gender was made that could inspire the constructed tweets. As to not prime participants or affect their responses, it was important that the stimulus material did not contain known names or were recognizable quotes. For example, there are plenty of comments of this nature from Donald Trump (Prasad, 2019), but because participants might know these quotes and have particular opinions of him, they cannot be used in this case.

The tweet for the gendered attack treatment was inspired by a UK Independence Party (UKIP) politician, whose old tweets were republished after he announced his run for leadership of the party. One of the tweets said, regarding the Scottish First Minister: “Can someone just like… tape Nicola Sturgeon’s mouth shut? And her legs, so she can’t reproduce. Thanks. #ITVEURef” after an EU-referendum debate (Allegretti, 2016). The quote was changed slightly in the stimulus material, to avoid participants recognizing the quote and this affecting their responses. The non-gendered attack was inspired by an attack by a surrogate for the senator of Missouri that of the senator’s opponent said the opponent: “is a liar…. She’ll cheat to get elected” (as quoted in Fridkin and Kenney (2008, p. 694). This quote was chosen because it attacked the politician in an uncivil

12 manner, underlining her gender by naming her “she”, but without the content being related to her gender. Meanwhile, the gendered non-attack was inspired by supportive comments about “the squad”, four progressive female Democratic US Congresswomen whom Donald Trump attacked on Twitter (Forgey, 2019). The final treatment stimulus (non-gendered non-attack) simply mentioned the female candidate in a short acknowledgement of a good debate.

Procedure

The survey was conducted as follows. Participants were first introduced to the experiment, then asked a series of background questions of age, country of origin, gender, and education. These are variables that might have an effect on the dependent variable and should therefore be used as controls (Mutz, 2011, p. 123). Participants were subsequently presented with biographies of two British politicians, see the material in Appendix 3.1 and 3.2.

As mentioned, both politicians are real and so was the information given about them. The two politicians were chosen for several reasons. They were not Danish, and since it was expected that most of the participants would be from Denmark and there are only 179 members of the Danish Parliament, the risk of participants knowing he politician and this affecting the results was

significant. British politicians were chosen because the Parliament has 650 members and many members who do not hold a ministerial position or a position in the party are not well known in the public. Further, these two politicians were chosen because of their respective gender and because they come from the same party, Labour Party. Both had served maximum one term as MP’s and are about the same age, to control for these external factors as well. To avoid a situation where some participants knew of the politicians, because of a previous scandal or fame, a search on both their names were conducted and found that this was not the case.

13 After reading biographies of the politicians, participants were told that the two politicians had met in a debate during the UK election campaign leading up to the election in December 2019. They were told that Neil Coyle, the male politician, had made a post on Twitter after the debate, this post was the stimulus that participants saw in the next step (see Appendix 4.1-4.4). Immediately after having seen the tweets, participants were asked questions for the dependent variable,

favorability of the sponsoring candidate. Favorability for the attacked candidate, Stephanie Peacock, and a series of emotion-related and vote intention questions were asked following this. A set of control questions about participant’s view on feminism were placed after the favorability questions, so as not to prime participants. Lastly, participants were debriefed and made aware that the tweets were constructed for the experiment.

Measurements

Favorability was chosen to measure the effects of the attacks and backlash because previous research has often used this very method (American National Election Studies, 2016; Craig & Rippere, 2016; Fridkin & Kenney, 2008; Fridkin & Kenney, 2011) and results are therefore

comparable. Favorability can be measured in several ways, this thesis measured it in the most used manner. As in American National Election Studies (2016) and Fridkin and Kenney (2008),

favorability was measured from a feelings thermometer, using the phrasing “I would like you to rate Neil Coyle below what is called a feeling thermometer, which measures how much he is perceived as warm and favorable”. In this thesis however, as opposed to ANES (2016), the question was split into two responses, one measuring warmth (M = 3.22, SD = 1.72) and one measuring favorability (M = 3.21, SD = 1.69), because these are two different questions that cannot be answered in one

14 measurement. The dependent variable was thus only the favorability measurement and was

measured on a 7-point scale.

Furthermore, a set of control questions measuring level of feminism within the participants was created. The set was used as control, as it can be expected that a person’s level of feminism would moderate their reacting to a gendered attack. Participants were asked “Do you consider yourself a feminist [strong feminist, feminist, not a feminist] as well as “How well does the term “feminist” describe you?” and “How important is it to you to be a feminist?”. The latter two were measured on 5-point scales. A factor analysis by principal axis factoring with Direct Oblimin rotation strategy showed an Eigen value above 1 (2.94) and that 83.14% of the variance in the original items is explained by the measurement of feminism. A Cronbach’s Alpha of α = .87 shows a high internal consistency, and the three items were therefore recoded to one feminism control variable (M = 2.75, SD = 0.89).

Manipulation check

A manipulation check was performed to test whether the stimuli had the expected effect. Oneway ANOVA tests were performed for the randomization variable’s effect on the manipulation variable, asking if the tweets were respectively an attack, gendered and sexist. The attack

manipulation asking the question “The tweet I saw from Neil Coyle was an attack toward Stephanie Peacock” showed diverse means across treatment groups (M gendered attack = 6.43), (M gendered non-attack = 2.89), (M non-gendered attack = 6.12), (M non-gendered non-attack = 2.37) and an effective manipulation F (3, 198) = 108,81, p = .000. The gender manipulation asked, “The tweet I saw from Neil Coyle was related to the gender of Stephanie Peacock” and showed significant results F (3, 197) = 92.86, p = .000 across treatment groups, (M gendered attack = 5.96), (M gendered non-attack= 5.75), (M non-gendered attack =

15 2.48), (M non-gendered non-attack = 2.14). Finally, asking “The tweet I saw from Neil Coyle was sexist” showed the following means across treatments: (M gendered attack = 5.98), (M gendered non-attack = 4.00), (M non-gendered attack =2.23), (M non-gendered non-attack = 2.12). The sexist manipulation check showed

significant results F (3, 197) = 63.88, p = .000. The manipulations are therefore considered successful. See result tables in Appendix 2.1-2.3.

Analysis

A straight liners check via the standard deviation of battery method (Kim et al., 2019) was performed as attention check instead of one specific question. This choice was made to avoid that the attention check would not be noticed by respondents who gave otherwise useful answers and thus take out relevant results from the collected data. The check was performed on a variable set of feelings toward Neil Coyle and showed 20 participants who had a standard deviation of 0 across the variable set. The participants were examined to check for other signs of negligence in their

responses instead of excluded to avoid excluding useful responses or create a skewed treatment distribution (Jovancic, 2019). Ultimately they were kept in the survey, as there was no clear indication of low attention levels throughout the survey.

Hypothesis 1 will in the following results section show tested, with a dichotomous

independent variable, coding the randomization into attack (non-gendered and gendered) or neutral (non-gendered and gendered). The analysis will be performed via a oneway-ANOVA, in order to test whether the stimuli created a difference on the dependent variable, favorability of Neil Coyle. Hypothesis 2 will also be tested with a dichotomous variable, coded in gendered comment treatment (attack and neutral) or non-gendered comment treatment (attack and neutral). The dependent

16 variable will remain to be favorability of Neil Coyle, and this test will show whether the effect has been moderated by including the dichotomous gender variable in the analysis.

Results

A randomization check was performed for four control variables, that were thought to affect the dependent variable (Mutz, 2011, p. 109) – age, education, country, and gender. A oneway ANOVA test for age showed that the mean age of the treatment groups was not statistically different F (3, 208) = 0.21, p = .888. A chi-square test of gender showed that the groups were not statistically significant differently from each other χ2 (3, N = 222) = 1.04, p = .719. The chi-square test of countries between treatment groups also showed insignificant results χ2 (99, N = 192) = 99.45, p = .469 and a chi-square test examining education showed an insignificant result χ2 (15, N = 225) = 8.86, p = .885. Finally, a test for the feminist control variable showed that it did not have a significant effect on the treatment groups F (10, 191) = 0.77, p = .656. Because none of the variables showed significant difference between the treatment groups, the randomization is considered successful and the tested variables will not be controlled for in the analysis. See descriptive results for the variables in Appendix 1.1-1.5.

To test H1 (whether an attacking comment leads to lower favorability than a non-attacking comment for a male politician attacking a female politician), a oneway ANOVA was performed. It tested the effects of the dichotomous attack variable had on favorability of Neil Coyle, (M attack comment = 1.97, SD = 1.11; M non-attack comment = 4.21, SD = 1.23; F (1, 210) = 320.99, p = .000. Put in other words, an attacking comment (compared to a non-attacking comment) leads to lower

favorability for a male politician attacking a female politician. Therefore, this analysis supports H1. The findings are shown graphically in Figure 3 below.

17 Figure 3 –Distribution of mean favorability by treatment group (attack or no attack)

A two-way ANOVA was conducted to test Hypothesis 2. It tested whether seeing a

gendered comment (compared to a non-gendered comment) changes the way the attack from a male politician on his female opponent is viewed amongst voters. Thus, it examines whether the content of the attack (being gendered or not) moderates the main relationship between attacks and

favorability. The two-way ANOVA did not show a statistically significant moderation between an attacking tweet and a gendered tweet F (1, 208) = 2.48, p = .117. This analysis therefore did not find support for Hypothesis 2. In other words, there was not found evidence that the specific gendered content of the comment had a different effect on the respondents than the non-gendered content. The findings are depicted graphically in Figure 4 below. See descriptive statistics and between subject effects in Appendix 6.

1 1,5 2 2,5 3 3,5 4 4,5 5 Non-attack Attack

Fav

o

rab

il

it

y

18 Figure 4 – Distribution of mean favorability by treatment group (attack) and moderator (gender)

Discussion

By showing that attacks lead to lower evaluations of the sponsoring politician and thus supporting H1, this thesis supports findings (Craig et al, 2014; Fridkin & Kenney, 2011) that show that a politician’s evaluations can be significantly damaged by attacking opponents. The finding that attacks lead to less favorability for the sponsor, can be viewed as support for the relevance

argument (Fridkin & Kenney, 2011), as participants rejected the messages that were not relevant for making their voting decision. The non-gendered attack made accusation that, if true, would make Stephanie Peacock’s performance in office worse. But since there was no examples or proof of this bad behavior, it is possible that it would not be believed by the voters and therefore not be taken into consideration while making their voting decision.

1 1,5 2 2,5 3 3,5 4 4,5 5 Non-gendered Gendered F avora bil it y Experimental treatment Non-attack Attack

19 Similarly, the difference between the attacks and the non-attacking comments can be seen as driven by incivility in the communication from Neil Coyle. Both suggesting that someone should “shut Stephanie Peacock’s mouth” and that they should be “a liar” would by most voters be viewed as uncivil (Fridkin & Kenney, 2009), especially when not substantiating this accusation with specific information about why this is the case (Craig and Rippere, 2016, p. 393). Expectation violation theory states that people have a certain expectation of certain societal groups act (Cassese & Holman, 2018, p. 788), and politicians are not expected to use uncivil language while in their politician role, as it is not expected to see uncivil language or behavior in the public sphere in general (Fridkin & Kenney, 2008, p. 698). In finding support for H1, this thesis supports the findings of Fridkin and Kenney (2009) who found that men take severe backlash from attacking women in an uncivil manner. With this in mind, it can be questioned why support for H2 is not found, if using Stephanie Peacock’s gender in the attack is considered more uncivil than the calling her a “liar”. The following discussion will suggest explanations for the rejection of H2.

There are several options for why support was not found for Hypothesis 2. As mentioned, both the attack treatments could be considered as being uncivil, and it might be because of this the difference of the attack (compared to no attack) is driven, while no effect of a gendered comment (compared to a non-gendered comment) was seen. Both the attack stimuli were made to be uncivil to be comparable. Because of the gendered attack stimulus being so aggressive the non-gendered attack was made to be comparably aggressive, so a potential difference was not driven because of this. This will further be discussed later in this section, when discussing the limitation of these findings.

Furthermore, it can be argued that there is a ceiling for acceptable attacks, and that passing this ceiling does not further increase the effect of the attack. So even though it might be less acceptable to use a woman’s gender in an attack against her, calling her a liar has passed a certain

20 point of unacceptability that adding the gendered factor does not further decrease his favorability. Therefore, the addition of the gendered attack would not make the attack worse in voter’s eyes and the subsequent decrease in favorability would not be bigger.

A final explanation for the gendered attack not moderating the attack itself is that voters do not find a gendered attack worse than any other attack. Even though there have been many

examples of gendered attacks and comments receiving severe criticism (Allegretti, 2016; Cuddy, 2018, Crowley, 2020), it could be that voters have simply been saturated with this topic and have moved on to care about other topics in politics or that they have come to expect this of male politicians. Therefore, the gendered addition would not further change their view of the sponsoring politician.

When reviewing the analysis and results, two methodological choices are apparent and can possible have affected these. The effects of those are found in the following discussion. Firstly, it was chosen not to used forced responses in the survey, thereby letting participants skip questions. This choice was made because when forcing a response, there is a risk that participants might choose something that is not how they truly feel and therefore distort the data (Herndon, 2014). However, this choice can lead to lower numbers of answers, especially in the latter parts of the survey as participants grow tired or uninterested. This was shown to happen in this survey, for example in the higher number of responses in evaluation of Neil Coyle (32 missing responses for favorability) compared to the number of responses in Stephanie Peacock’s evaluation (41 missing responses for favorability). With that being said, the difference in participants between the

treatment groups in the analysis (n non-gendered non-attack = 39, n gendered non-attack = 42, n non-gendered attack = 43, n gendered attack = 42) is low, with a difference of four people between the smallest and the largest group. Therefore, it is not expected that this methodological choice can have affected the results.

21 While there is a small visible difference, this is not enough to drive the difference in the results of the dependent variables.

Secondly, it was chosen not to have a classic attention check in the survey, but instead use a straight liners check via the standard deviation of battery method (Kim et al., 2019). This choice was made because of a worry that an attention check would not be noticed by respondents who gave otherwise useful answers and thus remove relevant results. For the same reason, the participants who failed the straight liners test were not pulled away from the survey, but simply inspected (Jovancic, 2019) to test for other signs of negligence in their responses. This could risk not pulling away respondents that were not answering the survey properly and thus distorting the results. However, rather than ascribing these results to methodological choices, the low difference in favorability between the gendered versus non-gendered groups is believed to be because of the stimulus material as there was a clear significant difference between the attack groups, that likely could have disappeared had the participants not been paying attention. Furthermore, the participants who were outliers in the straight liners check were inspected and no other sign of low attention was seen in their responses, their low standard deviation can therefore be explained by an effective stimulus material that made them greatly dislike the sponsoring politician.

What can be seen as limitation for the findings of this thesis, is the generalizability of the results. If the same experiment were to be tested on people of similar demographic background, it can be expected that this test will generate similar results because of the significant difference that was seen between the groups in this thesis. However, as formerly noted, an actual election is more complex, and voters are throughout a long period of time given mounts of information, and it cannot be expected that they pay as much attention to a piece of information during an election campaign as they do while participating in a survey experiment (Malloy & Pearson-Merkowitz, 2016, p. 2). As Mutz noted “if an effect can only be obtained when unrealistically high levels of

22 attention are directed toward a stimulus, then the effect probably does not occur as often in the real world” (2011, p. 13). Based on the findings of this thesis however, and the strong effect that the stimulus material has shown to have, it would be expected that a similar effect can be found in an actual political campaign, although most likely not as strong because of the higher complexity.

There have been many scandals related to politicians and other prominent people during the last few years, the current president of the United States often being an example, where a comment was posted on social media and thereafter picked up by mainstream media. Specifically, attacking comments which are known to get more attention on social media (Lee & Xu, 2018; Stromer-Galley et al., 2018) are often the ones that move on to become a story with other news organizations. Following this discussion of external validity, it is relevant to ask whether the treatments in the tweets themselves were realistic, as an unrealistic stimulus would be an important limitation. Too strong or too weak attacks would not have given an accurate vision of what voters react on in an election scenario and would therefore not have been generalizable. However, because the treatment tweets were based on tweets from real politicians, and these real tweets also caused criticism of that politician, the results can be considered to be generalizable. Most likely however, the effects would not be as strong because of the voters having more actual information about the candidates and the noise occurring during a campaign.

Finally, a limitation to this research can be that the effect is possibly driven by the uncivility combined with irrelevance in the constructed tweets in the stimulus material. Both the gendered and the non-gendered attack were created to be uncivil and irrelevant because the tweets that were used for inspiration where a male politicians attacked a female opponent in a way that was related to her gender were often uncivil (Allegretti, 2016; Fridkin & Kenney, 2008). Therefore, it can be

questioned whether the incivility and irrelevance drove the effect or the fact that it was a man attacking a woman drove it. While this cannot be tested with the current data, it is important to note

23 as a limitation. However, seeing as both the gendered and the sexist manipulation proved effective, it is believed that a combination of incivility and irrelevance in the comments as well as the

gendered aspect drove the effect of the tested attacks.

The research question of this thesis asked to what extent attacks on politicians change voter perception of the sponsor of the attack and which role gender plays in this relationship. The analysis of former literature on the topic showed a mixed picture in that some articles thought that an attack on a fellow candidate during a political campaign would hurt their evaluations, while some claimed that the sponsor could gain ground by attacking their opponent. The analysis in this thesis have showed that any attack on the opponent hurts the sponsor via the backlash effect, therefore resulting in a lower evaluation than before the attack. The results further showed that bringing gender into the relationship, by it being a male candidate attacking a female one for something related to her

gender, did not moderate the effect of the attack on favorability of the sponsor. However, it should be noted that both attacks that were tested in this relationship were uncivil and irrelevant and that generalizability for the findings is limited due to the controlled nature of experiments.

Conclusion

This thesis has researched and analyzed the effects of negative campaigning in a modern setting. Specifically, the effects of a male politician attacking a female politician were tested with the addition of a gendered moderation to this relationship. Tests showed that an uncivil and irrelevant attack hurt the sponsoring candidate’s favorability to a great deal but that this effect was not enhanced by the attack being gendered. Thus with both tested attacks being uncivil, the thesis confirms previous findings that civility affects the effect of the attack (Fridkin & Kenney, 2008; Fridkin & Kenney, 2011) and that backlashes can occur for the sponsoring candidate (Craig et al.,

24 2014; Craig & Rippere, 2016; Fridkin & Kenney, 2011). This research is a thus further evidence that an attack can result in backlash for the sponsor, specifically when it is a male candidate attacking a female candidate in an uncivil and irrelevant manner.

The thesis’ results are limited by the fact that they were found via a survey experiment. An experiment is a good way to secure the causal relationship and control the effect of the attack

without interference from other variables and spurious relationships. However, in an election, voters receive mounts of information and they can easily overlook one attack amongst the many comments made by politicians during a campaign. Therefore, it cannot be expected that the effect shown in the thesis, will be as strong in a real-life election scenario, where partisanship, individual tolerance for negativity, political interest and the noise of other political communication can disturb this effect. With this said, the stimulus material of the experiment was made to look like social media

comments from real politicians, whose similar comments did receive criticism by the public. Therefore, a smaller effect, but still an effect can be expected if a male politician made similar attack comments toward a female politician in a current political campaign.

Future research on this topic could benefit from testing this effect in moderation of more realistic settings than in an experimental. This could be by observing reactions when a similar case happens online or by following a certain race in a time series study, hoping that attacks will occur that they can analyze the effects of. Furthermore, it could be interesting to reverse the genders in the attack. Even though gendered attacks from female candidates toward their male counterparts are not often seen, it would also be interesting to test the effects of such in an experimental setting first, as to avoid interference. Along the same lines, it would be interesting to test if the effect is the same when it comes to a different aspect of the politician, particularly race is prevalent at this moment in time. As awareness of race inequality has increased during 2020, it would be interesting to test

25 whether positions have changed, by collecting reactions to old racist tweets that were made before this recent increase in awareness and compare these to reactions to current or future comments.

References

Allegretti, A. (2016, October 28). Ukip Leadership Candidate Raheem Kassam Apologises For ‘Violent,

Crass, Unacceptable’ Sturgeon Joke. Huffington Post UK Edition. Retrieved from:

https://www.huffingtonpost.co.uk/entry/ukip-leadership-frontrunner-raheem-kassam-nicola-sturgeon-twitter-apology_uk_5813943de4b0672ea688ddc0?guccounter=2

American National Election Studies. (2016). 2016 Time Series Pre-Election Questionnaire. American National Election Studies. Retrieved from:

https://electionstudies.org/wp-content/uploads/2018/03/anes_timeseries_2016_qnaire_pre.pdf

Bauer, N. (2015). Emotional, Sensitive, and Unfit for Office? Gender Stereotype Activation and Support Female Candidates. Political Psychology, 36(6), 691–708. https://doi.org/10.1111/pops.12186

Beck, A. M. (2019, May 2). Mette Frederiksen raser efter uhørt angreb på gravid borgmester [Mette Frederiksen rages after unheard attack on pregnant mayor]. Tv2 Lorry. Retrieved from:

https://www.tv2lorry.dk/roskilde/mette-frederiksen-raser-efter-uhort-angreb-pa-gravid-borgmester

Bode, L. (2017). Closing the gap: gender parity in political engagement on social media. Information, Communication & Society, 20(4), 587–603. https://doi.org/10.1080/1369118X.2016.1202302

Cassese, E., & Holman, M. (2018). Party and Gender Stereotypes in Campaign Attacks. Political Behavior, 40(3), 785–807. https://doi.org/10.1007/s11109-017-9423-7

Craig, S., & Rippere, P. (2016). He Said, She Said: The Impact of Candidate Gender in Negative Campaigns. 12(2), 391–414. https://doi.org/10.1017/S1743923X15000550

26 Craig, S., Rippere, P., & Grayson, M. (2014). Attack and Response in Political Campaigns: An

Experimental Study in Two Parts. Political Communication, 31(4), 647–674.

https://doi.org/10.1080/10584609.2013.879362

Crowley, J. (2020, January 15) 5 Things to Know About Robert Hyde, Congressional Candidate in Lev

Parnas Documents. Newsweek. Retrieved from: https://www.newsweek.com/robert-hyde-lev-parnas-things-know-1482375

Cuddy, A. (2018, September 4). ‘Look at her knees’: Austrian politician expelled from ruling party over

‘sexist’ tweet. Euronews. Retrieved from: https://www.euronews.com/2018/09/04/-look-at-her-knees-austrian-politician-expelled-from-ruling-party-over-sexist-tweet

Dahlberg, B. G. (2019, May 14). Trods mistillid: Simon Simonssen fortsætter som folketingskandidat [Despite mistrust: Simon Simonsen continues as parliamentary candidate]. Altinget. Retrieved from:

https://www.altinget.dk/artikel/trods-mistillid-simon-simonsen-fortsaetter-som-folketingskandidat

Ertmann, B. (2019, November 27). Her er Danmarks nye ministerhold [Here is Denmark’s new team of ministers]. Jyllands-Posten. Retrieved from: https://jyllands-posten.dk/politik/ECE11469296/Her-er-Danmarks-nye-ministerhold1/

Evans, H., & Clark, J. (2016). “You Tweet Like a Girl!”: How Female Candidates Campaign on Twitter. American Politics Research, 44(2), 326–352. https://doi.org/10.1177/1532673X15597747

Fischer, S. (2018, November 6). Political ad spending hits new record for the 2018 midterm elections. Axios Media. Retrieved from: https://www.axios.com/record-midterm-ad-spend-explodes-money-was-no-object-1541450836-f92d1767-ad5f-4d85-99ee-96d9847e7691.html

27 Forgey, Q. (2019, July 16). Biden praises AOC and progressive congresswomen following Trump’s

attacks. Politico. Retrieved from:

https://www.politico.com/story/2019/07/16/joe-biden-democratic-congresswomen-trump-1416578

Fridkin, K., & Kenney, P. (2008). The Dimensions of Negative Messages. American Politics Research, 36(5), 694–723. https://doi.org/10.1177/1532673X08316448

Fridkin, K., & Kenney, P. (2009). The Role of Gender Stereotypes in U.S. Senate Campaigns. Politics & Gender, 5(3), 301–324. https://doi.org/10.1017/S1743923X09990158

Fridkin, K., & Kenney, P. (2011). Variability in Citizens’ Reactions to Different Types of Negative Campaigns. American Journal of Political Science, 55(2), 307–325. https://doi.org/10.1111/j.1540-5907.2010.00494.x

Hagemeister, M. L. (2019, May 3). S-kandidat føler sig misforstået: ’Selvfølgelig skal Joy have lov til at få

et barn alene’ [S-candidate feeling misunderstood: ’Of course Joy should be allowed to have a child

alone’]. DR. Retrieved from: https://www.dr.dk/nyheder/politik/s-kandidat-foeler-sig-misforstaaet-selvfoelgelig-skal-joy-have-lov-til-faa-et-barn

Halpern, S. (2019, October 24). The problem of political advertising on social media. The New Yorker. Retrieved from: https://www.newyorker.com/tech/annals-of-technology/the-problem-of-political-advertising-on-social-media

Herndon, M. (2014, December 5). To Force or no to Force (an answer): It’s a complicated question. Corona Insights. Retrieved from: https://www.coronainsights.com/2014/12/to-force-or-not-to-force-an-answer-its-a-complicated-question/

Jovancic, N. (2019, November 12). Attention Check Questions: Yay or Nay for Surveys? LeadQuizzes. Retrieved from: https://www.leadquizzes.com/blog/attention-check-questions/

28 Kim, Y., Dykema, J., Stevenson, J., Black, P., & Moberg, D. (2019). Straightlining: Overview of

Measurement, Comparison of Indicators, and Effects in Mail–Web Mixed-Mode Surveys. Social Science Computer Review, 37(2), 214–233. https://doi.org/10.1177/0894439317752406

Kristensen, T. (2019, April 28). Joy er gravid: Lykkelig Roskilde-borgmester – Breddam barselsvikar [Joy is pregnant: Happy Roskilde-mayor – Breddam is maternity replacment]. Roskilde Avis. Retrieved from:

https://sn.dk/Roskilde-Avis/Joy-er-gravid-Lykkelig-Roskilde-borgmester-Breddam-barselsvikar/artikel/835124

Lau, R., & Rovner, I. (2009). Negative campaigning. Annual Review of Political Science, 12(1), 285–306. https://doi.org/10.1146/annurev.polisci.10.071905.101448

Lau, R., Sigelman, L., & Rovner, I. (2007). The Effects of Negative Political Campaigns: A Meta-Analytic Reassessment. The Journal of Politics, 69(4), 1176–1209.

https://doi.org/10.1111/j.1468-2508.2007.00618.x

Lee, J., & Xu, W. (2018). The more attacks, the more retweets: Trump’s and Clinton’s agenda setting on Twitter. Public Relations Review, 44(2), 201–213. https://doi.org/10.1016/j.pubrev.2017.10.002

Malloy, L., & Pearson-Merkowitz, S. (2016). Going positive: The effects of negative and positive advertising on candidate success and voter turnout. Research & Politics, 3(1).

https://doi.org/10.1177/2053168015625078

Mutz, D. C. (2011). Population-Based Survey Experiments. Princeton University Press.

Møller, A. S. and Jørgensen, M. K. (2019, May 2). Gravid borgmester bliver kaldt egoistisk af

partikollega: - Han kender intet til mit liv [Pregnant mayor is called selfish by party colleague: - He

knows nothing about my life]. Tv2. Retrieved from: https://nyheder.tv2.dk/samfund/2019-05-02-gravid-borgmester-bliver-kaldt-egoistisk-af-partikollega-han-kender-intet-til-mit

29 North, A. (2019, October 4). 7 positive changes that have come from the #MeToo movement. Vox.

Retrieved from: https://www.vox.com/identities/2019/10/4/20852639/me-too-movement-sexual-harassment-law-2019

Nørgaard, M. (2019, May 1). Partifælle om donor-gravid borgmester: Selvcentreret og egoistisk [Party colleague about pregnant mayor: Selfcentered and selfish]. Estra Bladet. Retrieved from:

https://ekstrabladet.dk/nyheder/politik/danskpolitik/partifaelle-om-donor-gravid-borgmester-selvcentreret-og-egoistisk/7615844

Phillips, J. (2019). Negative Political Communication on Social Media and the Gender Gap: A Study of Men’s and Women’s Reactions to Presidential Candidate Attacks on Facebook in 2012 and 2016. Politics & Gender, 1–29. https://doi.org/10.1017/S1743923X1900059X

Prasad, R. (2019, November 29). How Trump talks about women – and does it matter? BBC News. Retrieved from: https://www.bbc.com/news/world-us-canada-50563106

Schultz, C., & Pancer, S. (1997). Character Attacks and Their Effects on Perceptions of Male and Female Political Candidates. Political Psychology, 18(1), 93–102.

https://doi.org/10.1111/0162-895X.00046

Simon, S., & Hoyt, C. (2008). Exploring the Gender Gap in Support for a Woman for President. Analyses of Social Issues and Public Policy, 8(1), 157–181.

https://doi.org/10.1111/j.1530-2415.2008.00167.x

Stromer-Galley, J., Zhang, F., Hemsley, J., & Tanupabrungsun, S. (2018). Tweeting the Attack: Predicting Gubernatorial Candidate Attack Messaging and Its Spread. International Journal Of

30 Van Der Pas, D., Aaldering, L., & Political Science and Public Administration. (2020). Gender Differences

in Political Media Coverage: a Meta-Analysis. Journal of Communication, 70(1), urn:issn:0021– 9916.

Vindum, K. B. (2019, April 29). Borgmester deler ‘den bedste af alle nyheder’: Skal være solo-mor [Mayor shares ‘the best of all news’: Is going to be solo mother]. Tv2 Lorry. Retrieved from: https://www.tv2lorry.dk/roskilde/borgmester-deler-den-bedste-af-alle-nyheder-skal-vaere-solo-mor

Wootson Jr., C. R. (2019, February 16). The lengthy history of white politicians wearing blackface – and

getting a pass. The Washington Post. Retrieved from:

https://www.washingtonpost.com/nation/2019/02/16/lengthy-history-white-politicians-who-wore-blackface-got-pass/

Øllgaard, A. K. (2019, May 7). Kredsformand om mistillid til S-kandidat: Nu er det nok [Local chapter chairman about mistrust to S-candidate: Now it is enough]. DR. Retrieved from:

31 Appendix

Appendix 1.1 – Descriptive statistics and One-Way Analysis of Variance of age between treatment groups Age Treatment group n M SD Gendered attack 55 8.53 5.36 Gendered non-attack 53 8.34 6.53 Neutral non-attack 52 7.81 3.98 Neutral attack 52 8.58 5.89 Total 212 8.32 5.49 Source df SS MS F p Between groups 3 19.46 6.49 .213 .888 Within groups 208 6346.37 30.51 Total 211 6365.83

Appendix 1.2 – Descriptive statistics and chi-square test of independence of gender between treatment groups

Gender

Treatment group Male Female Total

Gendered attack 20 36 56 Gendered non-attack 19 38 57 Neutral non-attack 16 39 55 Neutral attack 15 39 54 Total 70 152 222 χ2 (3, N = 222) = 1.04, p = .719.

32 Appendix 1.3 – Descriptive statistics and chi-square test of independence of country between treatment groups Country Gendered attack Gendered non-attack Neutral non-attack

Neutral attack Total

Australia 1 1 1 2 Canada 2 2 China 1 2 2 5 Curacao 1 1 Cyprus 1 1 Denmark 20 21 15 17 73 France 1 2 1 4 Germany 4 2 3 4 13 Greece 1 1 Hong Kong 1 1 Hungary 3 1 4 India 1 1 1 1 4 Italy 5 3 4 1 13 Japan 1 1 Latvia 1 1 Lithuania 1 1 1 3 Malta 1 1 Morocco 1 1 Malaysia 1 1 Norway 1 1 Poland 1 1 2 Portugal 1 1 Romania 1 1 Russia 2 2 South Africa 1 1 Spain 2 2 Sweden 2 2 Switzerland 1 1 Taiwan 1 1 The Netherlands 4 6 7 10 27 UK 1 3 2 2 8 USA 3 2 1 3 9 Venezuela 1 1 Vietnam 1 1 Total 45 50 47 50 192 χ2 (99, N = 192) = 99.45, p = .46

33 Appendix 1.4 – Descriptive statistics and chi-square test of independence of education

between treatment groups

Treatment group Education level Gendered

attack Gendered non-attack Neutral non-attack Neutral attack Total Primary school 1 1

Some high school 1 1

High school 6 6 6 5 23

Some higher education (1-2 years)

7 7 3 7 24

Higher education (3-4 years)

26 24 28 25 103

Long higher education (more than 4 years)

16 19 20 18 73

Total 56 57 57 55 225

χ2 (15, N = 225) = 8.86, p = .885

Appendix 1.5 – Descriptive statistics and One-Way Analysis of Variance of the feminist control variable between treatment groups

Level of feminism Treatment group n M SD Gendered attack 53 2.81 0.86 Gendered non-attack 49 2.89 0.97 Neutral non-attack 51 2.60 0.81 Neutral attack 49 2.69 0.92 Total 202 2.75 0.89 Source df SS MS F p Between groups 3 2.69 0.89 1.13 .337 Within groups 198 156.98 0.79 Total 201 159.68

34 Appendix 2.1 – One-Way Analysis of Variance for manipulation check, attack

The tweet was an attack on Stephanie Peacock

Treatment group n M SD Gendered attack 53 6.43 1.20 Gendered non-attack 49 2.89 1.64 Neutral non-attack 51 2.37 1.62 Neutral attack 49 6.12 1.28 Total 202 4.47 2.34 Source df SS MS F p Between groups 3 686.68 227.89 108.81 .000 Within groups 198 414.69 2.09 Total 201 1098.38

Appendix 2.2 – One-Way Analysis of Variance for manipulation check, gendered

The tweet was about Stephanie Peacock’s gender

Treatment group n M SD Gendered attack 53 5.96 1.58 Gendered non-attack 49 5.75 1.31 Neutral non-attack 51 2.14 1.40 Neutral attack 48 2.48 1.74 Total 201 4.11 2.34 Source df SS MS F p Between groups 3 640.59 213.53 92.86 .000 Within groups 197 453.00 2.30 Total 200 1093.59

35 Appendix 2.3 – One-Way Analysis of Variance for manipulation check, sexist

The tweet was sexist

Treatment group n M SD Gendered attack 53 5.98 1.55 Gendered non-attack 49 4.00 1.83 Neutral non-attack 51 2.12 1.46 Neutral attack 48 2.23 1.66 Total 201 3.62 2.27 Source df SS MS F p Between groups 3 510.51 170.17 63.89 .000 Within groups 197 524.75 2.66 Total 200 1035.26

36 Appendix 3.2 – Biography of Neil Coyle

37 Appendix 4.2 – Stimulus material – Neutral attack

38 Appendix 4.4 – Stimulus material – Gendered attack

Appendix 5 – Distribution of mean favorability by treatment group (attack or no attack) and oneway analysis of variance n M SD Non-attack 107 4.42 1.23 Attack 105 1.97 1.11 Total 212 3.21 1.69 Source df SS MS F p Between groups 1 317.88 317.88 230.99 .000 Within groups 210 288.99 1.38 Total 211 606.87

39 Appendix 6 – Distribution of mean favorability by treatment group (attack) and moderator (gender) and test of between subjects effects

n Mean Standard error

Non-attack Non-gendered 39 4.70 .15 Gendered 42 4.13 .15 Attack Non-gendered 43 2.52 .16 Gendered 42 1.47 .15 Source df SS MS F p Attack variable 1 310.38 310.38 256.67 .000 Gender variable 1 34.68 34.68 28.68 .000 Moderation (attack variable x gender variable) 1 2.99 2.99 2.48 .117 Corrected model 3 355.34 118.45 07.95 .000 Error 208 251.52 1.21 Corrected total 211 606.87