Self-reported and physiological responses to hate speech and criticisms of systemic social inequality: an investigation of response patterns and their mediation…

PDF

Download

Open document

Flip pages

Summarize

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content Self-Reported and Physiological Responses to Hate Speech and Criticisms of Systemic Social
Inequality: An
Investigation of Response Patterns and Their Mediation by Demographic and Psychometric
Variables
By:
Erin Ryan
A Dissertation Presented to the
FACULTY OF THE USC GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Fulfillment of the
Requirements for the Degree
DOCTOR OF PHILOSOPHY
(NEUROSCIENCE)
August 2024
Copyright 2024 Erin Ryan

ii
Dedication
For anyone who has ever had to fight for their right to exist. For anyone who the world
knocks down, but gets up to fight one more round anyway. For anyone who has ever worked for
a better world, even though they know they won’t be there to see it. For anyone who has ever
been told that they’re too different, too dark, too queer, too disabled, too feminine, too
masculine, too fat, too thin, too foreign, too loud, too anything at all, just for daring to take up
space. Keep daring.

iii
Acknowledgements
I have many, many people to thank for helping me complete a PhD program- this is as
much your accomplishment as it is mine. First, to the multiple generations of my family who
worked, and worked, and worked so that one day, someone in their family would have the
opportunity to achieve something bigger than themselves. Many of you never even got to meet
me, but there is no doubt that I could not be here, writing this, without you. Thank you, I hope
I’ve made you proud.
Second, to the friends I’ve made along the way- my life is so much better for having you
be a part of it. Thank you for every kind word, every encouragement, every cup of coffee, every
rant you listened to. Thank you for being my friends. I can only hope that I can live up to being
as good of a friend to you, as you are too me. Special thanks to Melissa, Julia, Jodie, Rita,
Heather, and Katelyn. You truly are some of the best people I’ve ever known, and ever will.
Drinks are on me.
Third, to Kate, thank you. Thank you for being my heart and soul. You are the kindest
person I have ever met, and when you laugh, I know what it’s like to hear angels sing. You are
the sunshine in all of my days. You’ve walked with me through the end of the world and back
again, and every day I wake up in awe of the fact that I get to share another day of my life with
you. I can’t wait to see what our next chapter has in store for us.
To my sister. Thank you for always being encouraging, for all the love you’ve given me,
and for being my Ride or Die. And thank you for making sure I stay caught up with all the latest
memes. I am your biggest fan. Love you kiddo.
To my parents. I don’t know that I’ll ever be able to find sufficient words to thank you.
You’ve told me I can do whatever I set my mind to from the day I was born, and helped me

iv
make that true. And more importantly, you helped me believe that it’s true. You’ve always been
my biggest supporters, and I credit all of my proudest achievements directly to you. It may have
taken 28 years of school, but now I am finally done with all of my homework. I love you both
forever.
And finally, to Layla, Gwen, and Akko. Layla, you were simply the best. I miss you
every day. Gwen, Akko, I don’t know who I’d be without you. You’ve taught me entirely new
dimensions of love and friendship. Thank you for waking me up every day at exactly 7:30 to get
up and fill your bowls, and head out for a walk, even on the days when I didn’t know if I could.
Thank you for being so happy to see me every time I walk in the door. Thank you for being the
best companions, and filling every day with laughter and fun. Yes, I am finally done with workgrab your toys, now we can go and play.

Table of Contents
Dedication………………………………………………………………………………………...ii
Acknowledgements………………………………………………………………………………iii
List of Tables……………………………………………………………………………………...v
List of Figures…………………………………………………………………………………….vi
Abstract………………………………………………………………………………………….xiii
Introduction………………………………………………………………………………………..1
Chapter 1: Experiment 1…………………………………………………………………………46
Chapter 2: Experiment 2…………………………………………………………………………86
Chapter 3: Experiment 1+2 Combined Analyses………………………………………………120
Chapter 4: Experiment 3………………………………………………………………………..152
Discussion………………………………………………………………………………………170
Conclusions……………………………………………………………………………………..192
Limitations……………………………………………………………………………………...195
References………………………………………………………………………………………196
Appendix-All Stimuli Used…………………………………………………………………….209

v
List of Tables
Table 1: Emotion Identification Patterns………………………………………………………...54
Table 2: Experiment 1 Principal Components Factor Loadings…………………………………60
Table 3: Experiment 2 Principal Components Factor Loadings…………………………………93
Table 4: Experiment 1+2 Principal Components Factor Loadings……………………………..123
Table 5: Experiment 3 Principal Components Factor Loadings………………………………..160

vi
List of Figures
Figure 1- Experiment 1 Ratings By Stimulus Category…………………………………………57
Figure 2- Experiment 1 Women’s Ratings of General
Hate vs Misogynist Hate Stimuli……….………………………………………………………..58
Figure 3- Experiment 1 Non-Women’s Ratings of
General Hate vs Misogynist Hate Stimuli……………………………………………………….58
Figure 4- Experiment 1 Variance Explained by Each
Principal Component…………………………………………………………………………….60
Figure 5- Experiment 1 Inverse Relationships between
PC1 and Rating Values…………………………………………………………………………..63
Figure 6- Experiment 1 Inverse Relationships between
PC3 and Rating Values…………………………………………………………………………..65
Figure 7- Experiment 1 Distributions of Participants’
Ambivalent Sexism Scores……………………………………………………………………....67
Figure 8- Experiment 1 Distributions of Participants’
Social Dominance Orientation Scores…………………………………………………………...68
Figure 9- Experiment 1 Distributions of Participants’
Belief in a Just World Scores…………........................................................................................69
Figure 10- Experiment 1 Distributions of Participants’
Grandiose Narcissism Scores…………………………………………………………………….69
Figure 11- Experiment 1 Distributions of Participants’
Mental Health Scores…………………………………………………………………………….70

vii
Figure 12- Experiment 1 Means and Distributions
of the Number of Skin Conductance Responses Across Stimulus Categories…………………..71
Figure 13- Experiment 1 Number of Skin Conductance
Responses Between Gender Groups by Stimulus Category……………………………………..72
Figure 14- Experiment 1 Women’s Mean Number of Skin
Conductance Responses to Misogynist Hate vs General Hate Stimuli………………………….73
Figure 15-Experiment 1 Non-Women’s Mean Number
of Skin Conductance Responses to Misogynist Hate vs General Hate Stimuli………………….73
Figure 16- Experiment 1 Mean Number of Skin Conductance
Responses by Category between White vs Non-White Participants…………………………….74
Figure 17- Experiment 1 Mean Number of Skin Conductance
Responses by Category between Straight vs Queer participants………………………………...74
Figure 18- Experiment 1 Area Under the Curve by Category…………………………………...78
Figure 19- Experiment 1 Mean Area Under the Curve
of Skin Conductance Responses to Threat Stimuli, between Genders…………………………..79
Figure 20- Experiment 1 Mean Area Under the Curve of Skin
Conductance Responses to General Hate Stimuli, between Genders……………………………79
Figure 21- Experiment 2 Mean Ratings of All
Categories of Stimuli, from All Participants…..............................................................................90
Figure 22- Experiment 2 Women’s Mean Ratings of Criticisms
of Systemic Racism, General Hate, and Misogynist Hate Stimuli………………………………91
Figure 23- Experiment 2 Non-Women’s Mean Ratings of Criticisms
of Systemic Racism, General Hate, and Misogynist Hate Stimuli………………………………92

viii
Figure 24- Experiment 2 Inverse Relationships
between PC1 Values and Stimuli Ratings….................................................................................95
Figure 25- Experiment 2 Participants’ Distributions
of Ambivalent Sexism Scores…………........................................................................................97
Figure 26- Experiment 2 Participants’ Distributions
of Social Dominance Orientation Scores.......................................................................................98
Figure 27- Experiment 2 Participants’ Distributions
of Belief in a Just World Scores………........................................................................................99
Figure 28- Experiment 2 Participants’ Distributions
of Global Grandiose Narcissism Scores........................................................................................99
Figure 29- Experiment 2 Participants’ Distributions
of Combined Mental Health Scores…………………………………………………………….100
Figure 30-Experiment 2 Relationship between Social
Dominance Orientation Scores and Rating by Stimulus Category……………………………..101
Figure 31- Experiment 2 Participants’ Mean Number
of Skin Conductance Responses to Each Category of Stimuli…………………………………103
Figure 32 -Experiments 2 Women’s Mean Number
of Skin Conductance Responses to Criticisms of Systemic Racism,
Misogynist Hate, and General Hate Stimuli……………………………………………………104
Figure 33 -Experiments 2 Non-Women’s Mean Number
of Skin Conductance Responses to Criticisms of Systemic Racism,
Misogynist Hate, and General Hate Stimuli……………………………………………………104

ix
Figure 34- Experiment 2 Mean Area Under the Curve
of Participants’ Skin Conductance Responses to Each Category of Stimuli……………………………109
Figure 35-Experiment 2 Mean Inverse Changes in Interbeat
Interval Across Stimuli Categories……………………………………………………………..115
Figure 36-Experiment 2 Inverse Mean Changes in Women’s Interbeat
Intervals in Response to Misogynist Hate vs General Hate Stimuli……………………………116
Figure 37-Experiment 2 Inverse Mean Changes in Women’s Interbeat
Intervals in Response to Criticisms of Systemic Racism vs Misogynist Hate Stimuli…………116
Figure 38-Experiment 2 Inverse Mean Changes in Women’s Interbeat
Intervals in Response to Criticisms of Systemic Racism vs General Hate Stimuli……………116
Figure 39-Experiment 2 Inverse Mean Changes in Non-Women’s
Interbeat Intervals in Response to General Hate vs Misogynist Hate Stimuli………………….117
Figure 40-Experiment 2 Inverse Mean Changes in Non-Women’s
Interbeat Intervals in Response to Criticisms of Systemic
Racism vs Misogynist Hate Stimuli……………………………….............................................117
Figure 41-Experiment 2 Inverse Mean Changes in
Non-Women’s Interbeat Intervals in Response to Criticisms
of Systemic Racism vs Misogynist Hate Stimuli……………………………….........................118
Figure 42-Experiment 1+2 Distributions of mean
Ratings of All Stimuli Categories………………………………………………………………121
Figure 43-Experiment 1+2 Distributions of Women’s Mean
Ratings of General Hate vs Misogynist Hate Stimuli ………………………………………….122

x
Figure 44-Experiment 1+2 Distributions of Non-Women’s
Mean Ratings of General Hate vs Misogynist Hate Stimuli……………………………………123
Figure 45-Experiment 1+2 Inverse Relationships between
PC1 Values and Rating Scores between Categories……………………………………………125
Figure 46- Experiment 1+2 Distributions of Ambivalent
Sexism Scores……………………..............................................................................................128
Figure 47-Experiment 1+2 Distributions of Social
Dominance Orientation Scores…………....................................................................................128
Figure 48-Experiment 1+2 Distributions of Belief in a Just World Scores…………………….129
Figure 49-Experiment 1+2 Distributions of Grandiose Narcissism Scores…………………….130
Figure 50-Experiment 1+2 Distributions of Mental Health Scores…………………………….130
Figure 51-Experiment 1+2 Relationship between Social Dominance
Orientation and Ratings for All Stimulus Categories…………………………………………..132
Figure 52-Experiment 1+2 Distributions of Mean Number
of Skin Conductance Responses between Stimulus Categories………………………………..133
Figure 53-Experiment 1+2 Distributions of Mean Number
of Women’s Skin Conductance Responses to General Hate
vs Misogynist Hate Stimuli……………………………………………………………………..134
Figure 54-Experiment 1+2 Distributions of Mean Number
of Non-Women’s Skin Conductance Responses to General
Hate vs Misogynist Hate Stimuli……………………………………………………………….135
Figure 55-Experiment 1+2 Number of Skin Conductance Responses
between Stimulus Categories for the Psychometric Baseline Model…………………………..137

xi
Figure 56-Experiment 1+2 Distributions of Mean Area Under
the Curve of Skin Conductance Responses between Stimulus Categories……………………..139
Figure 57-Experiment 1+2 Inverse Distributions of Mean
Changes in Interbeat Interval between Categories……………………………………………...146
Figure 58-Experiment 1+2 Inverse Distributions of Women’s
Mean Changes in Interbeat Intervals between General Hate
and Misogynist Hate Categories……………………………......................................................147
Figure 59-Experiment 1+2 Inverse Distributions of Non-Women’s
Mean Changes in Interbeat Intervals between General Hate
and Misogynist Hate Categories……………………………......................................................147
Figure 60-Experiment 1+2 Distributions of Different
Emotional Response Types by Stimulus Category……………………………………………..151
Figure 61- Experiment 1+2 Individual Outputs of Pairwise
Comparison of Emotional Response Types Between Stimulus Categories…………………….151
Figure 62-Experiment 3 Distributions of Mean Ratings of Stimulus Categories………………156
Figure 63-Experiment 3 Distributions of Women’s vs
Non-Women’s Ratings of General Hate Stimuli……………………………………………….158
Figure 64-Experiment 3 Distributions of Women’s vs
Non-Women’s Ratings of Misogynist Hate Stimuli……………………………………………158
Figure 65-Experiment 3 Distributions of Women’s Mean Ratings
of Criticisms of Systemic Racism, General Hate, and Misogynist Hate Stimuli………………159
Figure 66-Experiment 3 Distributions of Queer vs Straight
Participants’ Mean Ratings of Criticisms of Systemic Racism Stimuli………………………..159

xii
Figure 67- Experiment 3 Distributions of Non-Women’s
Mean Ratings of Criticisms of Systemic Racism, General Hate,
and Misogynist Hate Stimuli…………………………………………………………………...160
Figure 68-Experiment 3 Proportion of Variance Explained
by Each Principal Component….................................................................................................161
Figure 69-Experiment 3 Inverse Relationships between
PC1 and Rating Values For All Stimulus Categories…………………………………………..163
Figure 70- Experiment 3 Inverse Relationships between
PC2 and Rating Values For All Stimulus Categories…………………………………………..163
Figure 71-Experiment 3 Relationships between PC1 and
PC2 and Rating Values For All Stimulus Categories…………………………………………..164
Figure 72-Experiment 3 Distribution of PC1 Values
Between Gender Categories ……………………………………………………………………165
Figure 73-Experiment 3 Distribution of PC2 Values
Between Gender Categories…………………………………………………………………….165
Figure 74-Experiment 3 Distribution of PC1 Values
Between Race Categories………………………………………………………………………166
Figure 75-Experiment 3 Distribution of PC2 Values
Between Race Categories………………………………………………………………………166
Figure 76- Experiment 3 Distribution of PC1 Values
Between Orientation Categories………………………………………………………………..166
Figure 77-Experiment 3 Distribution of PC2 Values
Between Orientation Categories………………………………………………………………..166

xiii
Abstract
These studies investigate how people respond to hate speech vs other types of speech in
several modalities. We explored how people rated hate speech vs neutral speech in terms of
offensiveness and upsettingness, and stress responses measured via physiological changes in skin
conductance and heart rate. We found that hate speech was consistently rated as being
significantly more offensive/upsetting than neutral speech, hate speech elicited more skin
conductance responses, with greater magnitudes, measured via area under the curve, than neutral
speech, and that hate speech elicited greater changes in heart rate than neutral speech.
We explored how personal characteristics affected responses to hate speech, including
demographics, self-relevance, and psychometric traits. We compared women’s vs. non-women’s
responses to specifically misogynist hate speech, as compared to hate speech targeting a wide
range of other groups (“general hate speech,”). This investigated whether people who were
members of the group targeted by hate speech (in this case, women) exhibited heightened
responses compared to hate speech that was not personally relevant. We found that selfrelevance did not have an effect on offensiveness/upsettingness ratings- in fact, both women and
non-women consistently rated hate speech targeting a variety of other groups to be more
offensive than misogynist hate speech. We also found that self-relevance did not have an effect
on any of the physiological responses either. Interestingly, however, all participants’
physiological responses to misogynist hate speech were more frequent, and of greater magnitude,
than their responses to general hate speech. We hypothesize that this may indicate that
misogynist hate speech might be eliciting threat assessment and salience related responses, while
general hate speech may be eliciting emotional responses, as supported by out emotion
differentiation models.

xiv
We investigated whether being a member of a non-targeted marginalized group enhanced
responses to hate speech, via an “empathy,” effect. We examined how gender, race, and sexual
orientation affected response patterns. We found that, consistent with previous literature (Cowan
and Khatchadourian 2003, Cowan et al 2008, Lo Cricchio & Stefanelli 2023, LeMaire 2014)
women consistently rate hate speech as being more offensive/upsetting than non-women. Outside
of this effect, demographics did not have an effect on rating or physiological responses to hate
speech.
We investigated how social dominance orientation, belief in a just world, ambivalent
sexism, grandiose narcissism, and mental health impacted responses to hate speech. These
psychometric variables could be used to predict offensiveness/upsettingness ratings, but had no
bearing on the increased physiological arousal experienced when reading hate speech. Social
dominance orientation consistently had the greatest predictive ability, followed by ambivalent
sexism. Belief in a just world also had strong predictive ability for ratings, but not enough to
improve the linear mixed effects model of rating that already included social dominance
orientation and ambivalent sexism as predictors. Grandiose narcissism and mental health had
some predictive ability on the model, but did not have the same predictive strength, or
consistency among datasets, that the other variables had.
We also directly compared how people responded to hate speech compared vs. how they
responded to criticisms of systemic racism: non-hateful statements describing the differential
effects of systemic racism on white vs non-white people. These statements were meant to evoke
a sense of “reverse racism,” in some participants, so that the similarities and differences between
how responses to these stimuli and actual hate speech varied. We found that criticisms of
systemic racism were rated as being more upsetting than neutral speech, but not as upsetting

xv
as hate speech. Further, we found that criticisms of systemic racism elicited physiological
responses patterns that matched the patterns observed in response to misogynist hate speech, but
not general hate speech. We again hypothesize that this may be indicative of a salience-related
response, as opposed to the emotional responses elicited by general hate speech.
We found that queer participants rated criticisms of systemic racism as less offensive
than straight participants, but that demographics outside of this did not have an impact on rating
responses. Finally, we found that the same psychometric variables that predicted ratings of hate
speech also predicted ratings of criticisms of systemic racism, but in an opposite direction, since
people’s ratings of hate speech and criticisms of systemic racism had an indirect relationship.
Finally, we created a framework for integrating recorded changes in heart rate, heart rate
variability (via RMSSD), and skin conductance to assess what types of emotional responses the
stimuli elicited, on intersecting scales of approach-avoidance, and positive-negative valence,
with attention orienting responses removed and considered separately. We found that hate speech
elicited more instances of negatively-valent emotional responses, both approach and avoidance
motivated, than neutral speech, as well as more attention orientation responses. We found that
general hate speech elicited more anger-related responses (negative-approach), while misogynist
hate speech elicited more fear/threat responses (negative-avoidance).

1
Introduction
The experiments described in this thesis have sought to increase our understanding of
how hate speech is perceived, how it impacts people on a physiological level, and how personal
characteristics can alter responses to hate speech. Few previous studies have integrated results
from self-reported assessments, psychometric data, demographic data, and physiological data,
leaving gaps in our current understanding of the implications of exposure to hate speech. We aim
to synthesize these different types of data in order to produce a more comprehensive
understanding of responses to hate speech. We studied these phenomena through the lens of hate
speech and social media/internet usage, and specifically explore the following main research
questions:
1. Is hate speech perceived as more offensive/upsetting than non-hate speech, according to
self-reports?
2. Does hate speech elicit more physiological responses than non-hate speech?
3. Do people respond differently, both via self-reported assessment and via physiological
responses, to self-relevant hate speech? For this study, hate speech that targets a group
that an individual is a member of is considered self-relevant hate speech.
4. Do people who are not being targeted by a given example of hate speech, but who are
members of a different marginalized group, exhibit an “empathy effect,” and rate hate
speech as being more offensive/upsetting than non-marginalized participants? Would this
“empathy effect,” extend to physiological responses as well?
5. Do peoples’ scores on psychometric inventories predict their response patterns to hate
speech, both in terms of self-assessment and physiological responses?

2
6. Do members of a non-marginalized group respond to criticisms of the unjust social
structures that grant them social privilege in a similar way to how they respond to hate
speech, both in terms of assessment of offensiveness/upsettingness, and physiological
responses? Would these responses mirror how marginalized groups respond to selfrelevant hate speech? Do members of marginalized groups respond the same way?
Self-Reported Assessments of Hate Speech via Rating Systems
Is hate speech more offensive than non-hate speech?
Previous studies have investigated how people perceive hate speech, but none have
completely explained the phenomenon, and few have directly explored how hate speech affects
its targets. In addition, much of the previous research surrounding the topic has produced
contradictory results, and left room for questioning the ecological validity of their results.
Our studies depart from how previous studies have been conducted by considering
several key methodological factors. First, our studies center the experiences of people being
targeted by hate speech, as opposed to focusing primarily on the experiences of bystanders, or
those using hate speech, as has more extensively been investigated (Soral et al 2017, Cowan &
Hodge 1996, Cowan et al 2008).
In bystanders, one of the most prominent research questions has examined how frequent
exposure to hate speech relates to desensitization towards, and increased tolerance of, hate
speech. For bystanders, higher frequency of hate speech exposure has been shown to lead to
increased levels of desensitization towards future hate speech, and lowered perceptions of how
offensive hate speech is. In turn, increased desensitization to hate speech has been shown to lead
to increased out-group prejudice in bystanders (Soral et al 2017).

3
Since recurrent exposure to hate speech is likely a source of stress for its target, it could
potentially have wide-reaching social and health implications for already marginalized people
(Atdijan & Vega 2005, Calabrese et al 2014, Holden et al 2014). It is important that this study
compares the responses of those targeted by hate speech with different groups of non-target
people, in order to more fully understand the experience of being regularly exposed to selfrelevant hate.
Many studies establish that hate speech is considered “harmful,” but not offensive or
upsetting. Both direct (including use of slurs) and indirect (using terms like “you people,” where
the target of the speech is implied, but not explicitly stated) forms of hate speech have been
demonstrated to be considered harmful (Leets 1999, Boeckmann & Liew 2002, Leets 2002,
Cowan & Khatchadourian 2003). These results suggest that hate speech is likely to also be
considered offensive, since they establish that hate speech is perceived differently, and more
negatively, than non-hate speech.
While “harmfulness” vs “offensiveness” or “upsettingness” may sound similar, they have
very different possible applications, especially with regards to how hate speech affects one’s
internal emotional states in response. We posited that asking about harmfulness asks participants
to engage with the stimuli in an external way- they aren’t necessarily being asked if they think
hate speech harms themselves, or what type of harm it might specifically cause, only whether it
could be considered harmful in general. The implications of these results are certainly valuable
for understanding how hate speech is culturally perceived, but these past research questions did
not directly address the question of how hate speech is perceived internally by an individual-how
they assess whether they themselves were specifically harmed by such speech. Our studies thus
aimed to specifically probe this separate line of questioning.

4
Do People Respond Differently to Self-Relevant Hate Speech?
Marginalized Out-Group Members’ Responses to Hate Speech- Stigma by Prejudice
Transfer and Empathy
The next research question we wish to address is whether or not people of one
marginalized identity assess hate speech against a marginalized group they are not a member of
as being more offensive/upsetting than people who are not members of a marginalized group.
One of those most prominent examples of this phenomenon is that women are also
known to be less tolerant of hate speech than men in general, regardless of whether they are also
a member of the group being targeted or not. The exact mechanism underlying this trend is not
entirely understood, however. One factor that has been shown to mediate these gender
differences were scores on inventories measuring empathy, with women generally being more
empathetic than men (Cowan and Khatchadourian 2003, Cowan et al 2008).
Further, girls have been shown to be less likely to commit ethnically motivated bullying
against out-group minorities in real life than their male peers. Boys demonstrate a positive
correlation between exposure to ethnically targeted hate speech, and themselves perpetuating inperson bullying of ethnic minorities. For boys, this correlation is, however, mediated by social
tolerance. Girls, however, are unlikely to carry out ethnically motivated bullying, regardless of
their prior exposure to hate speech, or their social tolerance levels. (Lo Cricchio & Stefanelli
2023). Heterosexual women have also been shown to find anti-LGBT hate speech more
disturbing than heterosexual men, and to have more favorable attitudes towards LGBT people.
This suggests that women may not only tend to have higher degrees of empathy for out-group
members, they may have a greater appreciation for social diversity (LeMaire 2014).

5
Boeckmann & Liew 2002 demonstrated that this potential empathy effect between
marginalized out-groups is not limited to women, but also exists between separate marginalized
racial groups as well. They found that Asian American participants considered anti-Asian and
anti-Black hate speech to be harmful, but did not demonstrate a significant difference between
how harmful both types of hate speech were. They hypothesized that this was due to shared
minority status.
We suspected that these observed effects may have been at least partially a result of a
phenomenon known as stigma by prejudice transfer. Sanchez et al 2017 demonstrated that if
white women learned that a given individual was racist, they expected them to also be sexist,
exhibit more social dominance orientation-related traits, and to treat them more poorly. Black
men, upon learning that a given individual was sexist, expected that the individual was more
likely to be racist, exhibit more social dominance orientation-related traits, and treat them more
poorly. Conversely, white men did not exhibit these expectations (Sanchez et al 2017). It thus
follows that marginalized people have a learned, intuitive sense of the existence of social
dominance orientation, and its influence on how people are likely to treat marginalized people,
even if they have never heard of the term “social dominance orientation”.
Additional studies demonstrate that investment in group identity has been found to
mediate the perceptions of harm from hate speech, with higher investment resulting in higher
perceived harm when the in-group is targeted, and higher social preference for the prioritization
of in-group members (Sidanius et al 2010). Perhaps, then, there is a sense of common in-group
identification across marginalized communities, driven by shared experiences.
We thus hypothesized that members of one marginalized group, being aware of the
similarity of experiences they shared with members of other marginalized groups, would also be

6
more likely to assess hate speech targeting marginalized outgroup members as more
offensive/upsetting than non-marginalized individuals would, and more likely to rate criticisms
of systemic racism as being less offensive/upsetting than non-marginalized people.
Responses to Criticisms of Systemic Inequity
Also crucial to understanding the dynamics of hateful interactions is understanding how
non-marginalized people respond to identity-based criticism. Often, when a marginalized person
speaks broadly about the systemic abuse they experience, the people who benefit from those
power structures take personal offense, even when they are not specifically mentioned, or only
passively benefit from these power structures (eg: they do not attempt to promote these
structures, but still benefit from the privilege granted by them). A well-known example of nonmarginalized people responding in this way is men’s usage of #NotAllMen in response to
women’s online discussions of how men, enabled by, and perhaps as a result of, our patriarchal
society, have harmed women (“Not all men: How discussing women’s issues gets derailed”,
Hayes & Luther 2018).
While social stigma and shame are usually studied in the context of hardships and
inequalities faced by minorities, they may also be applicable to this “Not All Men” phenomenon
as well. Stigma has previously been defined as what occurs when, due to an action or
characteristic, a person is socially transitioned “from a whole and usual person to a tainted,
discounted one.” (Major & O’Brien 2005). We hypothesize that when minorities refer to the
abuse they receive under a collective label, members of the antagonizing group may experience
shame and stigma, and then try to prove why they are not deserving of such stigma through “Not
All Men,” type discourse. They may also realize that they engage in the abusive behaviors

7
mentioned, even if they were not intentionally being malicious, and be faced with further
identity-threat derived shame.
This newfound stigma may result in feelings of shame, and a sense of ingroup-inferiority,
as they see their identity groups publicly devalued. Previous research has shown that a sense of
ingroup-inferiority creates a sense of frustration, anger, and diminished self-worth, and can lead
an individual to attempt to further devalue or compete with the outgroup. This divide can further
result in feelings of pleasure at the failings or suffering of these outgroup members (Leach &
Spears 2008). This is significant because it belies that these majority-group members do in fact
have an emotional investment in the status and power of their in-group identity. This investment
is important, as it may reveal deeper motivations for why people are seemingly so eager to
retaliate against criticisms. Both the satisfaction and individual gains from group inclusion and
validation, as well as the emotional pain of exclusion or criticism from their group, have been
found to be directly proportional to an individual’s degree of investment in their group identity,
emphasizing again the importance of investment (Bernstein et al 2010, Boeckmann & Liew
2002).
Power is also important in that it can dictate the “rules,” of these interactions, as it
determines how society will view and judge each participant, and how they will be punished, if
applicable. This can also embolden the party with more social power. For example, Jacobsen et
al 1994 found that in physically abusive relationships, the person with more power was always
the first to turn violent, even if they didn’t “start” the fight.
This leads us to a discussion of how power can differentiate between criticism and abuse.
While criticizing a majority group may result in emotional distress and stress responses in its
members, as shame and stigma have been found to elicit (Dickerson et al 2004), this should not

8
be equated with discrimination and harassment. When majority group members use their social
power to abuse marginalized people based on a facet of their identity, it elicits the same stress
responses (Dickerson et al 2004). Conversely, in the specific interactions we are examining in
these studies, marginalized people are: 1) criticizing people for the verifiable behavior of their
social group, not their identity, or false accusations and mischaracterizations of their social
group’s behavior, and 2) fighting against their own oppression, not oppressing others.
Understanding the similarities and differences in the perceptions and responses of marginalized
and non-marginalized people in response to both hate speech, and criticisms of power structures
and those who perpetuate them will enable us to devise more effective interventions to both
mediate the potentially harmful impacts of chronic exposure to hate speech, and to prevent the
spread of hostility towards marginalized groups by majority groups when power-structures are
criticized.
This defensive type of response to criticism of one’s privileged identity also has
implications for real-world harm and further marginalization of those speaking out about their
experiences. In the wake of the “#NotAllMen movement, many social media platforms began
enforcing bans on “gender-based hate.” In practice, however, what the scientific literature has
named, “gender-based censorship,” (Nourik 2019). While misogyny and sexual harassment of
women were quantitatively measured, and shown to still be rampantly present at the same levels
on these platforms, researchers recorded abundant cases of women speaking out about their
experiences with men, or criticizing men as a whole for the way male privilege had been used to
abuse them, found that their posts were being removed. They were told that their posts were
specifically in violation of the platforms’ Terms of Service Agreements regarding the use of
“gender-based hate.” Thus, even after the implementation of supposed gender-based protections,

9
the same patriarchal systems that had not prioritized their safety or experiences with harassment
online still existed, and continued to deprioritize their needs. Women still reported facing the
same frequency of abuse and hate online that they did before, but now found themselves unable
to talk about it, in the name of equality (Nurik 2019). This example illustrates why
considerations of social power and privilege are vital when discussing hate speech, and what
constitutes it.
Finally, previous works have demonstrated that social power/status does have an impact
on how explicitly, and harmfully, the intentions of the speaker are perceived to be, though the
exact mechanism of that impact remains unclear. Leets et al 1999 found that Caucasian
participants viewed both direct and indirect statements aimed at Caucasian people to be equally
harmful, but did not view the indirect statements aimed at Asian participants to be as harmful.
Conversely, Asian participants viewed indirect anti-Asian statements as more harmful than direct
anti-Asian statements. All participants also viewed explicit and implicit statements aimed at
Caucasian participants to be equally explicit, while Caucasian participants considered indirect
statements spoken to an Asian person to be less explicit than Asian participants did.
Roussos & Dovidio 2018 investigated how participants with high and low anti-Black bias
differed in considering anti-Black speech vs anti-White speech a hate crime vs being protected
by US free speech laws. In their first experiment, they found that participants high in anti-Black
bias were less likely to consider anti-Black speech a hate crime, as compared to anti-White
speech. People high in anti-Black bias were also more likely to consider anti-Black speech as
protected free speech, as opposed to anti-White speech. Conversely, the opposite patterns were
found in participants low in anti-Black bias. They hypothesized that this may be explained by

10
participants with low anti-Black bias having a higher level of white-privilege remorse, and/or
having a higher sensitivity to the history of anti-Black persecution in the US.
In Roussos & Dovidio 2018’s second experiment, a replication of their first experiment,
they did not replicate these results. In this experiment, all participants considered anti-White
speech to be less protected by free speech, and more likely to consider anti-White speech a hate
crime, as compared to anti-Black speech. The results of these experiments are contradictoryhowever, they point to social privilege and the strength of in-group identity having an impact on
how prejudice against marginalized vs majority groups is perceived.
While there might not be an established understanding of how social power dynamics
directly affect perceptions of hate speech, there is previous research that expands on this
knowledge base, and allows us to make hypotheses about how non-marginalized people will
respond to non-hateful critiques of their social groups. For these experiments, we investigate
how white vs non-white participants respond to criticisms of systemic racism, and we
hypothesize that white participants are more likely to consider criticisms of systemic racism to be
more offensive/upsetting than non-white participants.
Psychometric Variables and Responses to Hate Speech
Psychometric Variables Measured
Investigating how personality traits correlate with perceptions of harm or magnitude of
response to hate speech has also not been extensively explored. Here, we have chosen to assess
subjects’ scores on Grandiose Narcissism, Belief in a Just World, Ambivalent Sexism, and Social
Dominance Orientation inventories, and correlate those scores with how offensive they rated
hate speech to be, how offensive they rated criticisms of systemic racism to be, and whether their

11
scores on these inventories had a relationship with their physiological responses. Below, we
describe the individual psychometric variables measured in detail.
Belief in a Just World (BIAJW) investigates the extent to which an individual believes
that a person’s privilege or circumstances are determined by how hard working they are, or how
good of a person they are, and vice-versa (Lerner 1980, Lipkus 1991). BIAJW attitudes have a
complicated relationship with perceptions of the victims of bullying, aggression, and hate.
People with high scores on Global BIAJW, when asked in the abstract, are less tolerant of
bullying than people who score lower. After all, if one believes that bad deeds deserve
punishment, then it follows that they would endorse punishments for acts of cruelty. (Dalbert
2001,Voss & Newman 2021). Sang et al 2023 specifically investigated whether there was a
relationship between an individual’s score on BIAJW inventories, and how frequently they
engaged in cyberaggression. They found that there was an indirect relationship- the more one
endorsed BIAJW views, the less likely they were to engage in cyberaggression. However, this
only captures one aspect of people’s behavior, not the whole picture.
In order to preserve their world view, people with strong BIAJW attitudes may infer that
targets of hate speech and victims of hate crimes “deserved” their treatment. It perhaps also
logically follows that, when confronted with someone being abused, someone who endorses
BIAJW might presume that the victim deserved their punishment- if they were undeserving of it,
it wouldn’t be happening to them. This is a coping mechanism of sorts- if an observer cannot
find a way to restore justice by standing up for victims, punishing abusers, etc., then they can
“restore justice psychologically,” by concluding that the abusers were right to harm their victims
(Dalbert 2001, Voss & Newman 2021).

12
This can also lead to the adoption of further discriminatory attitudes towards entire
groups, when the targets of hate are entire populations. There is even evidence that, in order to
maintain their BIAJW views, people will actually commit acts of hate against members of
targeted groups, since they have mentally concluded that, since the world is a just place, the
targets deserve to be punished (Dharmapala et al 2009). This study aims to further investigate
how BIAJW views specifically impact perceptions of online hate speech. Given that this study
relates to perceptions of hate, and not committing acts of hate, we hypothesize that participants
who endorse BIAJW views will engage in the “coping” behavior of presuming the targets of hate
speech deserve to be punished, and will not assess hate speech as being as upsetting as
participants who do not score high on the BIAJW endorsement inventory.
The ambivalent sexism scale investigates how much an individual endorses both benign
and harmful sexism, where benign sexism includes ideas such as women being inherently more
delicate and weaker than men, and it thus being the responsibility of men to protect and provide
for them. The attitude may call for women being treated well and cared for, but the reason
behind this attitude is an inherently sexist one (Glick et al 1996). There is little current research
investigating how endorsement of ambivalent sexism interacts with perceptions of general hate
speech. However, quite a bit of research has been done to explore how ambivalent sexism
correlates with Social Dominance Orientation, Belief in a Just World, and perceptions of sexism
and sexist/gender-based harassment. Even among women, perceptions of both mild and lewd
street harassment by men were mediated by hostile sexism scores, though not benign sexism
(Moya-Garofano et al 2022.)
Russell & Trig 2004 demonstrated that hostile sexism levels were the strongest predictors
of tolerance of sexual harassment- even stronger than gender itself, where in-group/out-group

13
dynamics were at play. They also found that social dominance orientation scores were a
predictor for tolerance of sexual harassment, though not as strong of a predictor as hostile sexism
or gender. Christopher et al 2013 further demonstrated that social dominance orientation can
actually mediate the strength of hostile sexism, while right wing authoritarianism predicted
benevolent sexism scores. Rollero et al 2021 also found that social dominance orientation and
ambivalent sexism scores were predictors of whether physical violence, emotional violence, and
the restrictions of freedoms of women were recognized as harmful, and were better predictors
than gender itself. However, gender was still a significant predictor- thus, the predictive ability
of gender is mediated by social dominance orientation and sexism scores.
Further, both hostile and benign sexism scores, and belief in a just world scores predict
whether someone will engage in victim blaming in cases of sexual violence (Pederson &
Stromwall 2013, Feather & McKee 2012, Sakalli-Ugurlu et al 2007). Belief in a just world
scores were also shown to predict attitudes towards victims and perpetrators of domestic
violence. In instances where participants were told what incited a male partner to violence
towards a female partner (wanting to separate, wanting to visit male friends, and wanting to take
trips with female friends without their partner), high belief in a just world scores predicted lower
victim blaming and lower exoneration of the perpetrator. However, if no mention of what
instigated the perpetrator to violence was given, there was no longer an observable relationship
between belief in a just world scores and victim blame/perpetrator exoneration. (Valor-Segura et
al 2013). This could suggest, then, that while people with high belief in a just world scores do
not condone “unnecessary” violence against women, they operate on a “guilty until proven
innocent” framework, where women are more likely to have “deserved” the violence, in keeping

14
with the belief in a just world ideology- if they weren’t guilty, why would they have been
punished?
We thus hypothesize that ambivalent sexism will predict how people rate the
offensiveness/upsettingness of hate speech, especially misogynistic hate speech. We hypothesize
that gender will also act as a predictor of ambivalent sexism, and of ratings of hate speech’s
offensiveness. However, we also hypothesize that social dominance orientation and belief in a
just world scores will mediate the relationships between gender, ambivalent sexism, and
perceptions of hate speech.
Narcissism, outside of Narcissistic Personality Disorder, can describe how much more
importance an individual may place on themselves and their wants, feelings, etc., as compared to
everyone else. Here, we use the revised Grandiose Narcissism Scale, which measures seven
aspects of narcissism: entitlement (belief that one is owed what they want), exhibitionism (belief
that one should be noticed and paid attention to), authority (belief that one should be in charge),
self-sufficiency (belief that one does not need others), superiority (belief that one is better than
others), vanity (how one self-evaluates), and exploitativeness (the belief that one can take
advantage of others because it is what they are owed), (Foster et al 2015).
Narcissism has demonstrated relationships with the use of offensive language. People
with high scores on assessments of narcissism were asked to assess how attention-grabbing they
thought different offensive words were, how often they used those words, and how offensive
they thought the words were. People with high narcissism scores were found to rate offensive
words as being more attention-grabbing, and less offensive than people with low scores, and
used the words more frequently than people with low scores. This suggests that people with high
narcissism scores are more likely to use offensive language as a way to seek attention, and that,

15
in general, they are less sensitive to the offensiveness of such language (Adams et al 2014). We
thus hypothesize that participants who score high on the Grandiose Narcissism inventory will
rate hate speech as being less upsetting than participants who score low on the narcissism
inventory.
However, narcissism also has known relationships with perceptions of hate speech usage.
In a study where Dark Triad traits (Psychopathy, Machiavellianism, and Narcissism) were
measured and compared to ratings of how acceptable different “deviant” behaviors were on
social media, psychopathy and Machiavellianism were positively correlated with ratings of hate
speech usage as being acceptable. Narcissism, however, was not (Withers et al 2017). We posit
that, while the results of this study may seem to contradict the results of the previous study, they
actually may support the previous study. Based on the previous results, people who score high on
narcissism know that using offensive language breaks social acceptability rules, and seem to
specifically engage with it as a means of getting attention. It would follow, then, that they would
score hate speech as not being socially acceptable, even if they do find it less offensive than
people with low narcissism scores. Additionally, knowing that something is not socially
acceptable, does not automatically translate to abstaining from the behavior. Thus, we maintain
our previously stated hypothesis.
A form of narcissism called collective narcissism has known interactions with other
forms of prejudice, especially in the form of collective narcissism. Collective narcissism is
essentially narcissism held in common by members of a given group, usually a gender, religion
or nationality. This group holds its identity as being superior to others, and entitled to increased
external recognition, authority, respect, and resources, in much the same way that a narcissistic
individual might. Collective narcissism differs from personal narcissism in an important way,

16
though- despite these groups commonly not facing any actual marginalization, it stems from a
collective insecurity, and view that one’s group is not being respected enough, or given enough
recognition/social deference from groups considered to be inferior (de Zavala & Bierwiaczonek
2020, Cichoka & Cislak 2020, Cichoka 2016, de Zavala et al 2009).
Collective narcissism differs from healthy levels of pride for one’s membership in a
group, eg: personal heritage, sexual orientation, religious identity, etc., by the insecurity of its
members, and perception of threat from outgroups. Healthy pride results from security in one’s
identity, lack of need for external validation in the form of deference, and lack of perceiving the
existence (but not necessarily the actions) of outgroups as threats to the ingroup receiving the
external validation they believe they deserve (de Zavala & Bierwiaczonek 2020, Cichoka &
Cislak 2020, Cichoka 2016, de Zavala et al 2009). Healthy collective pride also specifically does
not result in harm towards either ingroup or outgroup members, while collective narcissism
specifically seeks harm against outgroups, and often harms ingroup members in the process
(Cichoka 2016). Collective narcissism has been shown to result in intergroup aggression,
endorsement of the use of military action against outgroup members, and endorsement of traits
such as right wing authoritarianism, social dominance orientation, and uncritical patriotism (de
Zavala et al 2009).
Demonstrating these concepts, collective narcissism with regards to male gender identity,
Catholic religious identity, and national identity among Polish participants was correlated with
higher endorsement of both benevolent and hostile sexism, in both men and women. Feelings of
collective narcissism among Catholics was also associated with greater tolerance of physical
violence against women, among both men and women. Conversely, feelings of security in one’s
gender, religious, or national identity amongst Polish participants was found to result in low

17
feelings of collective narcissism, and low tolerance of sexism and violence against women (de
Zavala & Bierwiaczonek 2020).
Given the inherent aspects of social dominance orientation and individual narcissism
required for collective narcissism to exist, we expect that these aspects will be captured both by
individual grandiose narcissism measures and social dominance orientation measures, and will
result in negative relationships between scores on narcissism and social dominance orientation
inventories with assessments of the offensiveness/upsettingness of hate speech, and positive
relationships between these scores and assessments of offensiveness/upsettingness of criticisms
of systemic racism.
Social Dominance Orientation describes to what degree an individual supports and
reinforces social hierarchies, and the belief that some groups of people are “better” than others
(Sidanius et al 1994a, Sidanius et al 2010, Sidanius et al 1994b, Pratto et al 2000). In addition to
having complex relationships with the previously described variables, social dominance
orientation has well-established relationships with tolerance of hate speech and prejudicial
attitudes, such as racism.
In fact, social dominance orientation can act as a predictor of endorsement of prejudicial
attitudes writ large, not just against marginalized groups considered individually. Social
dominance orientation is a likely mediator for why people who exhibit prejudice against one
given marginalized group are more likely to exhibit prejudice against another marginalized
group. (Pratto et al 1994, Castellanos et al 2023). High levels of social dominance orientation
have been demonstrated to have a direct effect on rates of in-person and online hate speech
usage. It has also been shown to have oppositional effects on other factors known to decrease
hate speech usage, such as empathy, altruism, and tolerance (Castellanos et al 2023, Pratto et al

18
1994). Further, social dominance orientation scores were found to predict support for, and
participation, in real life violence against marginalized groups (Gordon 2021, Schrader et al
2024). We hypothesize, then, that social dominance orientation scores will have an inverse
relationship with assessments of offensiveness/upsettingness of hate speech, but a direct
relationship with assessments of offensiveness/upsettingness of criticisms of systemic racism.
We will measure symptoms of depression, anxiety, and trauma using the DSM-5 SelfRated Level 1 Cross-Cutting Symptom Measure—Adult (American Psychiatric Association
2013) and the PCL-C for DSM-IV (Weathers et al 1993), to investigate whether these additional
traits may be affecting physiological responses. Anxiety and depression are known to decrease
heart rate variability, which could impact participants’ physiological outputs during the
experiment, or could result in lessened physiological responses to stimuli (Chalmers et al 2014,
Gorman & Sloan 2000, Pittig et al 2013).
We also aim to determine whether there is a relationship between overall mental health
and ratings of how upsetting or offensive stimuli containing hate speech are. Right-Wing
Authoritarianism, which has known correlations with social dominance orientation (Rocatto &
Ricolfi 2010, Mayer et al 2020), narcissistic rivalry (Mayer et al 2020, Grigoropoulos 2023), and
belief in a just world (Grigoropoulos 2023), has been demonstrated to lead to an increase in
depressive symptoms over time (Duriez et al 2012). RWA has also been demonstrated to have a
negative correlation with having positive emotions, even when degree of social fit in one’s
community was controlled for (Van Hiel & Kossowska 2006). Thus, we hypothesize that there
may be a correlation between mental health and ratings of how offensive or upsetting hate
speech is perceived to be.

19
This work will build on the existing literature by examining how these traits are
associated with not just each other, but also social power and privilege, perceptions of harm via
hate speech and criticisms of power structures to others and themselves, and social media
behavior. Previous studies that investigated the relationships between narcissism and social
dominance orientation often investigated how these traits correlate with each other, the Big Five
personality traits, and right-wing authoritarianism through the lens of identifying a personality
trait that makes an individual more likely to ascribe to social dominance orientation, prejudice,
and support of right-wing authoritarianism, as opposed to seeking to understand how these traits
may or may not correlate with actual behavior in more natural settings (Cichoka et al 2017, Di
Meo 2007, Mayer et al 2020, Bizer et al 2012). This approach has left several gaps in our
understanding of how these traits relate to each other as behavioral predictors, not just as
predictors of how someone will or won’t endorse a given ideology.
Bilewicz et al 2017 provides an example of the importance of extrapolating how these
traits relate to each other into behavioral tasks. Their study examined how social dominance
orientation and right-wing authoritarianism scores predicted opinions on whether or not hate
speech usage online should be punished. Among the study participants, right wing
authoritarianism and social dominance orientation scores both correlated strongly with out-group
prejudices. While social dominance orientation and right-wing authoritarianism are known to
have a strong positive correlation with each other, participants with high social dominance
orientation, but low right wing authoritarianism, opposed punishments for hate speech usage.
Participants with high right wing authoritarianism scores, however, strongly supported
punishment for hate speech use. The researchers hypothesized that this may be due to right wing
authoritarianism instilling a strong disapproval of violating cultural norms, regardless of whether

20
the individual agreed with the statements. Thus, we see the importance of investigating how each
of the previously named traits impacts behavior, and not treating them as interchangeable,
despite the high correlation between inventory scores, when we investigate what drives
responses to hate speech.
We also performed principal components analyses (PCA), a data reduction technique, to
investigate the relationship between these variables and responses to the stimuli. PCA take
several component factors, in this case, the psychometric variables, and create new principal
components, composed of varying amounts of influence from the component factors, to use as
predictive variables.
Since all of these psychometric variables often have correlational relationships, if they
were found to have explanatory relationships with responses, there may have been overlaps of
explained variance between multiple variables. By conducting data reduction, we would be able
to study the predictive abilities of each variable outside of the effects of overlapping variance, in
order to determine if any of the psychometric predictors could or should be removed from the
models due to not actually being a behavioral predictor, only having overlapping explanations of
variance with other predictors. In practice, however, these analyses created principal components
that largely showed equal contributions to explaining response variance from almost all the
psychometric variables. As a result, to better understand which psychometric variables were the
most responsible for observed variance in responses, we instead used more traditional methods
of examining whether each variable independently improved our models, and to what extent.
Physiological Responses to Hate Speech
In addition to asking participants to rate how offensive/upsetting they found different
stimuli to be, we recorded physiological data, in order to gain insights on what their internal

21
states were. Rating scores alone don’t necessarily tell the whole story of how someone is feelingmuch emotion happens at the subconscious level. Physiological responses can reflect these
processes (Isenberg 1999). Additionally, sometimes people don’t accurately identify their
feelings, either unintentionally/because they are genuinely unaware of what their feelings are, or
because they have a desire to conceal them (Lang et al 1983). Gender roles and social pressure
often lead men to reduce their emotional expression, and downplay their feelings (Carlton et al
2019). People may also conceal their true emotions in order to conform with social expectations
and norms regarding what emotions are appropriate to feel in certain situations (Bica 2022,
Scheff 1988, Heerdink et al 2013). Mental health disorders, such as depression, alexythemia,
suppression of emotional expression, and anxiety can impact someone’s ability to accurately
identify their feelings, or the intensity of their feelings, as compared to their physiological
responses (Vanman et al 1998, Brown et al 2020, Zhu et al 2019).
This may explain why in previous studies, such as Boeckmann & Liew 2002, people who
are members of the group addressed by hate speech seem to not be very strongly impacted.
Culturally, there is pressure to “keep a stiff upper lip,” and “not let it get to you.” These selfreported responses may be more indicative of that than of how people are actually feeling.
Conversely, non-targeted people can know that agreeing with such a statement is “wrong,” and
can feel pressure to respond as such due to actual empathy, or due to the desire to be perceived as
“good,” by society, whether they endorse such statements or not. This anxiety is well
documented via behavioral and physiological responses in individuals interacting with
stigmatized others. Subjects were also more accommodating to these stigmatized others
(different race, prominent facial birthmark, perceived lower socioeconomic status) than to
ingroup members (Blascovich et al 2001, Paolini et al 2015). As such, relying solely on self-

22
reported responses may not be presenting researchers with an accurate idea of how subjects feel.
Thus, we supplemented the rating data with physiological recordings in order to gain the best
possible understanding of how people cognitively and emotionally responded to hate speech.
The autonomic nervous system (ANS) is largely responsible for all the involuntary
processes happening in an organism. The ANS is divided into two main branches, the
sympathetic and parasympathetic pathways. Together, these branches regulate processes in
almost every tissue in the body. The sympathetic nervous system is largely responsible for
initiating responses to stress-inducing situations, or for creating the “fight or flight,” response.
The parasympathetic nervous system, conversely, works to promote “rest and digest,” responses.
The efforts of both systems are what allow us to both adapt to stressful stimuli in our
environments, and to restore homeostasis once the stressor is gone. (McCorry 2007). Thus, many
of these measures provide a non-invasive way to better understand someone’s internal states,
especially with regard to stress and emotional responses. Our experiments utilized some of the
most well-known measurements: skin conductance, heart rate, and heart rate variability.
Do People Produce More Physiological Responses to Hate Speech than NonHate Speech?
Electrodermal Activity and Skin Conductance
Changes in electrodermal activity are regulated solely by the activity of the sympathetic
nervous system, and, consequently, are a metric of increases or decreases in stress and arousal.
These changes can occur in response to multiple types of arousal, including changes in emotional
state, cognitive load, and attention orientation (Lohani et al 2019, Nourbaksh et al 2012, Nagai et
al 2004, Cuthbert et al 2002). These sympathetic responses are often at the subconscious level,

23
and are thus an extremely useful tool for measuring internal states that participants may not even
be aware of to articulate.
The sympathetic portion of the autonomic nervous system innervates the eccrine sweat
glands in the skin. (Critchley 2002, Magai et al 2004). The alteration in the activity of these
glands alters the electrodermal activity of the skin. One aspect of electrodermal activity is skin
conductance, aptly named as it measures the changes in the electrical conductivity of the skin.
When recording skin conductance, a small electrical current is run through the skin via
electrodes, and the level of resistance the skin produces is recorded. While participants are
completing a research-related task, each sweat tube associated with the eccrine sweat glands
increases or decreases sweat production in response to sympathetic nervous system activity.
Each gland acts as an electrical resistor in a parallel series when activated. The measure of skin
conductance, then, is representative of the sum total electrical resistance, and is indicative of the
magnitude of eccrine activation, and thus sympathetic activation of the ANS. (Nourbaksh et al
2012, Calvo et al 2017, Aparicio-Betancourt et al 2017). These alterations in conductivity reflect
the magnitude of ANS activity, and thus, the magnitude of the stress response (Critchley 2002,
Aparicio-Betancourt et al 2017).
Electrodermal activity measurements can be broken down further into multiple different
components, which are used for different types of research. The two main branches of
electrodermal activity measurements are tonic and phasic components (Lohani et al 2019,
Braithwaite et al 2015, Aparicio-Betancourt et al 2017). Tonic measurements are what the
overall skin conductance level is derived from, and indexes electrodermal activity on a broad
scale. Changes in the overall skin conductance level are slower than changes in the phasic
components of electrodermal activity, are a measure of the overall state of bodily arousal, and

24
are associated with longer-term emotional or cognitive load states. (Lohani et al 2019, AparicioBetancourt et al 2017). Changes in the phasic components of electrodermal activity represent
much faster changes, and responses to specific stimuli, as opposed to representing one’s overall
emotional state. These responses, known as skin conductance responses, can even be used in
research to measure a person’s subconscious arousal to an experimental stimulus or set of
conditions (Lohani et al 2019, Braithwaite et al 2015, Critchley 2002).
The presence of the skin conductance response (SCR) is inherently an indicator that some
type of a cognitive response has occurred. However, not all responses will represent the same
magnitude of response (Smets et al 2019, Aparicio-Betancourt et al 2017). There are multiple
ways to quantify the magnitude of a response using skin conductance, including the number of
SCRs produced, the peak height of the SCR, the duration of the SCR, and the area under the
curve of the SCR (Bach et al 2010).
Each SCR, representing a temporary increase in electrical conductivity of the skin, has
the same basic composition: the waveform onset, the waveform rise, the waveform peak, the
waveform fall, and the waveform end (Braithwaite et al. 2015). The intensity of the response,
then, is represented in the rate of SCR production, how quickly the response reaches its peak, the
amplitude of the peak, and the duration of the response (Smets et al 2019). Using the area under
the curve of the response is a good way to create a composite representation of all of these
aspects, and can provide a more balanced interpretation of the response than examining the max
amplitude, duration, or number of SCRs individually (Bach et al 2010).
The use of skin conductance as a reliable measure of cognitive load is a fairly recent
development. Studies have demonstrated the efficacy of using skin conductance to measure
cognitive load by having participants complete several kinds of tasks, including reading

25
comprehension, arithmetic, and auditory presentation-verbal response working memory tasks,
with multiple tiers of difficulty. They found that skin conductance levels changed in response to
these tasks, and that the magnitude of the response increased with each increasing level of task
difficulty. Prior to this, due to incorrect ways of analyzing skin conductance outputs, and
problems with study designs (eg: tasks that were not challenging or engaging enough to
accurately reflect cognitive load), there was not solid evidence that skin conductance had a
proportional relationship with cognitive load (Mehler et al 2012, Nourbakhsh et al 2012).
Skin conductance is regulated by several brain regions. An indirect correlation between
skin conductance level (SCL) and neural activity in the ventromedial prefrontal cortex
(VMPFC), and the orbitofrontal cortex (OFC) has been observed. As SCL decreases, activity in
these regions increases, regardless of whether the subject is engaging in a mental task or not; this
is unsurprising considering that as cognitive load increases, activity in the VMPFC and OFC
ceases. Thus, these areas are believed to be associated with a default mental and emotional state
when an individual is awake. This was found to be in contrast to the production of SCRs. SCR
production is associated with changes in the internal emotional, cognitive, or physical states
(Lohani et al 2019, Nourbaksh et al 2012, Nagai et al 2004, Cuthbert et al 2002). SCR
production is positively correlated with activity observed in the thalamus, hypothalamus, lateral
prefrontal cortex, anterior striate, extrastriate visual cortices, anterior cingulate, and the insular
cortices (Nagai et al 2004).
Changes in skin conductance are also well-documented as being directly correlated with
amygdala activity. The amygdala is primarily responsible for producing fear and threat
responses. It is no surprise, then, to learn that its activity modulates the activity of the peripheral
nervous system in response to threatening and salient stimuli. When the amygdala is activated by

26
the presence of a threatening stimulus, SCRs are produced in response (Lazarus et al 1963,
Wood et al 2014). The amygdala has also been shown to be partially responsible for assessing
the emotional context of a given stimulus: Wood et al 2014 showed that the amygdala helped
mediate the skin conductance response when both a perceived threat AND emotionally arousing
stimuli (both positive and negative) were present, demonstrating the amygdala is involved in
contextualizing the environment.
SCRs can also be elicited by written words. Isenberg et al 1999 demonstrated that a
normed list of words associated with threat, when presented alone in written form, elicited
amygdala activity, and thus a subconscious fear response. This importantly demonstrates that a
stimulus need not be capable of inflicting bodily harm on a person for threat responses to be
generated. While a word on a screen can’t do anything to physically hurt someone, its linguistic
meaning can convey messages of threat, which the body responds to in a similar way as it would
if a physical threat were present. We thus hypothesize that exposure to written hate-based
rhetoric from a social media post can elicit SCRs, and will elicit more SCRs than social media
posts that do not contain known threat words or hateful/emotionally valent content.
The magnitude of the resultant SCRs also vary in direct proportion to both the level of
arousal caused by the stimulus, and the amount of amygdala activation that has been elicited
(Cuthbert et al 2002, Wood et al 2014). This responsiveness is enhanced when strong emotional
arousal accompanies a threat, whether the emotion is of positive or negative valence. When
strong feelings are felt, the same threatening stimulus elicits an SCR with a larger change in
amplitude than when the stimulus is presented in an emotionally neutral setting. This indicates
that both the threat perception and emotional regulation systems work together to modulate
responses to the environment.(Cuthbert et al 2002, Wood et al 2014). Additionally, the

27
magnitude of skin conductance response also varies proportionally to the level of emotional
arousal felt, regardless of valence, even in the absence of threat (Cuthbert et al 2002). This
supports the hypotheses that hateful content may elicit larger physiological responses than
exposure to words known to elicit a threat response, when presented in an emotionally neutral
context, and that we can measure how much exposure to hateful content emotionally arouses an
individual via skin conductance.
We hypothesize, then, that examples of hate-based rhetoric will elicit more SCRs than
stimuli that do not, and that stimuli that contain words known to elicit a threat response,
presented outside of hateful or threatening contexts, will elicit more SCRs than stimuli that
contain only neutral statements.
Heart Rate
Skin conductance is not the only physiological response that reflects changes in an
individual’s internal state. Heart rate also varies with autonomic arousal. As stress and anxiety
increase, and the sympathetic nervous system is activated, heart rate increases. As relaxation and
parasympathetic nervous system activity increase, heart rate decreases. Thus, feelings of fear and
threat are accompanied by increases in heart rate (Petry & Desdirato 1978, Wager et al 2009,
Arza et al 2019, Eisenbarth et al 2016). The resultant increases in heart rate can happen on both
short and long-term time scales. After the stimulus is removed, the parasympathetic nervous
system is activated once again, and heart rate returns to rest. If an individual is chronically
exposed to stressors, however, then the baseline heart rate itself is often elevated as a result, and
may increase the risk of, or cause, heart disease (Vrijkotte et al 2000, Schubert et al 2009, Brand
et al 2000, Noushad et al 2021, Eisenbarth et al 2016).

28
Short-term changes in heart rate have also been observed in response to transient
emotional states. Increases in heart rate have been observed when individuals read statements
with fearful content, both silently and aloud. Moreover, these increases were determined to not
be due to increases in cognitive load, as neutral sentences did not elicit that same degree of heart
rate increase as the fearful sentences (Vrana et al 1989).
Considering that hate based rhetoric is, by definition, meant to incite violence and hatred
in its supporters, and fear in its targets, we hypothesize that social media posts that contain hate
based rhetoric will elicit increases in heart rate, and that these increases will be greater than those
observed when participants are presented with social media posts with emotionally neutral
content.
However, cardiac responses to stress differ from changes in electrodermal activity in an
important way: electrodermal activity is only regulated by input from the sympathetic nervous
system, while cardiac activity is directly regulated by both the sympathetic and parasympathetic
nervous systems (Schubert et al 2009, Eisenbarth et al 2016, Iversen et al 2010, Ataee et al
2014). Additionally, the parasympathetic and sympathetic nervous systems are not individual
monoliths- each is formed of sub-pathways and circuits, which are regulated by different brain
areas. This allows the parasympathetic and sympathetic nervous systems to respond to different
body cues independently from one another, and not simply in response to the activation of one
another. Thus, while it might be expected that both electrodermal activity and heart rate would
rise or fall in parallel, this is not always the case, as both the sympathetic and parasympathetic
nervous systems can be activated or deactivated simultaneously. As a result, we recognize that
measures of electrodermal activity and cardiac output are not interchangeable, and provide

29
unique insights into an individual’s internal state (Eisenbarth et al 2016, Abdullahi Shehu 2023,
Wager et al 2009, Shaffer & Ginsberg 2017).
Skin conductance and heart rate, when measured simultaneously, can also be used to
identify involuntary attention orientations, or when someone changes the focus of their attention
(Sokolov et al 1960, 1963,1975, Zimmer & Richter 2013, Gruss et al 2019). These orientations
are marked by increases in skin conductance/the presence of SCRs, and heart rate decelerations.
Saliency is correlated with attention orientations, and threatening stimuli are more salient than
neutral stimuli. This is also known to be true for written words (Isenberg et al 1999, Porges et al
1969).
Based on this previously existing knowledge, we have two hypotheses regarding heart
rate activity and hateful vs non-hateful stimuli. We hypothesize that stimuli that contain hate
speech, and/or threat words, will elicit more attention orientation responses than stimuli that do
not, represented by simultaneous increases in skin conductance and decreases in heart rate. We
also hypothesize that hateful stimuli and stimuli that contain threat words will result in more
increases in heart rate than neutral stimuli, as a result of emotional arousal and threat responses.
Heart Rate Variability, and Determination of Basic Emotional States
Another well-known measure of cardiac changes due to autonomic nervous system
activity is Heart Rate Variability, or HRV. HRV is not interchangeable with heart rate, despite
the similarities in name. While heart rate is a measure of the number of heart beats in a given
time unit (generally per minute), HRV is the variation in the time between each individual beat.
So, the number of beats recorded in a minute (HR) could be the same in two different recordings,
but the time differences between beats in each of the recordings could still differ from each other
(Shaffer & Ginsberg 2017).

30
HRV has been shown to involve activity in the anterior cingulate gyrus, thalamus,
bilateral insula, ventromedial prefrontal cortex, amygdala, hippocampus, hypothalamus,
midbrain, and brainstem. It has been suggested that the right insula is involved in promoting
sympathetic regulation of cardiovascular activity, as well as in regulating cardio-vagal function.
The left insula, conversely, patterned more strongly with activation of the parasympathetic
pathways (Vargas et al 2016). Additionally, Gray et al 2004 demonstrated that parasympathetic
regulation of cardiac activity is heavily modulated by ganglion within the heart itself. They
demonstrated that the sinoatrial (SA) and pulmonary artery (PA) ganglion found in the heart
interact with each other directly in an interganglionic circuit, and then interact directly with the
sinoatrial node, to mediate vagal parasympathetic activity in the heart. This work further
supports the emerging hypothesis that cardiac ANS activity is regulated by a distinct cardiac
nervous system.
HRV rhythms have two main components, the Low Frequency (LF) and High Frequency
(HF). The sympathetic and parasympathetic systems, and the Vagal response all work in tandem
to modulate HRV via the balance of these two components (McDuff et al 2014). The LF portion
is modulated by both the sympathetic and parasympathetic systems-this balance is what allows
for the homeostatic regulation of blood pressure, while the HF portion is modulated solely by the
parasympathetic system as part of the respiratory sinus arrythmia, or the quickening of the heart
rate during autonomic breathing.
Changes in HRV can reflect several types of stress and arousal, including cognitive load ,
emotional changes/affective load, and threat responses (Taelman et al 2009, costalado et al 2019,
Schubert et al 2009,Richter et al 2021, Moses et al 2007, Hjortskov et al 2004, Moriguchi et al
1992). There are several methods by which changes in HRV can be measured. One possible

31
method is to examine the effects of the sympathetic and vagal systems on the sinus node of the
heart (Sztajzel 2004). The ratio of LF/HF output changes with increased sympathetic input, and
is thus indicative of an increased sympathetic and decreased vagal input ratio. This in turn is
representative of an increased arousal state (McDuff et al 2014).
Another method considers only the HF contributions, as it is primarily an index of
parasympathetic nervous system arousal, whose activation can then be more clearly interpreted
(Elliot et al 2011). Further breaking this measure down, the Root Mean Square of Successive
Differences (RMSSD) is often used as a way to quantify the HF contributions. RMSSD
measurements are reflective of changes in input from the vagal nerve, and can measure changes
in HRV in ultra-short-term intervals (between 10-60 seconds). This allows us to assess changes
in HRV to specific stimuli (Shaffer & Ginsberg 2017).
Additionally, some progress has been made towards beginning to dissociate the
simultaneous responses to each type of stimuli. Changes in heart rate and changes in skin
conductance have been shown to index different components of ANS activity. As summarized by
Britton et al 2006, valence, or the spectrum of unpleasant to pleasant responses, is associated
with changes in heart rate, while skin conductance is associated with arousal, or the spectrum of
calm to excited responses. Further, quite a lot of work has investigated and found different
physiological response patterns between different discrete emotional states (e.g. happy, sad, fear,
disgust, etc). These studies showed differential heart rate responses to emotionally provocative
stimuli, designed to induce single emotional state. Ekman et al 1983 showed that there were
unique heart rate changes with signature responses to comparisons between dichotomies of
related expressions, such as fear and anger, and that, when combined with the responses to
additional physiological measures, unique response patterns could be found for a wider range of

32
specific positively affective emotions. Collet et al 1997 found that when using skin conductance
responses in addition to other physiological markers, such as skin temperature, that even
negative emotions could be distinguished between each other. The problem remains, however,
that there are overlaps between reaction patterns, and a lack of a single, standard way to classify
emotions psychologically.
Such measures are often reliant on the participant only experiencing a single emotion at a
time, which is not extremely ecologically valid. Britton et al 2006 did find, however, that these
signals can be used together to distinguish between positive and negative valence and arousal
states, as well as positive and negative responses to social vs non-social stimuli. Social arousal
stimuli generated greater skin conductance responses than non-social, and heart rate
decelerations were greater to non-social stimuli than to social. Heart rate decelerations are also
known to be indicative of attention orienting. Porges et al 1969 found that both heart rate and
HRV decelerate with increases in attention. Conversely, heart rates have been shown to increase
as part of the process of coping with the feeling of threat (Fowles 1980). These results seem to
support each other in making these markers, used in conjunction, adequate measures to
differentiate between cognitive and affective responses, so long as the experiment is well
designed.
Hate Speech and Known Physiological Responses
We now build upon this knowledge and understanding of how the autonomic nervous
system responds to different emotional changes to assess what the current literature has
demonstrated regarding physiological responses to hate speech specifically.
While there is not extensive research into physiological responses to hate speech, and
some of what has been found is contradictory, it does provide evidence for hate speech eliciting

33
some type of physiological responses. Thus, we believe it is reasonable to expect hate speech to
elicit emotional responses, and thus corresponding physiological responses, if only at the
subconscious level.
Isenberg et al 1999 demonstrated that reading threat/taboo words, such as “gun,” can
elicit amygdala responses (Isenberg 1999). Emotionally valent sentences also elicit physiological
responses (Vrana et al 1989), and hate speech can elicit emotional responses (Cowan et al 1996).
Self-relevant stimuli are also associated with greater physiological responses than general
stimuli, indicating that a higher proportion of attention is allocated to them (Fan et al 2013).
Thus, we can speculate that hate speech, which is emotionally valent (Cowan et al 1996), and is
by definition based on some aspect of the target’s personal identity, would likely: a) elicit
physiological stress responses, and b) elicit a greater physiological and emotional response in
members of the targeted group than in non-target bystanders.
Further exploring these hypotheses, Soral et al 2023 used heart rate measurements to
investigate whether or not exposure to hate speech led to increased desensitization towards
subsequent exposure. They based their hypotheses largely on the General Aggression model,
which demonstrates that regular exposure to physical violence leads to lower levels of emotional
arousal when exposed to subsequent violence, which is correlated with the performance of more
aggressive behaviors (Anderson & Carnagey, 2004; Carnagey, Anderson, & Bartholow, 2007;
Carnagey, Anderson, & Bushman, 2007, Soral et al 2023).
Soral et al 2023 further incorporated these ideas with the knowledge that heart rate
deceleration occurs as part of an attention orienting response when an individual is exposed to
negatively emotionally valent stimuli, before the heart rate returns to a baseline state (Bradley
2009, Himichi & Ohtsubo 2020). They thus hypothesized that additional exposures to hate

34
speech would result in emotional desensitization, which could be measured via ratings of
offensiveness of hate speech, and the presence of heart rate decelerations when presented with
hate speech stimuli.
They sorted 56 participants between two groups: those who would be primed with hate
speech, and those who would be primed with neutral speech. The groups were asked to read an
initial text discussing politics and current issues that either contained hate speech, or only neutral
speech. Then, participants were asked to rate the offensiveness of 18 hate speech statements,
which targeted Black people, gay people, Ukrainian people, Jewish People, Muslim people, and
Roma people, on a scale of 1-7. (Note: this study was conducted in Poland, where these groups
are often recipients of hate and prejudice.) They then compared changes in heart rate, and ratings
of offensiveness between the two groups.
They found that participants who had been primed with hate speech did not exhibit
decreases in heart rate during the second portion of the experiment, while participants who had
been primed with neutral speech did. They concluded that, at least physiologically,
desensitization was occurring. In contrast, they did not observe significant differences in ratings
of offensiveness of the hate speech statements between the two groups. Their results, however,
did not have enough power to formally confirm the absence of an association between prior
exposure to hate speech and future ratings of hate speech.
These results somewhat contrast with results from a previous study from these authors, as
previously described. Soral et al 2017 found that there were, in fact, direct connections between
exposure to hate speech, desensitization, and perceived offensiveness of hate speech, but which
did not include a physiological measurement component.

35
Abuin-Vences et al 2022 used similar methods to investigate physiological responses to
hate speech within the context of desensitization to hate. They chose to focus on how exposure to
hateful content delivered by politicians can lead to desensitization to, and promote acceptance of,
removing democratic rights and freedoms from out-groups (Allport 1954, Bilewicz & Soral
2020, Soral et al 2017). They also investigated how political views, known to impact perceived
veracity of political statements (Clementson 2018), interacted with emotional responses, and
whether it mediated emotional responses.
They used skin conductance and heart rate variability to measure emotional responses
and desensitization to in-group vs. out-group political hate speech. They used HRV
measurements to represent the proportion of parasympathetic nervous system activity to
sympathetic nervous system activity to represent emotional state. SCRs were used in conjunction
to identify which responses were attention orientations as opposed to emotional responses.
They had two main hypotheses for the experiment: 1) People’s ratings of veracity and
upsettingness, and physiological responses, would depend on whether or not they shared political
views with the speaker, and 2) That ratings and physiological responses would also depend on
whether or not hate speech was present, or if only neutral discussions took place.
The experiment took place in Spain. First, participants were asked to complete an
inventory measuring how liberal or conservative their political views were. Participants were
then shown 4 video clips, each of which was 20 seconds long. Two videos featured speeches
from a prominent Spanish Socialist leader, and a prominent Spanish Far-Right Conservative
leader. Two videos featured each speaker talking about political issues in emotionally neutral
ways, and the other two videos featured each speaker using what the researchers deemed to be
hate speech against immigrants, and inter-party political opposition.

36
Participants watched each video in randomized order, while their physiological measures
were recorded. Participants were then asked to rate both how credible they thought each video
was, and how angry the video made them feel, on 7-point Likert scales.
For analyses, physiological measurements were synthesized onto a single measurement
of “biometric emotion,” by NeuroLynq/Shimmer software. Further details of how this
measurement was synthesized, or what it represented, were not provided.
One-way ANOVAs and Tukey's Post-Hoc analyses were used to analyze results. AbuinVences et al 2022 found that Liberal participants exhibited higher “biometric emotion” in all
categories than Conservative participants: Conservative participants had little to no emotional
responsiveness to both Far-Right videos, and Conservative responses to the Socialist videos were
greater than Conservative responses to Far-Right videos, but lower than Liberal participants’
responses to Socialist videos. Liberal participants showed higher “biometric emotion” level to
both Socialist videos, compared to Far-Right videos, and higher “biometric emotion” to the FarRight video containing hate speech than to the Far-Right neutral video.
With regards to self-reported measures, Conservative participants did not exhibit
significant differences in reported anger between any of the videos, and had relatively low
ratings (between 3.14 and 4.42, out of 7). Liberal participants showed lower anger ratings to both
Socialist videos than to both Far-Right videos, but did not have significantly different ratings
between neutral and hateful Far-Right videos, or neutral and hateful Socialist videos.
Conservative participants did not exhibit significant differences in assessments of credibility of
the videos- they assessed all 4 videos to be moderately credible.
The researchers concluded that the lower “biometric emotion” in all participants to
hateful content than to neutral content, within political categories, was due to emotional

37
desensitization as a result of exposure to hate speech, and that Conservative participants’ lowered
“biometric emotion” responses to all videos was due to previous regular exposure to hate speech
from Conservative politicians. However, details regarding what emotions were measured were
limited to “negative” without further detail, and measurements of how often participants were
exposed to hate speech before the experiment were not taken. As such, these results should be
reconfirmed with further testing.
The results do, however, demonstrate that physiological measures can provide insights
beyond self-rating measures. This is exemplified by the lack of significant difference in anger
ratings among Conservative participants across all videos, despite Conservative participants
having a higher level of “biometric emotion” when watching the hate speech video with a
Socialist speaker. They also reveal that there is, to some extent, an effect of being the “target” of
hate speech on responses, as participants listened to hate speech against those who shared their
political beliefs. However, the exact nature of this effect was not determinable from the data
collected, and should be explored further.
The researchers also concluded that the differential ratings of anger, truthfulness of
message, and truthfulness of speaker only had an interaction with political affiliation for Liberal
participants. Reasons explaining why these differences were only observed among Liberal
participants were not established, and need to be further explored.
Further research has probed deeper into whether or not ingroup-outgroup dynamics are
the best way to assess the impacts of hate speech, or if they are insufficient to describe the
complex processes occurring when someone is exposed to hate speech. Given the need for
mediating variables to connect exposure to hate speech with out-group relationships, it seems
plausible that this may be the case.

38
Pluta et al 2023 investigated physiological relationships between exposure to hate speech
and desensitization to pain, with more nuanced results. They investigated the cognitive
mechanism underlying this desensitization, via neural activity measured with fMRI. They
measured activity in the Anterior Cingulate Cortex (ACC), Anterior Insula (AI), and right
Temporal Parietal Junction (rTPJ) brain areas. The ACC and AI are associated with empathy.
Differential activation has been observed when participants are shown empathy-inducing stimuli
depicting in-group vs out-group members in pain, with decreased levels of activation occurring
when out-group members were observed (Peng et al 2020, Xu et al 2009). Activity in the rTPJ is
associated with “Theory of Mind,” cognitive processes, which allows for the ability to attribute
mental states to others, which is an important part of empathy (Mai et al 2016, Schaefer et al
2021). Together, the ACC, AI, and rTPJ are known as the “pain axis,” and is activated both when
the individual experiences pain themselves, and when they observe someone else experiencing
pain. Empathy and theory of mind allow someone to “walk in someone else’s shoes,” and
understand that when something that would cause themself pain happens to someone else, that
person will also experience pain (Dvash & Shamay-Tsoory 2014).
Pluta et al 2023 proposed 3 potential hypotheses for what may be driving desensitization
in response to verbal violence, and decreases in empathy towards others being targeted by such
violence.
First, they propose that multiple exposures to hate speech would result in decreases in
empathy, specifically towards out-groups. As described by Soral et al 2017 and Bilewicz & Soral
2020, recurrent exposures to hate against an out-group has been demonstrated to increase outgroup prejudice. Empathy interventions have been shown to reverse the effects of
desensitization, demonstrating that desensitization specifically results in diminished empathy

39
(Soral & Bilewicz 2022, Hangartner et al 2021). If this were what was occurring, during this
experiment, Pluta et al 2023 expected that there would have been reduced activity in the ACC
and AI regions when observing pain in out-group members, but not in-group members.
Second, they proposed that exposure to hate speech could decrease empathy for outgroup
members specifically by decreasing brain activity in regions responsible for theory of mind. If
this were occurring, they expected to see decreased activity in the right Temporal-Parietal
Junction (rTPJ) brain region, which is strongly associated with perspective taking, a crucial
aspect of theory of mind, but only for out-group members.
Finally, they proposed the idea that exposure to hate was simply resulting in “compassion
fatigue,” or “empathic numbing,” a well-known phenomenon in which a person becomes
emotionally “burnt-out” after exposure to stressful and/or upsetting content and interactions.
Compassion fatigue is often observed in professional caregivers, social workers, medical
professionals, etc., after recurring exposure to traumatic stories and experiences. It is not a result
of lack of caring, but a protective mechanism to help an individual cope with the recurrent stress
of exposure to upsetting scenarios (Robino 2019, Potter et al 2019). Compassion fatigue has been
documented outside of caregivers as well. Studies have shown that news and media coverage,
with chronic portrayals of negative world events, leads to compassion fatigue in the general
population, and avoidance of news coverage (Kinnick et al 1996, Cameron et al 2011). If this
were what was occurring during desensitization, the researchers expected there to be a decrease
in the ability to recognize the pain experienced by others, as assessed via decreases in the neural
“pain matrix”- the ACC, AI, and rTPJ brain regions, regardless of whether the person depicted in
pain was an in-group or out-group member.

40
The study took place in Poland, and participants consisted of 30 native-born and
culturally-identifying Polish people. The thirty participants were split into two groups of 15.
Each group read through priming material, and then read stories about Arabic migrants (who
have been the subjects of much hatred and intolerance in Poland) and other Polish people
experiencing pain. The stories did not contain any hateful content. The stories describing people
experiencing pain were the same for both Polish and Arabic subjects, with changes to the names
of the characters, and the photos of people meant to represent the characters, to reflect the
differences in race.
Two days before the experiment, participants were given SDO and RWA inventories to
complete. Then, on the day of the experiment, while fMRI activity was being recorded, the
participants were first shown either a neutral article about current events, or an article containing
hate speech against Arabic immigrants. Then, all participants were shown story vignettes in
which Polish or Arabic characters were experiencing pain.
The results most supported the third hypothesis, that compassion fatigue had occurred.
Participants that had been primed with hate speech saw a decrease in rTPJ activity in response to
all stimuli, not just those depicting outgroup members. Since this decrease was not seen
differentially between in-group and out-group depictions, this does not quite support the second
hypothesis, that this was as a result of decreased theory of mind. This decrease was not seen in
the neutrally primed participants. However, there was no change in the ACC or AI activity,
implying that there had not been a change in empathy. The researchers suggest that their results
may not be conclusive, since all participants had low SDO and RWA scores. Since SDO and
RWA are associated with in-group identification, the participants may have simply not perceived
a difference between stimuli subjects, instead considering all persons depicted to be in-group

41
members. If this was the case, then the results would support the second hypothesis, that
desensitization to hate speech is regulated via decreases in theory of mind processing.
These studies demonstrate that, while the nature of exactly how hate speech
physiologically impacts people is not fully understood, we do collectively know that hate speech
elicits a physiological response, and those physiological responses help contribute to decreases in
empathy and desensitization. Further, these studies demonstrate that there are more than simple
ingroup-outgroup dynamics mediating how people respond. Our studies help further our
collective understanding of these underlying processes by not just asking if people respond to
hate speech, but by specifically interrogating how participants from different demographics
respond to hate speech, how physiological responses to hate speech align with participants’ selfreported responses, and how psychometric values mediate all of these responses.
Physiological Measures and Descriptions of Emotional States
We now build upon this knowledge of physiological stress responses with more context
from research on a wider range of emotions. We will use this broader context of physiological
correlates of emotional arousal to attempt to describe which specific emotional states are being
elicited by hate speech and criticisms of systemic racism. While this is not a primary research
question for these experiments, we developed a paradigm to try to determine if hate speech
elicited specific emotional responses, and not just general stress.
There is some evidence that rudimentary emotional states can be deduced from
comparing physiological measures in tandem. Specifically, heart rate variability (HRV), heart
rate (HR), and skin conductance responses (SCRs). One method for defining emotional states
integrates whether they are approach/avoidance, and whether they have positive or negative
valence. Here, “approach,” refers to emotions that seek to engage with the stimulus, in an

42
incentivized way, often in order to achieve a sense of reward. Contrastingly, “avoidance” refers
to emotions that seek to get away from the stimulus, often due to fear or anxiety of a threat the
stimulus presents to an individual (Elliot et al 2013). Positive valence refers to emotions that
induce enjoyment or pleasantness, while negative valence refers to emotions that are unpleasant
and induce discomfort (Bradley et al, 2022). Emotional states with the same valence can have
opposite approach/avoidance motivations, and vice versa. For example, enjoyment is an
approach state, with positive valence, while anger is an approach state with negative valence
(Carver & Harmon-Jones 2009). This is represented physiologically, as well. Anger and
happiness have been shown to have similar cardiovascular patterns, despite their contrasting
valences (Wu et al 2019, Brossocht & Thayer 2003, Dechert et al 2005).
HRV has been found to be an indicator of approach vs avoidance during an emotional
response. Wu et al 2019 found that significantly higher HRV values were found for “approach,”
related emotions, namely amusement and anger, when participants viewed videos from the
standard Chinese affective video system (CAVS), which contains videos validated to elicit the
intended basic emotional responses. Rumpa et al 2021 also demonstrated that, when asked to
watch a video intended to induce anger, participants’ HRV values generally increased by a
significant amount. Conversely, HRV decreases when avoidance related emotions, like fear, are
experienced (Deits-Lebehn et al 2023, Ghiasi et al 2020).
By contrast, heart rate has fairly well-established correlations with emotional valence,
where increases in HR represent negatively valenced emotions, such as fear or anger, and
decreases in HR represent positively valenced emotions, such as happiness or contentedness
(Brosschot & Thayer 2003, Shi et al 2017, Vachiratamporn et al 2014). As HR can change
independently from HRV, we proposed that HR changes could be compared to HRV changes to

43
determine whether people’s emotional states in response to stimuli fell into one of 4 states:
approach/positive valence (happiness, enjoyment), approach/negative valence (anger),
avoidance/negative valence (fear), and avoidance/positive valence (boredom, emotional
disengagement).
Heart Rate and HRV are commonly used metrics in video game design, as a way to
assess players’ emotional experiences. They have been especially used to assess player responses
to horror games, where, as fear and anxiety increased, heart rate increased, and HRV strongly
decreased (Vachiratamporn et al 2014, Porter & Goolkasian 2019, Lobel et al 2016). Different
games designed to help children improve anger management skills found that anger and
frustration elicited increases in heart rate, while internal regulation to a calmer state lowered
heart rate (Ducharme et al 2021). These results, taken in conjunction with Wu et al 2019’s
results, and Rumpa et al 2021’s results, suggest that decreased HRV is related to “avoidance,”
related emotions, in addition to negatively valenced emotions. We thus hypothesize that
decreases in HRV will be associated with avoidance-related emotions.
Based on this knowledge, we will sort responses into quadrants, along axes of
positive/negative valence and approach/avoidance. Responses will be considered positively
valenced, approach oriented responses when heart rate decreases while heart rate variability
increases. We will refer to these responses as “happiness,” as a general label, but acknowledge
that this is likely an oversimplification, and that these responses may also include feelings such
as contentedness, or general interest in the stimulus (Elliot et al 2013, Bradley et al 2022).
Responses will be considered negatively valenced, approach oriented responses when
there is an increase in both heart rate and heart rate variability. We will refer to these responses
as “anger,” as anger is an emotion well known for being negatively valenced and approach

44
oriented (Dechert et al 2005, Wu et al 2019, Brossocht & Thayer 2003, Elliot et al 2013, Bradley
et al 2022).
Responses will be considered negatively valenced, avoidance oriented responses
when heart rate increases, and heart rate variability decreases. We will refer to these responses as
“fear,” though these responses may also encompass other similar emotions, such as threat or
anxiety (Vachiratamporn et al 2014, Porter & Goolkasian 2019, Lobel et al 2016, Elliot et al
2013, Bradley et al 2022).
This leaves us with the question of identifying what emotional state would occur during
positively valenced avoidance oriented responses. Using our quadrant system, these responses
would have a decrease in both heart rate and heart rate variability.
Positively valenced avoidant emotions can be difficult to conceptualize. However, some
researchers observed decreases in both heart rate and heart rate variability when participants
were playing a video game stage intentionally designed to induce boredom by being repetitive
and easy (Giakoumis et al 2010). We believe that boredom, or emotional disengagement, could
capture the experience of a positive-avoidant emotion. Boredom could reflect not wanting to
approach/engage with a stimulus, which would fall under avoidance (Elliot et al 2013), but not
necessarily being made to experience strong negatively valent emotions, such as fear or anger
(Bradley et al 2022).
Finally, changes in heart rate, in conjunction with the elicitation of a skin conductance
response, can be associated with attention orientation. Many studies have demonstrated that the
presence of an SCR, when observed while HR is decreasing, represents changes in attentional
focus (Graham & Clifton 1966, Zimmer & Richter 2022). As such, we identified any stimuli

45
responses where this was observed, and removed them from the analyses of specific emotional
identification to avoid false positive emotional assignments.
Summary of Hypotheses
Based on all of the previous literature surrounding these topics, our main hypotheses
were:
1. Hate speech would be rated as more offensive/upsetting than non-hate speech.
2. Hate speech would elicit more skin conductance responses than non-hate speech.
3. Hate speech would elicit skin conductance responses with greater area under the curve
than non-hate speech.
4. Hate speech would elicit greater changes in heart rate than non-hate speech.
5. Criticisms of Systemic racism would be responded to, both in terms of physiological
arousal and offensiveness/upsettingness ratings, as non-hate speech by non-white
participants, and be responded to as hate speech by white participants.
6. Self-Relevance would mediate these differential response patterns.
7. Being a member of a non-target marginalized group would mediate these response
patterns.
8. Scores on psychometric inventories would mediate these differential response patterns.
9. Hate speech would elicit more anger and threat/fear emotional patterns than non-hate
speech.
10. Hate speech would elicit more attention orientation responses than non-hate stimuli.

46
Chapter 1- Experiment 1
Methods
Participants
47 participants took part in Experiment 1. There were 25 women, 18 men, 2 non-binary
participants, and 2 participants who declined to state their genders. There were 3 gay/queer
participants, 1 demisexual-straight participant, 7 bisexual participants, 1 participant who declined
to provide their sexual orientation, and 35 straight participants. There were 2 Black participants,
15 Asian participants, 14 White participants, 4 Middle Eastern participants, 5 Hispanic
participants, 6 mixed race participants, and one participant who declined to provide their racial
identity. The mean age of participants was 20.15 years old. An additional 7 participants were
removed from the dataset because they did not exhibit changes in skin conductance response,
because they voluntarily removed themselves from the experiment before completion, because
there was an issue with the data recording equipment, or because they clearly answered the rating
questions with nonsense answers, eg: giving responses outside of the range of options, only
answering with one rating for all items, etc.
Stimuli
Each participant was shown and responded to eighty stimuli. There were four categories
of stimuli: Neutral Speech, Threat Speech, Misogynist Hate Speech, and General Hate Speech.
Participants were shown twenty items from each category. As each stimulus was presented to
participants, they were asked to assess how they felt about each item. For Experiment 1,
participants were asked, “How offensive is this statement?” and were given a Likert scale of 1 to
7, where 1 was defined as “Not at all offensive” and 7 was defined as “Extremely Offensive.”

47
Items (which can be found in the supplemental index) in the Neutral Speech (AKA
“Baseline”) Category did not discuss sensitive topics, or contain words that may elicit an
amygdala response. This category acted as one of the control conditions. Threat Speech items did
not contain content that would be considered a threat, or discuss sensitive topics, but did contain
words that are known to elicit a subconscious amygdala activity, and thus produced a
subconscious threat response (Isenberg 1999). This category also acted as a control condition, to
examine how responses to hate based rhetoric compared and contrasted to responses to words
known to elicit physiological threat responses, when read outside of a hateful context.
Items in the Misogynist Hate category contained items validated for containing hatebased rhetoric, that specifically targeted women. This category acted as an experimental
condition in two ways: responses to these items were used to assess how responses to hate based
rhetoric as compared to non-hateful content, and to assess how self-relevancy affected responses
to hateful content. Responses from women to these items were compared to responses from
people of other genders.
Items in the General Hate category contained items validated for containing hate-based
rhetoric targeting groups other than women, including Race, Sexual Orientation, and Religious
Beliefs. This category acted as an experimental condition in two ways: responses to these items
were used to assess how responses to hate based rhetoric as compared to non-hateful content, and
to assess how self-relevancy affected responses to hateful content. Responses from women to
items from the Misogynist Hate category were compared to responses to these items, to
determine if they were rated differently than self-relevant stimuli. Conversely, responses from
people of other genders to these items were compared to responses to items in the Misogynist

48
Hate category, to determine if there were differential responses to items in these categories,
despite the items in both categories not being self-relevant.
While it was possible that participants may have found some of the general hate stimuli
to be self-relevant, these stimuli were not considered as such during analyses. There is no way to
create a set of truly non-self-relevant stimuli to be used with participants from the general public,
given the inherently intersectional nature of human identities and social power dynamics. As
such, we made sure that several target groups were included, so that it would be very unlikely
that a given individual would find more than a few items self-relevant. We chose this method to
try to preserve ecological validity- if we were to intentionally seek participants who met very
controlled demographic criteria, in order to ensure that participants would not find any degree of
self-relevance in the stimuli, we would not be able to generalize any results to the general
population, and would potentially miss the insights that could arise from sampling a diverse
population.
Psychometric and Demographic Inventories
To assess how variables affecting participants’ world view and mental health, we used 5
psychometric inventories: The Ambivalent Sexism Inventory, the Social Dominance Inventory,
The Grandiose Narcissism Scale, The Belief in a Just World inventory, and the CPLC CrossCutting Symptoms Measure-Adult. The CPLC Cross-Cutting Symptoms Measure-Adult was
modified from its original form. In full, it includes questions measuring “Personality
Functioning,” that measure how sure someone is in their own identity and life choices. Given
that the participants for this study were all college students, who were all exploring their adult
identities, and what they want to do with their lives, scores were universally very low in this
category, which was affecting the aggregate score. As such, responses to these questions were

49
omitted from analyses. Participants also provided information about their race, sexual
orientation, and gender.
Equipment
All physiological responses were recorded using a Biopac MP 160 system. Stimuli were
presented to participants via custom MATLAB script on a desktop computer. Participants
entered rating responses to stimuli on this computer as well, and the stimulus delivery times,
stimulus IDs, and participants’ corresponding rating responses were output to a MATLAB file.
The MP160 system was connected to the desktop computer using a Little Black Box TTL device,
so electrical stimulus delivery signals could be sent from the desktop to the MP160, and
converted to a digital signal. The MP160 was connected to a separate laptop via ethernet cable,
and stimulus delivery timestamps and physiological output data were collected, integrated,
recorded, and saved using AcqKnowledge 5 software.
To record skin conductance responses, Biopac EL-509 single-use recording electrodes
were placed on the palm of the participants’ non-dominant hand. The non-dominant hand was
used to avoid movement artifacts in the data from participants typing their rating responses into
the desktop. Biopac GEL 101 Isotonic gel was used to amplify skin conductance for better
recording accuracy.
To record heart rate data, three Biopac EL-500 single-use recording electrodes were
placed on the participants’ chests, under their shirts. For Experiment 1, one electrode was placed
directly above the right collarbone, one directly above the left collarbone, and one directly above
the area between the two collarbones.

50
To record respiration data, a Biopac TSD101B Respiratory Effort Transducer was placed
snugly around each participant’s chest, below the approximate bust line, over their shirts.
Respiration data was collected to calculate heart rate variability (HRV) data.
Responses to psychometric and demographic inventories were once again collected via
Qualtrics, with participants completing these inventories using the laptop from the research
equipment set up.
Procedure
Data Collection
Participants were briefed about what the experiment would entail: that they would be
reading posts that were taken from social media, that some of them would contain hate speech,
and some would not, that they would be asked to rate the posts they were shown, and that they
would complete psychometric measures after they finished rating all of the posts. If participants
still wanted to participate in the experiment, they were given consent forms to sign, and any
questions they had were answered. Participants then had the physiological recording equipment
attached to them. Participants were asked to put their belongings and cell phones out of reach, so
they would not be able to use them during the experiment.
Participants were given practice examples of items to rate, with a researcher present, so
they could get used to the paradigm and ask any questions about the experiment. The researcher
then left the room, and the participants read and responded to each stimulus item without being
observed. Each stimulus was presented 5 seconds after the participant submitted their response to
the previous stimulus. The previous stimulus was not visible to the participant during this period.
This allowed for physiological responses to be better paired to which stimulus elicited them
during analysis, as well as providing a “wash-out” period for participants, to help prevent any

51
carry-over effects from one item to the next with regards to physiological responses, especially
changes in heart rate (Wascher 2021, Anttonen et al 2005). Physiological data was collected at a
sampling rate of 2000Hz.
Outside of this minimum inter-stimulus interval, participants responded to the stimuli at
their own pace. When the participants finished responding to all of the items, they let the
researcher know, and had the physiology-recording equipment removed. They then completed
the psychometric and demographic inventories.
Data Analyses
All physiological signal processing was done in AcqKnowledge 5 software. Each of the
signal channels were time-aligned, and compared to an additional channel where timestamps of
stimulus deliveries were recorded, and identified using the “Find Cycle,” function, set to detect
peaks in a data channel. The “Find Cycle” function was also used to identify individual segments
of the physiological signals between stimulus delivery markers on a separate channel, convert
them from tonic to phasic signals where applicable, and analyze them separately from the other
segments. Phasic data is more appropriate for assessing short-term physiological representations
of emotional responses to a continuous stream of stimuli, especially when the stimuli may not all
be designed to elicit the same type of emotional or physiological response (Ravaja et al 2006).
Skin Conductance
Skin conductance data was initially recorded as a tonic measurement of skin conductance
level. This data was then converted to a phasic measurement using the “Locate SCR,” function
using a 0.05Hz filter. Then, individual skin conductance responses were identified in the Phasic
data using the AcqKnowledge “Locate SCR,” function. Parameters recommended by Braithwaite
et al 2013 were used to determine the thresholds for what identified peaks in SCL should be

52
identified as a SCR, vs. rejected as background noise. Peaks with amplitudes less than 10% of
the amplitude of the highest peak in the recording, and with changes in amplitude of less than
0.01 microsiemens, were excluded. Braithwaite et al 2013 validated that these minimal exclusion
criteria could be used accurately when analyzing data specifically recorded with Biopac MP150
systems.
The Find Cycle function was used to identify the beginning, peaks, and ends of each
SCR, as well as calculating the change in amplitude between the onset and peak of each SCR,
and the peak and end of the SCR. Amplitude was measured in microsiemens. The Find Cycle
function was used in the same way to measure the area under the curve of each SCR. Area under
the curve was measured in square microsiemens. The Find Peaks and Find Cycle functions were
used to determine which stimuli elicited the specific SCRs. SCR waveform onsets had to begin at
least 1 second after stimulus delivery to be considered elicited by that stimulus, in accordance
with current known standards for minimum times between stimulus presentation and skin
conductance response onset (Sjowerman & Lonsdorf 2019, Boucsein et al 2012). If the
waveform onsets occurred earlier than this, or before the new stimulus was presented, it was
discarded as an artifact, likely due to physical motion from the participant.
AcqKnowledge 5 software was then used to convert all of these skin conductance outputs
into an .xlsx file. Using the most current versions of Microsoft Excel available when analyzing
each participant’s data set, skin conductance outputs were aligned with their corresponding
stimulus IDs, and added to a data table in a format that could be used in R for statistical analyses.
Heart Rate
Changes in heart rate were measured in changes in inter-beat interval durations, as is
standard for ECG recordings. Heart beats and their components were identified and labeled by

53
AcqKnowledge 5’s “Identify QRS Complexes” function. The “Find Rate” function, with the
settings for “HR from ECG-Human Resting,” was used to create a data channel representing
heart rate from the ECG data collected. The rejection rate was 5% of peak, heart rate segments
under 40 BPM, or over 120 BPM, for windows of 25 ms, to filter noise from the signal.
Then, the “Find Cycle,” function, with the “Locate Cycle from Events,” option selected, was
used to calculate the change in heart rate over each inter-stimulus interval. The changes in interbeat intervals were measured in milliseconds. The rate change data were then converted to .xlsx
files, matched with their corresponding stimulus IDs, and added to a dataset formatted for
compatibility with R for further analyses.
Heart Rate Variability
Heart rate variability was calculated from the existing heart rate and respiration data,
using the AcqKnowledge 5 function, “HRV and RSA,” with the “Multi-Epoch HRV and RSASpectral” option selected, and set to “Time between Event boundaries.” The outputs included the
RMSSD values used for analyses. These values were then converted to .xlsx files, matched with
their corresponding stimulus IDs, and added to the dataset formatted for analyses in R.
Emotion Identification
Finally, the physiological responses were classified by the combination of skin
conductance responses, RMSSD changes, and heart rate changes in response to each individual
stimulus. These combinations were identified in Excel using nested IF functions. Only one
response pattern was labeled for each stimulus interval. The combinations were classified as one
of four emotional responses, or an attention orientation response. The response patterns for each
response type are listed in the figure below.

54
Table 1: Emotion Identification Patterns
Response Type Skin Conductance Heart Rate Heart Rate Variability
Attention Orientation Present Increase N/A
Positive Approach Absent Decrease Increase
Negative Approach Absent Increase Increase
Positive Avoidance Absent Decrease Decrease
Negative Avoidance Absent Increase Decrease
Statistical Analyses
Several different analyses were used to assess how each response type (ratings, HR
changes, skin conductance responses) changed in response to the category of stimulus shown, the
participant’s demographic characteristics, and their personal characteristics, as assessed by
psychometric inventories.
Linear mixed effects models were used to examine the predictive abilities of different
variables, such as Category (of stimulus), psychometrics, and demographics, on the
physiological and Rating responses. These models were created in R using the lmer function.
The predictive effects of the psychometric variables were assessed in two different ways.
First, a principal components analysis was computed to attempt to determine which
psychometric variables were most impactful on the model, and to examine how these variables
related to each other. Each sequential principal component was then added as a predictor to a
linear mixed effects model, in addition to Category. ANOVAs were performed to determine if
the models with and without each principal components added were significantly different, and
which model was the best according to BIC values. A final set of linear mixed effects models
and ANOVAs were performed, with each predictor that created better models added one at a

55
time, to determine how many of these predictors should be added to create the best model
possible, according to BIC values.
The predictive effects of the psychometric variables were also assessed directly, by
adding each variable as a predictor to linear mixed effects models, to determine if it provided any
predictive ability to the model, for each type of response measured. The magnitudes of the
effects were calculated for psychometric variables that created significantly better models,
according to BIC values. The variables were then added one at a time to a final linear mixed
effects model, from largest effect to least, to determine which combination of variables created
the best predictive model via ANOVA tests and BIC values. For each individual model of the
effect of one of the psychometric variables individually being added as a predictor, the model
was examined to determine if the psychometric variable had relationships with the model as a
whole, including whether it had a main effect on the model, an interaction with the effect of
category’s effect on the model, and whether it had a main effect on the contrasts between the
responses to the baseline stimuli vs the categories of variable stimuli.
The predictive effects of demographic variables were assessed using linear mixed effects
models again. A model with both stimulus category demographics as predictors was created, and
examined to determine whether the demographic factor had any relationships with the model as a
whole, including having an interaction with category of stimuli, a main effect on the larger
model, and main effects on the contrasts between the baseline and variable stimuli. An ANOVA
was used to determine if the model including the demographic factor was a significantly better
model than the baseline model, via BIC values. This was repeated independently for the effects
of gender, sexual orientation, and race.

56
Mann-Whitney U-Tests were performed to determine if rating and physiological
responses to each stimulus category were significantly different between demographic groups, to
determine if the principal component values that significantly improved the linear mixed effects
models when added as a predictor were different between demographic groups, and to determine
if there were differences in how both women and non-women responded to misogynist hate vs
general hate, and hate speech vs criticisms of systemic racism.
Finally, the emotional classification data was analyzed using Chi-Squared tests. These
tests were used to determine if the distributions of emotional response types varied significantly
between the stimulus categories. Supplemental analyses included further Mann-Whitney U tests
to determine if individual response types varied between demographic groups, or between
stimulus categories.
Results
Rating Data
Effect of Category
First, we assessed whether there was a difference in rating between categories that
contain hate speech, and categories that do not, as the primary hypothesis to be confirmed before
proceeding with any further analyses. To do this, we established a “Baseline Model,” by
performing a series of tests where only Category was entered as a predictor for the Ratings of
each of the four stimuli categories. A linear mixed effects (LME) model was created, and found
that there was a significant main effect of Category, F(3,138)=1034.2. The main effects of the
contrasts between each individual stimulus category and the baseline condition (“Neutral
Speech”) were significant: General Hate and baseline, b = 4.9652, t(138)=44.3, Misogynist Hate

57
and baseline, b=4.3859, t( 138)=39.167, Threat and baseline, b=0.6109, t(138)=5.45, shown
below in Figure 1.
Figure 1
Demographic Factors
With the confirmation that stimuli containing hate speech are rated as more offensive
than either the “threat” speech or neutral speech categories, we proceeded to investigate whether
demographic factors affected Rating behavior. First, we examined the role of “self-relevancy,”
by examining gender differences in response to the 4 stimulus categories, with a special focus on
differences in Rating of the stimuli containing misogynist hate.
An LME was created, using both Gender and Category as predictors of Rating. An
ANOVA comparing the baseline model (BIC= 165.93) with only Category as a predictor, to the
model with Category + Gender (BIC= 181.56) as predictors was performed, and found that the
addition of gender did not create a better model than the baseline model.
Mann-Whitney U tests were performed to examine whether the ratings for each stimulus
category were significantly different between gender categories. There was a significant
difference only in the Misogynist Hate category, where women did rate the items as more
Experiment 1: Mean offensiveness ratings for each
category of stimulus, with data from all participants.
Distributions of mean offensiveness ratings by
category of stimulus.

58
offensive than people of other genders (W = 367.5, p-value = 0.04), women’s mean rating =
5.93, non-women’s mean rating = 5.24.
A Mann-Whitney U test was then performed to determine whether there was a
significant difference between women’s offensiveness ratings of General Hate stimuli vs.
Misogynist Hate stimuli. There was a significant difference, with women rating General Hate
stimuli as more offensive than Misogynist Hate stimuli (W= 487, p-value = 0.007), shown below
in Figure 2.
Figure 2
A Mann-Whitney U test was then performed to determine whether there was a significant
difference between other gender participants’ offensiveness ratings of General Hate stimuli vs
Misogynist Hate stimuli. There was a significant difference, with other gender participants rating
General Hate stimuli as more offensive than Misogynist Hate stimuli (W=312.5, p-value = 0.02).
Figure 3
Women’s mean ratings of the
offensiveness of General Hate
stimuli vs Misogynist Hate
stimuli.
Non-women’s mean ratings of General Hate
stimuli vs Misogynist Hate stimuli.

59
We further investigated whether belonging to a marginalized group, whether or not it was
the group targeted by a given instance of hate speech, was associated with Rating behavior by
creating LME models with Race and Sexual Orientation added as predictors. A model of the
Effects of Category + Race (BIC=179.65) was found to not be significantly different from the
baseline model of only the effect of Category (BIC=165.93). A model of Category + Sexual
Orientation (BIC=175.19) was not found to be a better model than the baseline model
(BIC=165.93).
Mann-Whitney U tests were performed to examine whether the ratings for each stimulus
category were significantly different between race categories. There were significant differences
in the Baseline/Neutral speech category (w=343.5, p=0.02), White mean rating =1.28, NonWhite mean rating =1.09, the Threat speech category (w=385.5, p=0.0009), White mean rating =
2.05, Non-White mean rating = 1.39, and the Misogynist Hate category (w=348.5, p=0.014),
White mean rating =5.95, Non-White mean rating = 4.95. There was not a significant difference
in the General Hate speech category, though significance was close to being achieved (w=324,
p=0.056), White mean rating = 6.40, Non-White mean rating = 5.78.
Mann-Whitney U tests were also performed to examine whether the ratings for each
stimulus category were significantly different between orientation categories. However, there
were no significant differences in ratings based on Orientation in any of the categories.
While there were several areas where demographics did have a strong relationship with
Rating and Category, the distributions of Rating scores between demographic pairs were not
entirely distinct- there were significant areas of overlap in the distributions of scores, suggesting
that Gender and Race alone are not responsible for the differences in Rating. To further

60
investigate what could be driving these behavior patterns, we next investigated the role of
psychometric variables on Rating.
Principal Components Analyses
Next, we began to investigate the role that traits measured by the psychometric
inventories on Rating behavior. To attempt to reduce the number of variables involved, we
performed a Principal Components Analysis (PCA), using participant scores on the psychometric
inventories to create the components. The Principal Component (PC) loadings are shown in
Table 2. The scree plot in Figure 4 shows how much of the total variance in rating is accounted
for by each principal component.
Table 2: Experiment 1 Principal Components Factor Loadings
Factor PC1 PC2 PC3 PC4 PC5
Ambivalent
Sexism
0.4957432 0.04861631 0.1762742 -0.84696230 0.05880005
Social Dominance
Orientation
0.5493423 -0.04656219 0.2061526 0.30993701 -0.74665600
Belief in a Just
World
0.2647755 -0.81177300 0.2712653 0.19262850 0.40028481
Narcissism -0.4574033 0.05240150 0.8720773 -0.09276512 -0.13752150
Mental Health 0.4160985 0.57770848 0.3038501 0.37534358 0.50981103

Figure 4

Variance explained by each principal component

61
We added PC1 to the baseline model and found that the model with PC1 included had a
lower BIC (391.07) than the baseline model (397.52), and the difference between the models was
significant, χ
2
(4)=27.31. In this model, there was a significant interaction between Category and
PC1, F(3,138) = 9.12. PC1 also had significant main effects in the contrasts between the
Misogynist Hate, b =0.287, t(138) = 2.83 and the General Hate, b =0.193 , t(138) = 4.19
categories, and the Baseline category. A model including PC2 did not improve prediction (BIC =
415.61) over the model including Category and PC1. We found no significant effect of PC2 or
interaction with Category in the larger model. A model including PC3 (BIC=392.15) also did not
improve prediction (BIC) over a model including Category + PC1 (BIC=391.07) as predictors,
but inspection of that model showed an interaction between PC3 and Category, F(3,138) = 7.82,
which merits further discussion. PC3 also had a significant main effect in the contrast between
the Misogynist Hate, b=0.51, t(138)=3.97, and the General Hate, b=0.33, t(138)=2.58,
categories, and the Baseline category.
We created a model with only Category + PC3 as predictors, to further examine the role
of PC3 as a predictor, without PC1 involved. We found that the Baseline model (BIC=397.52),
with only Category as a predictor, and the model with Category+PC3 (BIC=395.94) as predictors
had BIC values too close together to determine which model did a better job of explaining
variance in the model. A Chi-Squared analysis determined that the difference between the
models was significant, χ
2
(4)=22.44, though. We thus had to decide whether or not adding PC3
to our model would be beneficial or not. Given that PCs inherently explain less variance with
each successive iteration, that PC1 and PC3 only had significant main effects in the contrasts
between the hate categories and the baseline category, and that adding PC3 to a model with
Category and PC1 only produced a similar, but not lower BIC, we concluded that it would be

62
best not to include PC3 in the model going forward. It is also worth noting that having such
similar BIC values implies that the amount of variance accounted for by the models of Category
+ PC1 and Category + PC1 + PC3 is not easily distinguished. This suggests that it would be
beneficial to examine the effects of the psychometric variables that compose the PCs on a model
of Category and Rating directly. Especially of note is the fact that PC1 and PC3 have
overlapping factor loadings. PC1 has all of its constituent factors fairly evenly represented in its
composition. PC3 similarly has fairly even factor loading, with the exception of Narcissism,
which has a clear stronger loading than the other factors. However, since PC1 also has
Narcissism fairly well represented, it does not appear that PC3 is being driven by any factors that
are not also well represented in PC1.
A model including PC4 (BIC=414.63) did not improve prediction over the model using
Category + PC1 (BIC=397.52). We found no significant effect of PC4 or interaction with
Category in the larger model. A model including PC5 (BIC=408.79) did not improve prediction
over the model using Category + PC1 (BIC=397.52). We found no significant effect of PC5 or
interaction with Category in the larger model. The best model was thus determined to be the
model with Category+PC1 as predictors of offensiveness rating for the different categories of
stimuli.
Next, overly influential participants were identified for the model with Category + PC1
as predictors. Once removed, the models were recreated with the smaller participant pool, in the
same manner as before. A new baseline model of the effects of Category on ratings of the four
stimulus categories was created, where category was the only predicting variable, and found that
there was once again a main effect of Category in the larger model, F(3,102)=2052.7, and that
Category had additional main effects on the contrasts between the General Hate b=5.21,

63
t(102)=61.724, Misogynist Hate, b=4.68, t(102)=55.35, and Threat, b=0.58,t(102)=6.895,
categories, compared to the Baseline stimulus category (“Neutral Speech”).
PC1 was added as a predictor to this new baseline model, and significantly improved the
model. The model with PC1 + Category as predictors had a BIC of 168.23, while the baseline
model of only the effect of Category had a BIC of 171.5. The difference between these two
models was also found to be significant, χ
2
(4) = 22.93. There was also a significant interaction
between Category and PC1, F(3,102)=8.29. There was a main effect of PC1 on the contrasts
between the baseline stimulus category, and the General Hate, b=0.14, t(102)=2.39, and the
Misogynist Hate, b=0.21, t(102)=3.49, categories.
We observed that there was a steep positive interaction between PC1 scores, and Rating
in the General Hate and Misogynist Hate categories. By contrast, there was a weak negative
correlation between Rating and the Baseline/Neutral Speech, and Threat categories. It is worth
noting that the PCs are “sign reversed,” meaning that the direction of the graphed interactions is
the opposite of the true direction of the relationship between the PC and Rating behaviors. These
relationships are shown in Figure 5.
Figure 5
Inverse relationships between PC1 values and
rating values.

64
Because PC3 had interactions with Category in the previous models, we investigated how
PC3 acted as a predictor compared to PC1 after the overly influential participants had been
removed. We found that the model with PC3 + Category as predictors (BIC=175.01) was
significantly different from the new baseline model, with only Category as a predictor
(BIC=171.5), χ
2
(4)=10.45. PC3 had a significant interaction with Category, F(3,99)=3.62, in the
larger model, and a significant main effect on the contrast between the ratings of the Baseline
stimuli and the Misogynist Hate stimuli, b=0.28, t(99)=2.84. However, since the Baseline Model
had a significantly lower BIC value than the model with PC3 added as a predictor, PC3 was not
ultimately kept in the final model.
We also investigated the relationship between PC3, Rating, and Category. Like in the
analyses of PC1, we observed a strong positive correlation between PC3 scores and Rating in the
Misogynist Hate category, and a relatively weak negative correlation between PC3 and the
Baseline/Neutral Speech and Threat categories. Unlike PC1, however, PC3 had only a very weak
positive correlation with Rating in the General Hate category. As stated before, the PCs are “sign
reversed,” meaning that the direction of the graphed interactions is the opposite of the true
direction of the relationship between the PC and Rating behaviors. These relationships are
demonstrated in Figure 6 below.

65
Figure 6

PC1 Values between Demographic Variables
Next, we investigated whether there are significant differences in PC1 values between
demographic dyad categories. A Mann-Whitney U test also demonstrated that there was a
significant difference in PC1 values between gender groups (W = 2512, p-value = 0.005448),
with Women having higher PC1 values; importantly, the PC1 values are inversely correlated
with scores on the psychometric inventories, implying that women had lower scores on the
inventories than people of other genders.
A Mann-Whitney U test was performed to determine if there was a difference in PC1
values between race categories. There was a significant difference (W = 1392, p-value =
0.02664), with non-white participants having higher values, which indicates that they had lower
scores on the psychometric inventories used to compute the PCs.
Inverse relationships between PC3 values and
rating values.

66
A Mann-Whitney U test was performed to determine if there was a difference in PC1
values between orientation groups, and it was determined that the difference was significant (W
= 1072, p-value = 0.0001436). Queer participants had much higher PC1 scores, which indicates
that they had lower scores on the inventories contributing to PC makeup.
Thus, we concluded that the best model to explain Rating behavior in experiment 1 was
the model with PC1+Category as predictors. We re-examined the composition of PC1, to seek
insight into which psychometric variables were most strongly contributing to Rating behavior.
We found, however, the psychometric variables largely contributed to the composition in PC1 in
relatively equal amounts. The only notable exception being that BIAJW contributed less than the
other variables. Also of note, Narcissism contributed in a magnitude fairly equal to the other
variables, but in the opposite direction. Given that these results did not provide as much insight
as we had hoped for, we created LME models with each of the psychometric variables added as
predictors, one by one, to further investigate which psychometric variables played a predictive
role in regards to Rating behavior between stimulus categories.
Effects of Individual Psychometric Variables
We next considered that because the factor loadings of PC1 were fairly evenly distributed
between the component factors, and, given that there was overlap in the loadings in PC1 and
PC3, which also had significant impacts on the larger models, that the use of a PCA may not
have given us the level of insight into how each variable was driving Rating behavior. We
decided that it would be valuable to investigate each psychometric value as a predictor
individually.
A new Baseline model with Category as a predictor of Rating was created, as not
computing principal components removed the need for participants to have complete data for all

67
psychometric inventories to be included in analyses. In the previous models, several participants
were removed from the dataset because they had only partially completed some inventories. In
this model, only overly-influential participants were removed, after being identified. This new
Baseline model once again demonstrated a significant effect of Category on Rating via LME,
F(3,126)= 1669.4; the LME also showed that Category had a significant main effect on the
contrasts between the Baseline stimulus category (“Neutral Speech”), and all other stimulus
categories: General Hate, b=5.12, t(126)=55.95, Misogynist Hate, b=4.59, t(126)=50.19, and
Threat, b=0.63, t(126)=6.86.
Adding scores on the Ambivalent Sexism inventory as a predictor to the Baseline rating
model (BIC=276.13) created a significantly different model (BIC=286.55), but BIC values
indicate that the Baseline model is a better model. An ANOVA demonstrated that the difference
between these models was significant χ
2
(4)=10.08. An LME also demonstrated that Ambivalent
Sexism had a significant interaction with Category, F(3,126)=3.40. The mean ambivalent sexism
score was 34.01, with a standard deviation of 13.83
Figure 7

Adding scores on the Social Dominance Orientation inventory as a predictor to the
baseline rating model (BIC=276.13) created a model that was significantly different
(BIC=281.17), but BIC values showed that the Baseline model was still the better model. The
Distribution of participants’ Ambivalent
Sexism scores.
Ambivalent Sexism

68
difference between these models was significant, χ
2
(4)=15.46. An LME showed Social
Dominance Orientation had a significant interaction with Category, F(3,26)=5.28. The LME also
showed that SDO had a significant main effect on the contrasts between the Baseline stimulus
category, and the General Hate, b= -0.02, t(126)= -2.43, and the Misogynist Hate, b= -0.02,
t(126)= -2.94, stimulus categories. The mean SDO score was 30.46, with a standard deviation of
12.25.
Figure 8
The model with Category + Belief in a Just World (BIAJW)(BIC=294.10) did not create
a significantly different model compared to the Baseline model, with only Category as a
predictor (BIC=276.13), as determined via ANOVA, χ
2
(4)=2.53 We found no significant
interaction with Category or main effect of Belief in a Just World in the larger LME model. The
mean score on the Belief in a Just World inventory was 19.41, with a standard deviation of 6.47.
Distribution of participants’ Social Dominance
Orientation scores.
Social Dominance Orientation

69
Figure 9
Adding scores from the Grandiose Narcissism inventory to create a model with Category
+ Narcissism as predictors (BIC=295.43) did not create a significantly better model compared to
the baseline model, with only Category added as a predictor (BIC=276.13), as determined via
ANOVA. We found no significant interaction between Category and Narcissism or main effect
of Narcissism in the larger LME model. The mean score on the Grandiose Narcissism Inventory
was 51.43, with a standard deviation of 4.67.
Figure 10

We next created a model with the scores from the CPLC Cross-Cutting Symptoms
Measure-Adult, whose output is from here forward referred to as “Mental Health.” the model
with Category + Mental Health as predictors, (BIC=283.72), did create a model, χ
2
(4)=12.91,
Distribution of participants’ Belief in a Just
World scores.
Distribution of participants’ Grandiose
Narcissism Scores.
Belief in a Just World
Global Grandiose Narcissism

70
that was significantly different from the Baseline model, (BIC=276.13), but not necessarily a
significantly better one, as determined via BIC values. Further analyses via LME reveal that
Mental Health and Category do have a significant interaction in the larger model, F(3,126)=3.55.
An LME also revealed that Mental Health had significant main effects on the contrasts between
the Baseline stimulus category and the Misogynist Hate, b= -0.03, t(126)= -3.06, category. The
mean score on the CPLC Cross-Cutting Symptoms Measure-Adult (inventory of broad mental
health quality, where lower scores indicate better mental health) was 13.72, with a standard
deviation of 9.00.
Figure 11
Thus, we have determined that the best model of Rating was the Baseline model, with
only Category as a predictor.
Number of Skin Conductance Responses
Effect of Category
After the Rating data was analyzed, we analyzed the physiological data collected from
the participants in Experiment 1 and Experiment 2. First, for Experiment 1, we analyzed
Distribution of participants’ mental health
scores for each subcategory, combined.
Composite Mental Health Scores
Mental Health Score
Score

71
differences in the number of skin conductance responses (SCRs) between each of the stimuli
groups: Neutral Speech (aka Baseline), Threat Speech, General Hate, and Misogynist Hate.
We addressed first the primary hypothesis-whether there was an effect of Category of
stimuli on the number of SCRs produced, by creating a baseline model with only Category added
as a predictor. Category had a significant main effect on the model, F(3,117)=7.00. Category had
significant main effects on the contrasts between Baseline/Neutral Speech, and the Misogynist
Hate, b=0.14, t(117.00)=4.42, and Threat, b=0.09, t(117)=3.04, categories, but not the General
Hate, b=0.06, t(117)=1.87, category, shown in Figure 12.
Figure 12
Effect of Demographics
Next, we investigated the results of the secondary hypotheses: how demographics affect
SCR responses. We investigated the role of “self-relevancy” by examining the effect of Gender
on the model. An LME model was created, using both Gender and Category as predictors of
Rating. An ANOVA comparing the Baseline model (BIC= -29.04) to the model of Category +
Gender (BIC= -17.50, χ
2
(4)=8.65, was performed, and found that the two models were not
significantly different, and, based on the BIC values, that the Baseline model was better at
explaining variance than the model with Gender as an additional predictor. Additionally, Gender
Means and distributions
of the number of skin
conductance responses
across stimulus
categories.

72
had a main effect on the model, F(1,39)=7.57, but not an interaction with Category. We
investigated further, since it was unexpected that Gender would have a main effect on a model,
without creating a model significantly different from the Baseline model, or having any
interactions with Category. A series of Mann-Whitney U tests were used to compare the number
of SCRs in each category between stimulus groups, and found that in all categories, Other gender
participants had significantly more SCRs than Women: Baseline category, w=113, p=0.02,
Threat category, w=123.5, p=0.04, Misogynist Hate, w=3123.5, p=0.04, General Hate, w=117.5,
p=0.03. This phenomenon was not observed later in Experiment 2, so it is likely that this was
only a coincidence, given the relatively small n. Values are shown in Figure 13.
Further, a Mann-Whitney U Test revealed that women did not have a significantly
different number of skin conductance responses elicited by Misogynist Hate stimuli vs General
Hate stimuli, w=185, p=0.3769, shown in Figure 14. An additional Mann-Whitney U Test
revealed that non-women also did not have a significant difference between the number of skin
conductance responses elicited by Misogynist Hate stimuli vs General Hate stimuli, w=163.5,
p=0.6295, shown in Figure 15.
Figure 13

Number of Skin Conductance Responses
Gender
Other Gender
Women
Number of skin conductance responses per
category, separated by gender.

73
Figure 14 Figure 15
Next, we examined whether being a non-target member of a marginalized group
increased an “empathy/sympathy” effect, and had an effect on the number of SCRs produced for
each stimulus category, by examining the effect of Race and Sexual Orientation on the model.
An LME was created using Race and Category as predictors of the number of SCRs produced.
An ANOVA comparing the Baseline model to the model with Race included as a predictor
demonstrated that there was not a significant difference between the models, χ
2
(4)=8.74.
Additionally, the Baseline model (BIC= -29.04) had a lower BIC value than the Race model
(BIC= -17.58), and thus explained more variance than the Race model. Race did, however, have
a main effect on the larger model, F(1,39)=0.04, but did not have an interaction with Category in
the larger model.
An additional Mann-Whitney U Test revealed, however, that there was actually a
significant difference between the number of skin conductance responses elicited by the
misogynist hate stimuli between white and non-white participants, with non-white participants
exhibiting more responses than white participants, w=109, white mean number responses=0.43,
Women’s mean number of skin conductance
responses to Misogynist Hate stimuli vs
General Hate stimuli.
Other gender participants’ mean number of
skin conductance responses to Misogynist Hate
vs General Hate stimuli.

74
non-white mean number responses = 0.71. This was the only stimulus category for which the
difference between race categories where the difference in number of skin conductance
responses was significantly different.
An LME was also created with Orientation and Category as predictors of the number of
SCRs produced by category. An ANOVA comparing the Baseline model and the Orientation
model demonstrated that there was not a significant difference between the two models,
χ
2
(4)=2.43. The Baseline model (BIC= -29.04) also had a lower BIC than the Orientation model
(BIC= -11.27), and thus explained more variance than the Orientation model. Orientation did not
have a main effect on the larger model, or interaction with Category within the larger model.
Mann-Whitney U tests revealed that there were no differences in the number of skin conductance
responses produced by straight vs queer participants for any stimulus category.
Figure 16 Figure 17
Principal Components Analyses
Next, we began to investigate the role that traits measured by the psychometric
inventories on production of SCRs. We used the Principal Components (PCs) calculated Number of Skin Conductance Responses
Race
White
Non-White
Number of Skin Conductance Responses
Orientation
Queer
Straight
B GH MH T
Category B GH MH T
Category

75
previously for Experiment 1 (See Table 1). We then added PC1 to the Baseline model, which had
only Category as a predictor, and found that the model with Category + PC1 as predictors was
not significantly different from the Baseline model, χ
2
(4)=6.49. PC1 did not improve predictive
power, as the model of PC1+Category (BIC= -15.34) had a higher BIC than the Baseline model
(BIC= -29.04). PC1 did not have a main effect on the larger model, or a significant interaction
with Category within the model. PC1 did, however, have a significant main effect on the contrast
between responses to baseline/neutral stimuli, and general hate stimuli, b= -0.5, t(117)=0.02.
A model including PC2 did not improve prediction (BIC = -14.21) over the Baseline model
(BIC= -29.04), χ
2
(4)=5.36. We found no significant effect of PC2 or interaction with Category
in the larger model. A model including PC3 (BIC= -13.14) also did not improve prediction over
the baseline model (BIC= -29.04), χ
2
(4)=4.29. We found no significant effect of PC3 or
interaction with Category in the larger model. PC 4 did not improve prediction (BIC= -10.74)
over the baseline model (BIC= -29.04), χ
2
(4)=1.89. We found no significant effect of PC4 or
interaction with Category in the larger model. A model including PC5 (BIC= -14.73) did not
improve prediction over the baseline model, (BIC= -29.04), χ
2
(4)=5.89 . We found no
significant effect of PC5 or interaction with Category in the larger model. We thus concluded
that psychometric measurements did not seem to have an effect on the number of SCRs produced
by different stimuli.
Psychometric Variables
Here again we thought it important to investigate the impact of each individual
psychometric variable when it is used as a predictor of responses-in this case, the number of
SCRs produced. In the Principal Components Analysis, the factor loadings were fairly even
between the contributing variables. Given that there were some interactions between the PCs and

76
Category, we wanted to investigate whether or not any stronger relationships were found when
considering the psychometric variables individually.
We created a new Baseline model, with only Category as a predictor for number of SCRs
produced. Some participants were excluded from the previous analyses, because they did not
complete all of their inventories, and thus their data could not be used to compute the PCs. In the
following models, which consider each psychometric variable individually, all participants’ data,
excepting overly-influential participants, and SCR non-responders, could be used. Category had
a main effect on this new Baseline model, F(3,98.35)=6.77. Category also had individual main
effects on the contrasts between the Baseline stimulus category and the Misogynist Hate
category, b=0.13, t(98.57)=4.47, and the Threat stimulus category, b=0.09, t(98.57)=2.95.
We created a model with Category + Ambivalent Sexism as predictors of the number of
SCRs produced. This model (BIC= -35.17) was not significantly different from the Baseline
model (BIC= -52.31), χ
2
(4)=2.43, and did not explain more variance than the Baseline model,
according to BIC values. Ambivalent sexism did not have a main effect on the larger model, F(1,
35.03)=0.003, or a significant interaction with Category, F(3,98.36)=0.82, in the larger model.
We created a model with Category + Social Dominance Orientation (SDO) as predictors
of the number of SCRs produced, (BIC= -39.52). This model was not significantly different from
the Baseline model, (BIC= -39.52). The Baseline model explained more variance than the
model according to BIC values. SDO did not have a significant interaction with Category in the
larger model, F(3, 98.35)=2.21, or a significant main effect on the larger model, F(1, 35.06) =
0.37.
We created a model with Category + Belief in a Just World (BIAJW) as predictors of the
number of SCRs produced. This model (BIC= -38.21) was not significantly different from the

77
Baseline model (BIC= -52.31), χ
2
(4)=5.46, and did not explain more variance, according to BIC
values. BIAJW did have a significant main effect on the larger model, F(1, 35.22)=4.84, but did
not have a significant interaction with Category in the larger model, F(3,98.54)=0.33.
We created a model with Category + Grandiose Narcissism (Narcissism) as predictors of
the number of SCRs produced. This model (BIC= -35.39) was not significantly different from
the Baseline model (BIC= -52.31), χ
2
(4)=2.65, and did not explain more variance than the
Baseline model, according to BIC values. Narcissism did not have a significant main effect on
the larger model, F(1, 35.29)=0.03, or a significant interaction with Category in the larger model,
F(3,98.56)=0.89.
We created a model with Category + Scores from the CPLC Cross-Cutting Symptoms
Measure: Adult (“Mental Health”) as predictors of the number of SCRs produced. This model
(BIC= -35.87) was not significantly different from the Baseline model (BIC= -52.31),
χ
2
(4)=3.13, and did not explain more variance than the Baseline model, according to BIC values.
Mental Health did not have a significant main effect on the larger model, F(1, 34.93)=1.44, or a
significant interaction with Category in the larger model, F(3,98.26)=0.58.
We thus determined that the Baseline model, with only Category as a predictor of the
number of SCRs produced, was the best model.
Area Under the Curve of Skin Conductance Responses
Effect of Category
We next explored the effect of stimulus category on skin conductance responses by
measuring the Area Under the Curve (AUC) of each SCR, as a way to quantify not the number
of SCRs, but their magnitude. We again began by addressing the primary hypothesis-whether
there was an effect of Category of stimuli on the AUC of the SCRs produced, by creating a

78
Baseline model with only Category added as a predictor. We found that Category had a
significant main effect on the model, F(3,114)=3.94. Category also had a significant main effect
on the contrast between Baseline/Neutral Speech, and the Misogynist Hate, b=0.28,
t(114.00)=3.34, category, but not the General Hate, b=0.15, t(114)=1.85, or Threat, b=0.10,
t(114)=1.18, categories.
Figure 18

Effects of Demographic Variables
After establishing that Category did have an effect on AUC models, we addressed the
secondary hypotheses: how demographics impacted AUC. We again assessed the impact of selfrelevancy by examining whether Gender had an effect on the model. We created a model with
both Category + Gender as predictors of AUC (BIC=229.24), and found that it was not
significantly different from the Baseline model (BIC=214.47), χ
2
(4)=4.76, with only Category
as a predictor. Gender did not have a main effect, or a significant interaction with Category
within the larger model.
A Mann-Whitney U test demonstrated that women did not have a significant difference
between the area under the curve of skin conductance responses elicited by misogynist hate
stimuli vs. general hate stimuli, w=278.5, p=0.28. Another Mann-Whitney U test revealed that
Mean values of all
Experiment 1 participants’
area under the curve of skin
conductance responses, by
category, measured in
microsiemens2
.

79
non-women also did not have a significant difference in the area under the curve of skin
conductance responses elicited by the misogynist vs general hate stimuli, w=215, p=0.90.
However, there were significantly different mean areas under the curve for women’s
responses to Threat stimuli, mean AUC =0.17 microsiemens2
, vs non-women’s mean area under
the curve to Threat stimuli, mean AUC=0.43 microsiemens2
, shown in Figure 19. There were
significantly different mean areas under the curve for women’s responses to General Hate
stimuli, mean AUC =0.17 microsiemens2
, vs non-women’s mean area under the curve to General
Hate stimuli, mean AUC=0.53 microsiemens2
, shown in Figure 20.
Figure 19 Figure 20
We further investigated whether Race or Orientation had an effect on the AUC between
categories, in order to investigate whether being a member of a non-target, but still marginalized,
group created a “sympathy” effect. We created a model with Category + Race as predictors
(BIC=229.22), and found that it was not significantly different from the Baseline model
(BIC=214.47), with only Category as a predictor, χ
2
(4)=4.79, and did not explain more variance
Gender Gender
Mean area under the curve for skin
conductance responses to Threat stimuli,
between genders.
Mean area under the curve for skin
conductance responses to General Hate stimuli,
between genders.

80
than the Baseline model, via BIC values. We found that it did not have a main effect, or a
significant interaction with Category within the larger model.
We conducted a series of Mann-Whitney U tests to determine if White vs Non-White
participants exhibited significantly different areas under the curve for skin conductance
responses elicited by each category of stimuli. We found that there was a significant difference in
the Misogynist Hate category, w=109, p=0.04, White mean AUC=0.43, Non-White mean=0.71.
We then created a model with Category + Orientation as predictors (BIC=233.50), and
found that it was also not significantly different from the Baseline model, with only Category as
a predictor (BIC=214.47), χ
2
(4)=0.51, and did not explain more variance than the Baseline
model, via BIC values. Orientation also did not interact with Category, or have a main effect, in
the larger model. A series of Mann-Whitney U tests determined that there were no significant
differences in areas under the curve between straight and queer participants for any stimulus
category.
Principal Components Analyses
We next investigated whether psychometric variables had an impact on the AUC between
categories. We used the principal components (PCs) calculated previously for Experiment 1. We
made a model with Category + PC1 (BIC=239.69) as predictors of AUC, and compared it to the
baseline model with only Category as a predictor (BIC=227.15). We found that the model with
PC1 was not significantly different from the Baseline model, χ
2
(4)=7.56, and did not explain
more variance than the Baseline model, via BIC values. PC1 did not interact with Category or
have a main effect on the larger model.
We created a model with Category + PC2 (BIC=243.74) as predictors of AUC, and
compared it to the Baseline model (BIC=227.15). We found that the models were not

81
significantly different, χ
2
(4)=3.50, and that the model with PC2 did not explain more variance
than the Baseline model, via BIC values. PC2 did not interact with Category or have a main
effect on the larger model.
We created a model with Category + PC3 (BIC=231.40) as predictors of AUC, and
compared it to the baseline model (BIC=227.15). We found that the two models were
significantly different from each other, χ
2
(4)=15.84. PC3 interacted with Category in the larger
model, F(3,114)=4.24, and had main effects on the contrasts between the Baseline category and
the General Hate, b= -0.35, t(114)= -3.53, and Misogynist Hate, b= -0.22, t(114)= -2.18,
categories. However, the baseline model was determined to be the better model via BIC values.
We created a model with Category+PC4 (BIC=246.61) as predictors of AUC, and
compared it to the Baseline model (BIC=227.15). We found that the two models were not
significantly different from each other, χ
2
(4)=0.63. PC4 did not have an interaction with
Category or a main effect on the larger model.
We created a model with Category + PC5 (BIC=245.63) as predictors of AUC, and
compared it to the baseline model (BIC=227.15). We found that the two models were not
significantly different, χ
2
(4)=1.61. PC5 did not have an interaction with Category, or a main
effect on the larger model.
We next removed overly influential points from the model. We then made a new Baseline
model, with only Category as a predictor of AUC, with the smaller dataset. We found that
Category again had a main effect on the model, F(3,81)=5.13. Category also had a significant
main effect in the contrast between Baseline/Neutral Speech and Misogynist Hate,
b=0.10,t(81)=3.69, but no longer had significant main effects in the contrasts between

82
Baseline/Neutral Speech and the General Hate, b=0.03, t(81)=1.00, and Threat, b=0.02,
t(81)=0.42, categories.
We created another model with Category + PC1 (BIC= -75.50) as predictors of AUC, and
compared it to the new Baseline model (BIC= -86.60). We found that the two models were
significantly different from each other, χ
2
(4)=7.63. The Baseline model explained more variance
than the model with PC1 added, according to BIC values. PC1 did not have a significant
interaction with Category in the larger model, F(3,81)=1.79, or a significant main effect on the
larger model, F(1,27)=2.56.
We created another model with Category + PC2 (BIC=216.80) as predictors of AUC, and
compared it to the new Baseline model (BIC=214.47). We found that the two models were
significantly different from each other, χ
2
(4)=17.20. However, the BIC values were again too
close together to accurately determine which model is best. PC3 interacted with Category in the
larger model, F(3,99)=4.78, and had significant main effects on the contrasts between the
General Hate, b= -0.44, t(99)= -3.77, Misogynist Hate, b= -0.26, t(99)= -2.18, and Threat, b= -
0.25, t(99)= -2.11, categories and the baseline category. Because Principal Component Analyses
prioritize minimizing the number of variables, we will continue with the baseline model as the
better model.
We created another model with Category + PC3 (BIC=216.80) as predictors of AUC, and
compared it to the new Baseline model (BIC=214.47). We found that the two models were
significantly different from each other, χ
2
(4) =17.20. However, the BIC values were again too
close together to accurately determine which model is best. PC3 had a significant interaction
with Category in the larger model, F(3,99)=4.78, and had significant main effects on the
contrasts between the General Hate, b= -0.44, t(99)= -3.77, Misogynist Hate, b= -0.26,

83
t(99)= -2.18, and Threat, b= -0.25, t(99)= -2.11, categories and the baseline category. Because
Principal Components Analyses prioritize using the fewest variables possible, we concluded that
the Baseline model was the best model.
Psychometric Variables
We did, however, also investigate whether adding each psychometric variable
individually as predictors would create more meaningful models.
We again created a new Baseline model, with all participants’ data included, minus
individuals identified to be overly influential, or SCR non-responders. In the previous models,
participants who had incomplete psychometric data were excluded, as they could not be included
in calculating the PCs. In this new Baseline model, we found that Category had a significant
main effect on the model, F(3,85.61) =4.24. Category had a significant main effect on the
contrast between the Baseline/Neutral speech category, and the Misogynist Hate category,
b=0.25, t(85.88)=3.14.
We created a model with Category + Ambivalent Sexism (BIC= 173.94) as predictors,
and found that it was not significantly different from the Baseline model (BIC=157.42),
χ
2
(4)=2.47. Ambivalent Sexism did not have a significant interaction with Category, F(3,
85.63)=0.61, or a main effect, F(1,30.22)=0.70, in the larger model.
We created a model with Category + Social Dominance Orientation (SDO) as predictors
(BIC=170.08), and found that it was not significantly different from the Baseline model
(BIC=157.42), χ
2
(4) =6.33. SDO did not have a significant interaction with Category,
F(3,85.75)=2.05, or a significant main effect on the model, F(1,30.37)=0.38.
We created a model with Category + Belief in a Just World (BIAJW) as predictors
(BIC=171.25), and found that it was not significantly different from the Baseline model,

84
(BIC=157.42), χ
2
(4)=5.15. BIAJW did not have a significant interaction with Category, F(3,
85.72)=1.26, or a significant main effect on the model, F(1,30.28)=1.54.
We created a model with Category + Narcissism as predictors, (BIC=167.15), and found
that it was not significantly different from the Baseline model, (BIC=157.42), χ
2
(4)=9.25. The
model did come close to achieving being statistically significantly different from the Baseline
model, however, with a p-value of 0.055. Narcissism had a significant interaction with Category,
F(3,85.53)=3.01, in the larger model, but did not have a significant main effect on the larger
model, F(1, 30.12)=0.69.
We created a model with Category + Mental Health as predictors (BIC=168.72), and
found that it was significantly different from the Baseline model, (BIC=157.42), χ
2
(4)=0.10.
However, the model with Mental Health added had a larger BIC value than the Baseline model,
and thus explained less variance than the Baseline model, and was not a better model. Mental
Health did not have a significant interaction with Category, F(3,85.59)=1.45, in the larger model,
or a significant main effect, F(1, 30.12)=3.84, in the larger model.
Thus, it was determined that psychometric variables do have relationships with AUC, but
that the strongest predictor of AUC was simply the stimulus category.
Heart Rate Data
We continued to assess physiological changes in response to the four categories of
stimuli, Neutral Speech (AKA Baseline), Threat Speech, General Hate, and Misogynist Hate, via
changes in heart rate. We addressed the first primary hypothesis- whether there was an effect of
Category on whether the heart rate increases or decreases in response to a stimulus. We created a
baseline LME model with only Category as a predictor. Category did not have a significant main
effect on the model, F(3,164)=1.35. Additionally, further investigations revealed that adding

85
demographic factors, Principal Components, and psychometric variables did not have an effect
on the model, or produce models significantly different from the Baseline model. This continued
to be true even after overly influential participants were identified and removed as well.

86
Chapter 2- Experiment 2
Methods
Participants
37 participants took part in Experiment 2. There were 25 women and 12 men. There were
6 Bisexual/Pansexual, 5 Gay/Queer, and 26 straight participants. There were 10 White, 18 Asian,
3 Hispanic, 1 Black, and 5 Mixed-Race participants. The participants’ mean age was 21.03 years
old. An additional 10 participants were removed from the dataset because they did not complete
the experiment, there was an issue with the recording equipment, or they did not exhibit skin
conductance responses.
Stimuli
Experiment 2 was largely a replication of Experiment 1. There was, however, a slight
change made to the question participants responded to when assessing the stimuli. To better
ensure that internal assessments, as opposed to external, were made, participants were asked
“How upsetting is this statement?” instead of “How Offensive is this statement?” Participants
were again given a Likert scale of 1 to 7, but here 1 was defined as “Not at all upsetting” and 7
was defined as “Extremely upsetting.”
During Experiment 2 each participant was shown the same stimuli from the Neutral
Speech, Threat Speech, Misogynist Hate Speech, and General Hate Speech categories, that
participants in Experiment 1 were shown. However, for Experiment 2, an additional stimulus
category was added: the Criticisms of Systemic Racism category. All stimuli can be viewed in
the Appendix. As with the other categories, there were twenty stimuli in the Criticisms of
Systemic Racism category, which were length matched with the items from the other categories.
Thus, for Experiment 2, participants were shown a total of 100 stimuli. These stimuli were once

87
again presented in a pseudo-randomized order, where no more than 2 stimuli from a single
category were shown to participants in a row.
Items in the Criticisms of Systemic Racism category contained speech that criticized
social injustice and inequality in predominantly white societies. Some items discussed these
topics by referring to “White People,” instead of using a more non-specific epithet, such as,
“people who benefit from systemic racism due to their own race.” Allowance for such
descriptors reflected the way that people described their own experiences with racial oppression,
in their own words (Nurik 2019). However, items also did not incite hatred or violence toward
white people. Discussing one’s lived experiences, or the objectively true realities of systemic
oppression, was not considered to be incitement of hatred, or “reverse-racism.” These
discussions center on the actions of people who belong to a specific social group, that is only
permitted because of their membership in that group, and are thus not incitements to hatred
towards people based on baseless prejudice (Norton et al 2011, Ballinger 2021, Nelson et al
2018, Hawkins & Saleem 2022). Responses to these items from White vs Non-White
participants (according to self-identification) were compared in order to examine whether there
was a difference in how people from a socially privileged group perceived non-hateful criticisms
of the actions of their identity group, as compared to stimuli that contained hateful content.
Items from the Criticisms of Systemic Racism category were taken from the social media
platforms Reddit and Tumblr. Reddit consists of smaller communities, called “sub-reddits,”
where users discuss a specific topic. Tumblr is a microblogging platform. It was not possible to
take items for this category from the Gab Hate Corpus, as there were none present within the
corpus. We assessed and validated whether these items were considered an incitement to

88
violence or hatred using the guidelines set forth by Kennedy et al 2018, in the same way that the
researchers who created the Gab Hate Corpus did (Kennedy et al 2018, Kennedy et al 2022).
All items were length matched between categories, so that the mean word counts of each
post were not significantly different from one another, to control for effects of cognitive load on
physiological responses between categories. All items were then sorted into three different orders
of presentation, to control for order effect to responses. Each presentation order was pseudorandomized, so that no more than two items from a single category could be presented in a row,
to prevent responses from being either sensitized or desensitized to stimuli of a given category as
a result of exposure to many items in a row from the same category.
Psychometric and Demographic Inventories
The same psychometric and demographic inventories that were used in Experiment 1
were used again in Experiment 2. Participants again completed the ambivalent sexism, belief in a
just world, grandiose narcissism, social dominance orientation, and abridged CPLC CrossCutting Symptoms Measure-Adult mental health inventory. Participants again provided
information regarding their gender, sexual orientation, and race.
Equipment
The equipment used for Experiment 2 was the same as for Experiment 1, with only a
small change in set up. For Experiment 2, one electrode was placed over each collarbone, and
one was placed on the right-side rib cage just above the waist, as opposed to between the
collarbones. This positioning change was made in response to Biopac/AcqKnowledge updating
their recommended configuration for best accuracy.
Procedure

89
In addition to the same equipment, the same procedure for data collection that was used
during Experiment 1, was used again for Experiment 2. The same data processing steps from
Experiment 1 were also used to assemble a dataset of participant stimulus ratings, physiological
responses, emotional response classifications, psychometric data, and demographic data.
Statistical Analyses
As Experiment 2 was largely a replication of Experiment 1, the analyses methods did not
change- the same types of linear mixed effects models, principal components analyses, ChiSquared tests, and Mann-Whitney U tests were used to analyze the data from Experiment 2.
Results
Rating Data
Effect of Category
To investigate whether Category acted as a predictor of Rating for the dataset from
Experiment 2, we created an LME model with only Category added as a predictor. This model
acted as the Baseline model for comparison for all Experiment 2 calculations. It is also worth restating that in Experiment 2, an additional category of stimuli was added to what is otherwise a
replication of Experiment 1. Here, we added items in the new Criticisms of Systemic Racism
category, which, like the other items, were also taken from social media and length matched with
the other stimulus categories. This Baseline model demonstrated that there was a significant
main effect of Category on the model, F(4,103.10)=140.15. Category also had significant main
effects on the contrasts between each stimulus category and the Baseline condition (“Neutral
Speech”): Criticisms of Systemic Racism vs Baseline, b= 1.36, t(4,103.10)= 6.23, General Hate
vs Baseline, b = 3.99, t(4,103.10), and Misogynist Hate vs Baseline, b=3.52, t(4,103.10)=16.15,
shown in Figure 21.

90
Figure 21
Effect of Demographics
With a relationship between Category and Rating reconfirmed, as in Experiment 1, we
once again investigated the effects of self-relevancy on Rating behaviors, via the effect of
Gender. An ANOVA comparing the Baseline model (BIC= 165.93), with only Category added
as a predictor, to the model with Category + Gender (BIC= 181.56) as predictors was performed,
and found that the addition of gender did not create a better model than the baseline model, as
determined by comparing BIC values.
We again investigated whether being a member of another marginalized group increased
the likelihood of someone rating hate speech as offensive, even if they weren’t necessarily a
member of the targeted group themselves, via something like an “empathy effect.” We examined
how Race and Sexual Orientation acted as predictors added to the baseline model of the effect of
Category. Race was added as a predictor along with Category in an LME, and an ANOVA was
performed comparing it to the baseline model. It was determined that the Baseline model
(BIC=248.86) was a better model than Category + Race (BIC=271.38). Another LME was
created with Category + Orientation as predictors, and an ANOVA was performed to compare it
to the baseline Category model. The ANOVA demonstrated that the baseline model
All participants’ mean ratings of each category
of stimuli.

91
(BIC=248.86) was again a significantly better model than the model including Orientation
(BIC=262.79), via BIC values.
A series of Mann-Whitney U tests were performed to examine whether the ratings for
each stimulus category were significantly different between gender, orientation, or race
categories. There were no significant differences in Rating between any of the demographic
dyads.
A Mann-Whitney U test was performed to determine if women’s offensiveness ratings of
Misogynist Hate stimuli were significantly different from ratings of General Hate stimuli. The
ratings were not significantly different (W=401, p-value = 0.09).
A Mann-Whitney U test was performed to determine if women’s offensiveness ratings of
Misogynist Hate stimuli were significantly different from ratings of Criticisms of Systemic
Racism stimuli. Ratings of Misogynist Hate were significantly higher than ratings of Criticisms
of Systemic Racism (W=47, p-value =2.70e-07).
A Mann-Whitney U test was performed to determine if women’s offensiveness ratings of
General Hate stimuli were significantly different from ratings of Criticisms of Systemic Racism
stimuli. Ratings of General Hate were significantly higher than ratings of Criticisms of Systemic
Racism (W=23.5, p-value =2.14e-08).
Figure 22
Women’s mean ratings of
Criticisms of Systemic Racism,
General Hate, and Misogynist Hate
stimuli.

92
A Mann-Whitney U test was performed to determine if non-women’s offensiveness
ratings of Misogynist Hate stimuli were significantly different from ratings of General Hate
stimuli. The ratings were not significantly different (W=98.5, p-value = 0.13).
A Mann-Whitney U test was performed to determine if non-women’s offensiveness
ratings of Misogynist Hate stimuli were significantly different from ratings of Criticisms of
Systemic Racism stimuli. Ratings of Misogynist Hate were significantly higher than ratings of
Criticisms of Systemic Racism (W=16, p-value =0.001).
A Mann-Whitney U test was performed to determine if non-women’s offensiveness
ratings of General Hate stimuli were significantly different from ratings of Criticisms of
Systemic Racism stimuli. Ratings of General Hate were significantly higher than ratings of
Criticisms of Systemic Racism (W=5, p-value =0.0001).
Figure 23
Principal Components Analyses
Next, a Principal Components Analysis (PCA) was done with participant scores on the
psychometric inventories. The Principal Component (PC) loadings are shown in the Table 3
below.
Non-Women’s mean ratings of Criticisms of Systemic
Racism, General Hate, and Misogynist Hate stimuli.

93
Table 3 Principal Component Factor Loadings
Factor PC1 PC2 PC3 PC4 PC5
Ambivalent Sexism -0.5046963 0.26944958 0.19767314 0.77617363 0.17651746
Social Dominance
Orientation
-0.5657300 0.22299933 0.01504821 -0.28022049 -0.74261084
Belief in a Just
World
-0.5644968 0.06659533 -0.11696997 -0.50558106 0.63844675
Narcissism -0.1958051 -0.68978786 0.69413490 -0.05978447 -0.02134505
Mental Health 0.2612182 0.63041732 0.68205106 -0.24463151 0.09644093
A model with PC1 added as an additional predictor created a model (BIC=340.4) that
explained more variance than the baseline model (BIC=366.50), and was significantly different
from the baseline model, χ
2
(5)=50.4 There was a significant interaction between Category and
PC1 in the larger model, F(4,103.20)= 16.26. PC1 had significant main effects in the contrasts
between the Baseline stimulus category, and the Criticisms of Systemic Racism, b=0.36,
t(103.05)=3.19, General Hate, b= -0.41, t(103.05) = -3.61, and Misogynist Hate, b=0.39,
t(103.05) = -3.41, categories.
A model with Category + PC2 as predictors (BIC=386.73) did not improve prediction
over the model of Category+PC1 (BIC=340.4). We found no significant main effect of PC2 or
interaction with Category in the larger model.
A model including PC3 (BIC=382.94) did not improve prediction over the model with
Category+PC1 (BIC=340.4), according to BIC values. We found a significant main effect of
PC3 on the larger model, F(1, 26.16)= 6.10, but no interaction with Category.

94
A model including PC4 (BIC=390.4) did not improve prediction over the model of
Category+PC1 (BIC=340.4). We found no significant main effect of PC4 or interaction with
Category in the larger model.
A model including PC5 (BIC=390.44) did not improve prediction over the model of
Category+PC1 (BIC=390.44). We found no significant main effect of PC5 or interaction with
Category in the larger model.
Thus, the best model to move forward with was determined to be the model with
PC1+Category as predictors. Next, overly influential participants were removed from this model,
and the calculations repeated. Using the smaller dataset, another baseline model was computed.
In this new baseline model, with only Category as a predictor of Rating, Category still had a
main effect on the model, F(4, 83.22)=220.66. Category had significant main effects on the
contrasts between the Baseline stimulus category (“Neutral Speech”) and the Criticisms of
Systemic Racism, b=1.46, t(83.10)=7.86, General Hate category, b= 4.27, t(83.10)=22.97, and
the Misogynist Hate, b=3.79, t(83.10)=20.40 stimulus categories.
We then added PC1 as a predictor to this new Baseline model, and an ANOVA
demonstrated that the model with PC1+Category (BIC=239.87) as predictors was both a better
model than the Baseline model (BIC=248.86) according to BIC values, and significantly
different from the baseline model, χ
2
(5)=32.21. There was a significant interaction between
Category and PC1, F(4,83.31)=9.60 in the larger model. PC1 had a main effect in the contrast
between the Criticisms of Systemic Racism category, and the Baseline category, b=0.46,
t(83.09)=3.78. We investigated the relationship between Rating and PC1 further in the graphs
below, and found that PC1 and the item ratings in the Criticisms of Systemic Racism category
had a steep positive correlation, while item ratings had steep negative correlations with the

95
General Hate and Misogynist Hate categories, and ratings had only a weakly positive correlation
with the Baseline/Neutral Speech and Threat categories. It should be noted that when PC values
are calculated, the signs are reversed, these graphs reflect the inverse of the true direction of the
relationships between Stimulus Category and Rating. Of note, is the fact that the direction of the
relationships between the Misogynist and General Hate categories is the opposite of the direction
of the relationship between Rating and the Criticisms of Systemic Racism category. With the
sign reversal in mind, the figure demonstrates that lower PC1 values are correlated with lower
offensiveness ratings of Hate Based Rhetoric, and higher offensiveness ratings for Criticisms of
Systemic Racism.
Figure 24
PC1 Values between Demographic Groups
A Mann-Whitney U test also demonstrated that there was a significant difference in PC1
values between gender groups (W = 450, p-value = 0.02e-6), with Women having higher PC1
values; importantly, the PC1 values are inversely correlated with scores on the psychometric
inventories, implying that women had lower scores on the inventories than people of other
Inverse relationships between ratings and PC1
values between stimulus categories.

96
genders. Another Mann-Whitney U test demonstrated that there was not a significant difference
in PC1 values between race categories (w=1425, p=0.17). A final Mann-Whitney U test
demonstrated that there was not a significant difference in PC1 values between orientation
categories (w=1100, p=0.19)
As in Experiment 1, we examined the composition of PC1 to determine which
psychometric variables were most responsible for explaining the observed variations in Rating
behavior between stimulus categories. We found that Ambivalent Sexism, SDO, and BIAJW all
accounted for nearly identical portions of the composition of PC1. Notably, both Narcissism and
Mental Health accounted for significantly less of the composition of PC1, and Mental Health had
an inverse relationship with the other variables. We concluded that, as in Experiment 1, it was
necessary to proceed by exploring the effects of the individual psychometric variables on the
Rating model directly.
Psychometric Variables
We next examined the individual effects of the psychometric variables on the rating
models. As in Experiment 1, we created a new baseline model of the effect of Category, with all
participants included. In the models with principal components, participants who had incomplete
psychometric inventory responses were excluded, as their data could not be included in the
calculation of the principal components themselves. There were no overly-influential participants
to remove from this dataset. In this new Baseline LME model, Category once again had a
significant effect on Rating, F(4,146.51) = 250.25; the LME also showed that Category had
significant main effects on the contrasts between ratings of the Baseline stimulus category
(“Neutral Speech”), and the Criticisms of Systemic Racism, b= 1.43, t(146.58)=8.28, General

97
Hate, b=4.23, t(146.58)=24.43, and Misogynist Hate, b=3.67, t(146.58)=21.21, categories, but
not the Threat category, b=0.29, t(146.92)=1.69.
An LME model with Category + Ambivalent Sexism (BIC=480.03) as predictors was
found to be a significantly better model than the Baseline model, with only Category as a
predictor (BIC=488.17), χ
2
(5)=34.187, based on BIC values. The LME also demonstrated a
significant interaction between Ambivalent Sexism and Category, F(4,146.7)=0.061e-5.
Ambivalent sexism had significant main effects in the contrasts between the Baseline category
and the Criticisms of Systemic Racism, b=0.024, t(146.77)=2.20, General Hate, b= -0.031,
t(146.77)= -2.78, and Misogynist Hate, b= -0.031, t(146.77)= -2.80, categories. The mean score
on the ambivalent sexism inventory for Experiment 2 participants was 32.73 with a standard
deviation of 14.32.
Figure 25
An LME model with Category + Social Dominance Orientation (SDO) (BIC=471.13)
scores as predictors was found to be a significantly better model than the baseline model with
only Category as a predictor (BIC=488.17), χ
2
(5)=43.91, according to BIC values. SDO had
significant main effects in the contrasts between the Baseline category and the General Hate, b= -
Ambivalent Sexism
Distribution of Experiment 2 participants’
Ambivalent Sexism scores.

98
0.06, t(147.12)= -3.99, and Misogynist Hate, b= -0.05, t(147.12)= -3.54, categories. SDO was
also found to have a significant interaction with Category in the overall model, F(4,146.76) =
12.50). The mean SDO score among participants in Experiment 2 was 30.76, with a standard
deviation of 11.03.
Figure 26
An LME model with Category + Belief in a Just World (BIAJW) (BIC=486.88) scores as
predictors was found to be a significantly better model than the Baseline model with only
Category as a predictor (BIC=488.17), χ
2
(5)=27.34, according to BIC values. BIAJW had
significant main effects in the contrasts between the Baseline stimulus category and the General
Hate, b= -0.08, t(146.25)= -3.04, and Misogynist Hate, b= -0.07, t(146.25)= 12.59, categories.
BIAJW had a significant interaction with Category in the larger model, F(4, 146.26) = 7.47. The
mean score on the Belief in a Just World inventory among Experiment 2 participants was 20.43
with a standard deviation of 5.81.
Social Dominance Orientation
Distribution of participants’ Social Dominance
Orientation scores.

99
Figure 27
An LME model with Category + Grandiose Narcissism (BIC = 380.15) scores as
predictors did not create a better model than the baseline model with only Category (BIC=366.5),
χ
2
(5)=10.64, as a predictor of Rating, according to their BIC values. The mean score on the
Grandiose Narcissism inventory in Experiment 2 participants was 113, with a standard deviation
of 26.09.
Figure 28
An LME model with Category + Mental Health (measured via the CPLC Cross Cutting
Symptoms Measure-Adult) as predictors (BIC=505.52) was not found to be a better model than
the baseline model with only Category as a predictor of Rating (BIC=488.17), χ
2
(5)=8.70,
according to their BIC values. The mean score on the CPLC Cross-Cutting Symptoms MeasureBelief in a Just World
Distribution of participants’ Belief in a Just
World scores.
Global Grandiose Narcissism
Distribution of participants’ Grandiose
Narcissism scores.

100
Adult inventory, which quantifies mental health, in Experiment 2 participants was 13.38 with a
standard deviation of 9.53.
Figure 29
We next determined the effect size for each of the psychometric variables that improved
the baseline model. For Ambivalent Sexism, eta2= 0.21, for SDO, eta2=0.25, and for BIAJW,
eta2= 0.17. Thus, we proceeded to add these variables to the baseline model in the order of effect
size, SDO first, Ambivalent Sexism second, and BIAJW third.
Since we had already shown that the model with both Category and Social Dominance
Orientation was a better model than the baseline model, we next proceeded to determine if
adding Ambivalent Sexism and/or BIAJW improved the model of Category+Social Dominance
Orientation, via a series of ANOVAs. We determined that the model of Category+SDO
(BIC=471.13) was significantly different from the model of Category+SDO+Ambivalent Sexism
(BIC=510), χ
2
(10) = 13.23, but that, given the lower BIC, the model of only the effects of
Category+SDO was the better model.
We next determined that the model of Category+SDO (BIC=471.13) was not
significantly different from the model of Category+SDO+BIAJW (BIC=514.91), χ
2
(10) = 8.32,
and, given the BIC values, the model of Category+SDO alone was the best model.
Composite Mental Health
Distribution of participants’ combined scores
on mental health inventory.

101
We can thus conclude that while it is certainly not the only driving factor behind rating
behaviors, Social Dominance Orientation is the strongest driver, and is the best predictor for
future rating behaviors.
We next investigated the relationship between Rating and SDO further, and found that
there was a steep positive correlation between SDO and Rating in the Criticism of Systemic
Racism category, compared to steep negative correlations between Rating and the General Hate
and Misogynist Hate categories. There were also moderately positive interactions between SDO
score and Rating in the Baseline/Neutral Speech and Threat categories.
The relationships demonstrated in the figure below also suggest that SDO scores are
inversely related to PC1 values. Remembering that PC1 values are sign-reversed, this suggests
that SDO is largely responsible for mediating whether someone with find Hate Based Rhetoric,
and Criticisms of Systemic Racism, to be upsetting- people who score higher on the SDO
inventory are less likely to be upset by Hate Based Rhetoric, and are more likely to be upset by
Criticisms of Systemic Racism.
Figure 30
Social Dominance Orientation Scores
Relationships between ratings of stimulus
categories and Social Dominance Orientation
scores.

102
Number of Skin Conductance Responses
Effect of Category
For Experiment 2, we analyzed variations in the number of skin conductance responses
(SCRs) between the 5 stimulus groups: Criticisms of Systemic Racism, General Hate,
Misogynist Hate, Threat Speech, and Neutral Speech/Baseline.
We addressed first the primary hypothesis-whether there was an effect of Category of
stimulus on the number of SCRs produced, by creating a baseline model with only Category
added as a predictor. Category did not have a significant main effect on the model, F(4,92)=1.47.
Category only had a significant contrast between Baseline/Neutral Speech, and the Misogynist
Hate Category, b=0.58, t(92)=2.00. We identified and then removed overly influential
participants from the dataset, and reinvestigated the effect of Category on the model. After the
removal, Category still did have a main effect on the larger model, F(4,88)=1.44. Once again,
there was only a significant contrast between the Baseline category and the Misogynist Hate
category, b=0.61, t(88.00)=2.09. Given that the n after removing non-responders was only 19
participants, and that Category did have a significant main effect on the contrast between the
Baseline stimulus category and the Misogynist Hate stimulus category, we proceeded with the
following analyses to determine if any other insights could be gained, since the lack of
significant main effect of Category on the model was likely due to having too small of an n. We
find this likely, considering the results of Experiment 1, which did demonstrate a significant
main effect of Category on the model, with a significant main effect of Category on the contrast
between the Baseline Stimuli and the Misogynist Hate stimuli, and the Threat stimuli. This
seems to pattern with the Experiment 2 results demonstrating that Misogynist Hate stimuli may

103
elicit more skin conductance responses than the Baseline stimuli, but be too under-powered to
achieve significance with such a small n.
Figure 31
Effects of Demographic Category
Next, we again investigated the results of the secondary hypotheses: how demographics
affect SCR responses. First, we investigated the role of “self-relevancy” by examining the effect
of Gender on the model. An LME was created, using both Gender and Category as predictors of
Rating.
An ANOVA comparing the baseline model (BIC= 382.47) to the model of Category +
Gender (BIC= 396.11), χ
2
(5)=10.09, was performed, and found that the two models were not
significantly different, and, based on the BIC values, that the baseline model was better at
explaining variance than the model with Gender as an additional predictor. Gender did interact
with Category, F(4,92)=2.63, but did not have a main effect, F(1,23)=0.11, in the larger model.
Gender had a significant main effect on the contrast between Neutral speech, and Misogynist
Hate Speech, b= -1.33, t(92)= -2.29, but not on any of the other specific stimulus categories. We
investigated further, by performing a series of Mann-Whitney U tests to compare the number of
Number of Skin Conductance Responses by Category
Distributions of
participants’ mean number
of skin conductance
responses to each category
of stimuli.

104
SCRs in each category between stimulus groups. However, we did not find any significant
differences in the number of SCRs in any category between gender groups. Mann-Whitney U
tests revealed that there were no significant differences between the number of skin conductance
responses produced by women vs non-women for each category. Additional Mann-Whitney U
tests demonstrated that neither women nor non-women had a significantly different number of
skin conductance responses to Misogynist hate vs General hate vs Criticisms of systemic racism
stimuli.
Figure 32 Figure 33
Next, we examined whether being a non-target member of a marginalized group
increased an “empathy/sympathy” effect, and had an effect on the number of SCRs produced for
each stimulus category, by examining the effect of Race and Sexual Orientation on the model.
An LME was created using Race and Category as predictors of number of SCRs produced. An
ANOVA comparing the baseline model to the model with Race included as a predictor
demonstrated that there was not a significant difference between the models, χ
2
(5)=2.71;
additionally, the baseline model (BIC= 382.47) had a lower BIC value than the Race model
(BIC= 403.59). Race did not have a main effect on the larger model, F(1,23)=0.36, or interact
Distributions of women’s mean number of skin
conductance responses to Criticisms of Systemic
Racism, General Hate, and Misogynist Hate stimuli.
Distributions of non-women’s mean number of skin
conductance responses to Criticisms of Systemic
Racism, General Hate, and Misogynist Hate stimuli.

105
with Category in the larger model, F(4,92)=0.44. Race did not have any main effects on the
contrasts between the baseline stimulus category and any of the other categories.
An LME was also created with Orientation and Category as predictors of the number of
SCRs produced by category. An ANOVA comparing the baseline model and the Orientation
model demonstrated that there was not a significant difference between the two models,
χ
2
(5)=4.04 ; the baseline model (BIC= 382.47) also had a lower BIC than the Orientation model
(BIC= 401.54). Orientation did not have a main effect, F(1,23)=0.45, on the larger model, or
interaction with Category, F(4,92)=1.08, within the larger model. Orientation also did not have
any main effects on the contrasts between the baseline stimulus category and the other
categories.
Principal Components Analyses
Next, we investigated the relationship between the traits measured by the psychometric
inventories, and production of SCRs. We used the Principal Components (PCs) calculated
previously for Experiment 2, seen in Table 3. We then added PC1 to the baseline model, which
had only Category as a predictor, and found that the model with PC1 was not significantly
different from the baseline model, χ
2
(5)=2.74. PC1 did not improve predictive power, as the
model of PC1+Category (BIC= 403.46) had a higher BIC than the Baseline model (BIC=
382.47). PC1 did not have a significant main effect or interaction with Category in the larger
model.
A model including PC2 did not improve prediction (BIC = 403.37) over the baseline
model (BIC= 382.47), χ
2
(5)=2.83. We found no significant main effect of PC2 or interaction
with Category in the larger model.

106
A model including PC3 (BIC=400.51) was not significantly different from the baseline
model (BIC=382.47), χ
2
(5)=14.06, and did not improve predictive power. PC3 did not have a
significant main effect on the larger model, or significant interaction with Category in the larger
model. PC3 did not have a significant interaction with any of the stimulus categories.
A model including PC4 (BIC= 255.95) was significantly different from the baseline
model, but did not improve prediction over the baseline model, (BIC= 233.92), χ
2
(5)=14.51. PC4
had a significant interaction with Category in the larger model, F(4,92)=3.25, but no significant
main effect on the larger model, F(1, 23)=2.45. PC4 had a significant main effect on the contrast
between the Baseline stimulus category and the Misogynist Hate stimulus category, b=1.18,
t(92)=2.89. However, since PC4 did not improve predictive power compared to the baseline
model, it was not added to the final model.
A model including PC5 (BIC= 395.31) did not produce a better model than the baseline
model (BIC=382.47). PC5 had a significant interaction with Category F(4,92)=2.85, but not a
main effect, F(1,23)=0.15, on the larger model. PC5 did not have any main effects on the
contrasts between the Baseline stimulus category, and the other stimulus categories. Since PC5
did not improve the predictive power of the model, it was not added to the final model. However,
given that there were relationships between some of the PCs and the models, we thought it
important to further explore these relationships by evaluating how models with each
psychometric variable individually added compared to the Baseline model.
Psychometric Variables
We investigated these potential roles of individual psychometric variables by creating
another series of LME models, where Category and each individual psychometric variable were
added as predictors. Unlike in the previous models, where any participants with incomplete

107
psychometric data had to be excluded as part of the calculation of the PCAs, in these models, the
only participants removed were those identified as being overly influential to the Baseline model,
or SCR non-responders. First, a new Baseline model, with only Category as a predictor of the
number of SCRs produced, was created. Category now had a significant main effect on the
model, F(4,99.15)=2.81. Category also had an individual main effect on the contrast between the
Baseline category and the Misogynist Hate, b=0.23, t(99.70)=2.89, category, and the Criticisms
of Systemic Racism, b=0.22, t(99.70)=2.75 category. This stands in contrast to the Baseline
model created for the principal components analyses, which had a smaller n, and did not
demonstrate Category’s main effect on the contrast between Baseline and Criticisms of Systemic
Racism.
Next, a model with Category + Ambivalent Sexism (BIC=117.40) as predictors was
created and compared to the Baseline model (BIC=96.96). The models were not significantly
different from each other, χ
2
(5)=3.71, and, according to BIC values, the Baseline model was the
better model. Ambivalent sexism did not have a main effect on the model, F(1, 24.75)=0.0004,
or a significant interaction with Category, F(4,98.59)=0.94.
Next, a model with Category + Social Dominance Orientation (SDO) as predictors
(BIC=112.63) was created and compared to the Baseline model (BIC=96.96). The models were
not significantly different, χ
2
(5)=8.48, and, according to BIC values, the Baseline model was the
better model. SDO did not have a main effect on the larger model, F(1, 25.43)=0.69, or a
significant interaction with Category, F(4,99.17)=2.01.
We then created a model with Category + Belief in a Just World (BIAJW) as predictors
(BIC=118.04), and compared it to the Baseline model (BIC=96.96). The models were not
significantly different from each other, χ
2
(5)=3.07. BIAJW did not have a significant main effect

108
on the larger model, F(1, 24.56)=1.37, and did not have a significant interaction with Category,
F(4,98.46)=0.44.
We then created a model with Category + Mental Health (measured via the CPLC CrossCutting Symptoms Measure-Adult inventory) as predictors (BIC=114.98), and compared it to the
Baseline model (BIC=96.96). The models were not significantly different from each other,
χ
2
(5)=6.12. Mental Health did not have a main effect on the larger model, F(1, 25.90)=1.30, or a
significant interaction with Category, F(4,99.45)=1.29.
We finally created a model with Category + Narcissism as predictors, and compared it to
the Baseline model. To do so, we first had to create another Baseline model, as there were
participants who had incomplete Narcissism inventories, and thus could not be included. In this
new baseline model, however, Category did not have a main effect, F(4,75.99)=0.52. The model
with Category + Narcissism (BIC=109.22) was not significantly different from the new Baseline
model (BIC=88.63), χ
2
(5)=1.81. Narcissism did not have a main effect on the larger model, F(1,
19.13)=0.39, or a significant interaction with Category, F(4,75.84)=0.47.
Area Under the Curve
Effect of Category
We next explored the effect of stimulus category on skin conductance responses by
measuring the Area Under the Curve (AUC) of each SCR, as a way to quantify not the number
of SCRs, but their Magnitude. We again began by addressing the primary hypothesis-whether
there was an effect of Category of stimuli on the AUC of the SCRs produced, by creating a
baseline model with only Category added as a predictor. We found that Category did not have a
significant main effect on the model, F(4,84)=1.25. We identified overly influential participants,
and removed them from the dataset, to determine if there would be an effect of Category after

109
their removal. We found that, even after overly influential participants were removed, there was
still no effect of Category on the model, F(4,80)=0.35. We will continue describing the analyses
from the dataset with overly-influential participants removed.
Figure 34
Effect of Demographics
To address the secondary hypotheses of how demographics affect responses to the different
stimuli categories, we created models with Gender, Race, and Sexual Orientation added as
predictors to the Baseline model.
First, we investigated the role of self-relevancy on responses to the stimulus categories by
creating a model with Category+Gender as predictors, (BIC=123.03) and compared it to the
Baseline model (BIC=102.18). The models were not significantly different, χ
2
(5)=2.18, and,
based on BIC values, the Baseline model was the better model. Gender did not have a main
effect on the larger model, F(1, 20)=0.32, or an interaction with Category, F(4,80)=0.47. A series
of Mann-Whitney U tests were performed to determine if women vs non-women produced
significantly different areas under the curve for each stimulus category. However, there were no
differences between gender groups for any category. Another series of Mann-Whitney U tests
Mean area under the curve of participants’ skin conductance responses to
each category of stimuli.

110
were performed to determine if women produced significantly different areas under the curve for
Misogynist vs General hate stimuli, but there was no significant difference, w=297.5, p=0.95.
Non-women were also not found to have significant differences in areas under the curve between
Misogynist Hate and General Hate stimuli, w=69.5, p=0.91.
We next investigated whether being a member of a non-target marginalized group had an
effect on responses, via a “empathy” effect. We first investigated the effect of Race on the
model. We created a model with Category+Race as predictors (BIC=119.97), and compared it to
the Baseline model (BIC=102.18). The models were not significantly different from each other,
χ
2
(5)=5.23. BIC values still select the Baseline model as the better model. Race did not have a
main effect on the larger model, F(1, 20)=0.46, or a significant interaction with Category,
F(4,80)=1.23. A series of Mann-Whitney U tests were conducted to determine if there was a
significant difference between areas under the curve between White and Non-White participants,
and found that there were no differences in any stimulus category.
We created a model with Category+Orientation as predictors (BIC=107.41), and
compared it to the Baseline model (BIC=102.18). The models were significantly different,
χ
2
(5)=17.8, and the BIC values indicated that the Baseline model was a better model. Orientation
did not have a main effect on the larger model, F(1, 20)=2.89, but did have a significant
interaction with Category, F(4,80)=4.16. Orientation had a significant main effect on the contrast
between the Baseline stimulus category, and the General Hate category, b= -0.60, t(80)= -3.65. A
series of Mann-Whitney U tests were conducted to determine if there were significant
differences in the areas under the curve produced by straight vs queer participants for each
stimulus category. There was only a significant difference in the General Hate category,

111
w=206.5, p=0.04, straight mean area under the curve = 0.36 microsiemens2
, queer mean area
under the curve = 0.71 microsiemens2
.
Principal Components Analyses
We next began to investigate the effect of psychometric variables on the models by
adding the Principal Components generated earlier to the Baseline model. We began by creating
a model with Category + PC1 as predictors (BIC=124.00), and comparing it to the Baseline
model (BIC=102.18). The two models were not significantly different from each other, χ
2
(5)=1.21. BIC values favor the Baseline model as the better model of the two. PC1 did not have a
main effect on the larger model, F(1, 20)=0.18, or a significant interaction with Category,
F(4,80)=0.26.
We created a model with Category+PC2 as predictors (BIC=113.79), and compared it to
the Baseline model (BIC=102.18). The models were significantly different, χ
2
(5)= 11.42, but
BIC values indicate that the Baseline model is the better model. PC2 had a significant main
effect on the larger model, F(1, 20)=9.20, but not an interaction with Category, F(4,80)=0.99.
We created a model with Category+PC3 as predictors (BIC=121.17), and compared it to
the Baseline model (BIC=102.18). The models were not significantly different, χ
2
(5)=4.03, and
BIC values indicate the Baseline model is the better model. PC3 did not have a significant main
effect on the larger model, F(1, 20)=0.01 or a significant interaction with Category,
F(4,80)=1.03.
We created a model with Category+PC4 as predictors (BIC=118.79), and compared it to
the Baseline model (BIC=102.18). The models were not significantly different, χ
2
(5)=6.42, and
BIC values indicate that the Baseline model is the better model. PC4 did not have a significant
main effect on the larger model, F(1, 20)=0.96, or a significant interaction with Category,

112
F(4,80)=1.42. PC4 had a significant main effect on the contrast between the Baseline stimulus
category, and the General Hate category, b= -0.28, t(80)= -2.33.
We created a model with Category + PC5 as predictors (BIC=116.32), and compared it to
the Baseline model (BIC=102.18). The models were not significantly different, χ
2
(5)=8.89, and
the BIC values indicate that the Baseline model is the better model. PC5 did not have a
significant main effect on the model, F(1, 20)=0.43, or an interaction with Category,
F(4,80)=2.23. PC5 had a significant main effect on the contrast between the Baseline stimulus
category and the General Hate category, b=0.46, t(80)=2.85.
Psychometric Variables
We further investigated the effects of adding the individual psychometric variables as
predictors to the Baseline model. Given that some of the PCs had significant main effects on the
contrasts between the Baseline stimulus category and the General Hate category, it seems
possible that examining the psychometric models individually may reveal a relationship between
themselves and AUC.
First, we created a new Baseline model, in which only overly influential participants, and
SCR non-responders were removed. In the previous models, participants with incomplete
psychometric data could not be included in the calculation of the Principal Components, and thus
could not be included in the models analyzing them. Since these models do not use Principal
Components, all participants can be included in the creation of the Baseline model. We found
that Category did not have a significant main effect on the model, F(4,110.06)=1.21. We
identified overly influential participants, and removed them from the dataset, to determine if
there would be an effect of Category after their removal. We found that, even after overly
influential participants were removed, there was still no effect of Category on the model,

113
F(4,104.35)=0.28. We will continue describing the analyses from the dataset with overlyinfluential participants removed.
Next, we created a model with Category+Ambivalent Sexism as predictors
(BIC=427.15), and compared it to the Baseline model (BIC=403.03). The models were not
significantly different, χ
2
(5)=0.26, and BIC values indicate that the Baseline model is the better
model. Ambivalent sexism did not have a significant main effect on the larger model, F(1,
26.75)=0.06, or a significant interaction with Category, F(4,104.12)=0.05.
We created a model with Category+Social Dominance Orientation (SDO) as predictors
(BIC=425.41) and compared it to the Baseline model (BIC=403.03). The models were not
significantly different from each other, χ
2
(5)=2.00, and BIC values indicate that the Baseline
model is the better model. SDO did not have a significant main effect on the larger model, F(1,
27.09)=0.65, or a significant interaction with Category, F(4,104.40)=0.35.
We created a model with Category+Belief in a Just World (BIAJW) as predictors
(BIC=422.75), and compared it to the Baseline model (BIC=7403.03). The models were not
significantly different, χ
2
(5)=4.66, and the BIC values favor the Baseline model as the better
model. BIAJW did not have a significant main effect on the larger model, F(1, 26.66)=0.24, or a
significant interaction with Category, F(4,104.03)=1.13.
We created a model with Category + Narcissism as predictors (BIC=340.60), and
compared it to the Baseline model (BIC=322.27). The models were not significantly different,
χ
2
(5)=4.80, and the BIC values indicate that the Baseline model is the better model. Narcissism
did not have a significant main effect on the larger model, F(1, 20.94)=1.83, or a significant
interaction with Category, F(4,81.19)=0.78.

114
We created a model with Category+Mental Health (Measured via scores on the CPLC
Cross-Cutting Symptoms Measure-Adult) (BIC=422.82), and compared it to the Baseline model
(BIC=403.03). The models were not significantly different, χ
2
(5)=4.59, and the BIC values
indicated that the Baseline model was the better model. Interestingly, Mental Health did have a
significant main effect on the larger model, F(1, 27.57)=4.56, and did not have a significant
interaction with Category, F(4,104.78)=0.08.
While there were some interactions and effects between variables in the models of AUC
for the Experiment 2 data set, no model was produced that was able to demonstrate a significant
relationship between stimulus category and AUC.
Heart Rate
Effect of Category
We addressed the first primary hypothesis, regarding the effect of Category of stimuli on
changes in heart rate. We examined 5 categories of stimulus: Neutral Speech (aka Baseline),
Criticisms of Systemic Racism, General Hate, Misogynist Hate, and Threat Speech. We created a
baseline model with Category as a predictor. Category had a significant main effect on the larger
model, F(4, 125)=19.10. Category had a significant main effect on the individual contrasts
between the Baseline stimulus category and the Criticisms of Systemic Racism, b= -4.17,
t(125)= -7.27, Misogynist Hate, b= -3.11, t(125)= -5.42, and Threat speech, b= -1.64, t(125)= -
2.85, categories.

115
Figure 35
Effects of Demographic Variables
We addressed the secondary hypothesis of whether or not demographic data had an effect
on heart rate responses to the different categories of stimuli, but found that adding in Gender,
Race, and Sexual Orientation as predictors to models did not create models that were
significantly different from the baseline model. The demographic factors did not interact with
Category in the larger models, either. This remained true even after overly influential
participants were identified and removed from the dataset.
A Mann-Whitney U test revealed that there was a significant difference between
women’s mean change in heart rate interbeat interval in response to Misogynist Hate vs General
Hate stimuli, w=335, p=0.01.
A Mann-Whitney U test revealed that there was a significant difference between
women’s mean change in heart rate interbeat interval in response to Misogynist Hate vs
Criticisms of Systemic Racism stimuli, w=143, p=0.02.
A Mann-Whitney U test revealed that there was a significant difference between
women’s mean change in heart rate interbeat interval in response to General Hate vs Criticisms
of Systemic Racism stimuli, w=40, p=3.89e-7
.
Mean inverse changes in interbeat interval across stimuli categories.

116
Figure 36 Figure 37
Figure 38
A Mann-Whitney U test revealed that there was a significant difference between nonwomen’s mean change in heart rate interbeat interval in response to Misogynist Hate vs General
Hate stimuli, w=126, p=0.001.
Inverse mean changes in women’s interbeat
intervals in response to Misogynist vs General
Hate stimuli.
Inverse mean changes in women’s interbeat
intervals in response to Criticisms of Systemic
Racism vs Misogynist Hate stimuli.
Inverse mean changes in women’s interbeat
intervals in response to Criticisms of Systemic
Racism vs General Hate stimuli.

117
A Mann-Whitney U test revealed that there was not a significant difference between nonwomen’s mean change in heart rate interbeat interval in response to Misogynist Hate vs
Criticisms of Systemic Racism stimuli, w=76, p=0.84.
A Mann-Whitney U test revealed that there was a significant difference between nonwomen’s mean change in heart rate interbeat interval in response to General Hate vs Criticisms
of Systemic Racism stimuli, w=31, p=0.02.
A series of Mann-Whitney U tests were performed to determine if there were significant
differences between Women vs Non-Women, White vs Non-White, and queer vs straight
participants’ changes in heart rate in response to each category of stimuli, but there were no
significant differences.
Figure 39 Figure 40

Inverse mean changes in non-women’s
interbeat intervals in response to General Hate
vs Misogynist Hate stimuli.
Inverse mean changes in non-women’s
interbeat intervals in response to Criticisms of
Systemic Racism vs Misogynist Hate stimuli.

118
Figure 41
Principal Components Analyses
We investigated the potential effect of psychometric variables on the model by examining
the effects of adding the Principal Components, calculated previously for this dataset, as
predictors. We created a model with Category + PC1 as predictors (BIC=576.06), and found that
it was significantly different from the baseline model (BIC=565.60), with only Category as a
predictor. We found that the model with PC1 added was significantly different from the Baseline
model, χ
2
(5)=13.68. PC1 had a significant interaction with Category in the larger model,
F(4,125)= 3.60, but not a main effect, F(1,125)=0.07. PC1 also had a main effect on the contrast
between the Criticisms of Systemic Racism category, and the Baseline category, b=1.01,
t(125)=2.81. However, since the Baseline model had a lower BIC value than the model with
PC1 added as a predictor, PC1 was not kept in the model.
Further models, with PC2-PC5 added as predictors, did not produce models that were
significantly different from the Baseline model, and did not have significant interactions with
Category, or main effects on the larger models. Thus, it was concluded that the best model of the
Inverse changes in non-women’s interbeat
intervals in response to Criticisms of Systemic
Racism vs General Hate stimuli.

119
effects of Category on changes in heart rate was the baseline model itself, with only Category as
a predictor.
Psychometric Variables
Given that there were no significant interactions between any of the principal components
and Category, and the addition of principal components did not improve the Baseline model with
only Category as a predictor, there was not sufficient motivation to investigate the impact of each
individual psychometric variable on the Baseline model, as there was for previous analyses
where principal components with equally loaded factors did improve the Baseline model, in
order to determine if all of the psychometric factors contributed equally to explaining variance
within the model.

120
Chapter 3- Experiment 1+2 Combined Analyses
Given that Experiment 1 and Experiment 2 were largely replications of each other, we
decided to combine the datasets for additional analyses, in order to achieve a larger n. Only the
four stimulus categories that were shared between the experiments were analyzed- responses to
Criticisms of Systemic Racism were not analyzed for the combined dataset. The analyses used in
Experiments 1 and 2 individually were again used for this combined dataset.
Results
Rating Data
Effect of Category
We next combined the Rating results from participants in Experiment 1 and Experiment 2
into a single data set, in order to increase the sample size and power of the results. We
investigated the effects of psychometric and demographic values on Rating behaviors with this
larger dataset, to investigate how a larger sample size potentially changed results and
interpretations of the data.
As we did when we investigated each dataset individually, we first assessed whether
there was a difference in rating between categories that contain hate speech, and categories that
do not, as the primary hypothesis to be confirmed before proceeding with any further analyses.
To do this, we created another Baseline Model by creating an LME with only Category as a
predictor. There was a significant main effect of Category, F(3,215.27)=966.16 on the larger
model. There were significant main effects of Category in the contrasts between the baseline
condition (“Neutral Speech”) and the General Hate, b = 4.61, t(215.13)=42.64, Misogynist Hate,
b=4.07, t(215.13)=37.63, and Threat, b=0.50, t(215.41)=4.60 categories.

121
Figure 42
Effect of Demographic Variables
With a relationship between Category and Rating reconfirmed, we once again
investigated the effects of self-relevancy on Rating behaviors, via the effect of Gender. An
ANOVA comparing the Baseline LME model (BIC= 632.02) to the model with Category +
Gender (BIC= 645.25) as predictors was performed, and found that the addition of gender did
not create a better model than the Baseline model, χ
2
(4)=9.29.
We again investigated whether being a member of another marginalized group
influenced how likely someone would be to rate hate speech as offensive by examining the
effects of adding Race and Sexual Orientation as predictors to the Baseline model. It was
determined that the model of the effect of Category+Race (BIC=626.27) was a better model than
only the effect of Category (BIC=632.02), χ
2
(4)=28.27. Another LME was created with
Category+Orientation as predictors, and an ANOVA was performed to compare it to the baseline
Category model. The ANOVA demonstrated that the baseline model (BIC=632.02) was a
significantly better model than the model including Orientation (BIC=642.17), χ
2
(4)=12.37.
Distributions of participants’ mean rating of
stimuli by category.

122
A series of Mann-Whitney U tests were performed to examine whether the ratings for
each stimulus category were significantly different between gender, orientation, or race
categories. Interestingly, the only significant differences in Rating between any of the
demographic dyads were the Ratings of Baseline stimuli (w=941, p=0.01) and Threat stimuli
(w=902.5, p=0.04) between Race categories. The White mean rating of the Baseline stimuli was
1.24, and the Non-White mean rating was 1.20. The White mean rating of the Threat stimuli was
1.92, and the Non-White mean rating was 1.46.
A Mann-Whitney U test was then performed to determine whether there was a significant
difference between women’s offensiveness ratings of General Hate stimuli vs. Misogynist Hate
stimuli. There was a significant difference, with women rating General Hate stimuli as more
offensive than Misogynist Hate stimuli (W= 1696.5, p-value = 0.008).
Figure 43

A Mann-Whitney U test was then performed to determine whether there was a significant
difference between other gender participants’ offensiveness ratings of General Hate stimuli vs
Misogynist Hate stimuli. There was a significant difference, with other gender participants rating
General Hate stimuli as more offensive than Misogynist Hate stimuli (W=740, p-value = 0.01).
Women’s mean ratings of Misogynist Hate vs
General Hate stimuli.

123
Figure 44
Principal Components Analyses
Next, a Principal Components Analysis was performed using the psychometric data from
the combined dataset. The loadings of each component are shown in the figure below.
Table 4 Principal Components Factor Loadings
Factor PC1 PC2 PC3 PC4 PC5
Ambivalent
Sexism
0.5726655 -0.13021682 -0.04099685 -0.74721666 0.30835737
Social Dominance
Orientation
0.62645724 -0.06255206 0.02723028 0.17758882 -0.75588312
Belief in a Just
World
0.49329543 0.44902043 -0.24219949 -0.24219949 0.48337213
Narcissism 0.05578309 0.48093650 0.87100964 0.87100964 0.01875391
Mental Health 0.18208309 -0.73906127 0.42456964 0.42456964 0.31552487
An LME model with PC1 added as an additional predictor with Category created a model
(BIC=664.74) that explained more variance than the Baseline model (BIC=701.62), and was
significantly different from the Baseline model χ
2
(4)=59.51. There was a significant interaction
between Category and PC1 F(3, 215.4)= 21.10, and a main effect of PC1 on the larger model
Distributions of non-women’s mean ratings of
Misogynist Hate vs General Hate stimuli.

124
F(1, 72.3) = 3.9951 . PC1 had significant main effects in the contrasts between the Baseline
category and the General Hate, b= 0.35, t(215.12) = 5.022, and Misogynist Hate, b=0.41,
t(215.12) = 5.94, categories.
A model including PC2 did not improve prediction (BIC= 386.73) over the model
including Category and PC1 (BIC=664.74). We found no significant effect of PC2 or interaction
with Category in the larger model.
A model including PC3 (BIC=707.94) did not improve the model of Category+PC1
(BIC=664.74). However, PC3 did have a significant interaction with Category in the overall
model F(3, 215.46) = 4.32. PC3 had significant main effects between the Baseline category, and
the General Hate, b=0.34, t(215.14)=3.14, and Misogynist Hate, b=0.29, t(215.15)=2.64,
categories.
A model including PC4 (BIC=722.51) did not improve the model of Category+PC1
(BIC=664.74). We found no significant effect of PC4 or interaction with Category in the larger
model.
A model including PC5 (BIC=716.23) did not improve the model of
Category+PC1(BIC=664.74). PC5 did have a main effect on the larger model, F(1, 71.98) =
4.34.
Thus, the best model moving forward is once again the model with PC1+Category as
predictors. Overly influential participants were removed from this model, and the calculations
repeated. Using the smaller dataset, another Baseline LME model was computed (BIC=632.02).
This new baseline model, with only Category as a predictor of Rating, still had a main effect on
the model, F(3,209.25)=1164.8. Category also had significant main effects in the individual
contrasts between the Baseline stimulus category (“Neutral Speech”) and the General Hate, b=

125
4.69, t(209.11)=46.75, Misogynist Hate, b=4.16, t(209.11)=41.51, and Threat, b=0.52,
t(209.40)=5.15, categories.
We then added PC1 as a predictor to this new baseline model, and an ANOVA found that
the model of PC1+Category (BIC=592.05) was both a better model than the baseline model
(BIC=632.02) according to BIC values, and was significantly different from the Baseline model,
χ
2
(4)=62.50. There was a significant interaction between Category and PC1, F(3,209.39)=22.64.
PC1 had significant main effects in the contrasts between the Baseline category, and the General
Hate, b=0.32, t(209.12)=5.02, and Misogynist Hate, b=0.40, t(209.12)=6.20, categories. We
further examined the shape of the relationship in the graph below, and found that there is a steep
positive correlation between PC1 and Rating in the General Hate and Misogynist Hate
categories, and moderately steep negative correlations between PC1 and Rating in the
Baseline/Neutral Speech and Threat categories. It is important to note that when the principal
components were calculated, their signs were reversed, so these graphs reflect the inverse of the
true relationships between the principal components and ratings. Remembering this, then, the
figure demonstrates that items containing Hate Based Rhetoric are more likely to be rated as
being more upsetting or offensive by people with lower PC1 values.
Figure 45
Inverse relationships between PC1 values and
rating scores by category.

126
Since a model of Category + Race improved the baseline model of Category, as
mentioned previously in this section, we further investigated whether a model with
Category+PC1 as predictors explained more variance than a model with Category+Race as
predictors via ANOVA. We found that the model with Category+PC1 explained more variance
(BIC=592.05) than Category + Race (BIC=626.27) via BIC values; however, there was not a
significant difference between the models, χ
2
(0)=0. Based on the BIC information, we moved
forward with Category+PC1 being the best predictors.
We then looked at whether PC1 values differed between demographic groupings. A
Mann-Whitney U test demonstrated that there was a significant difference in PC1 values
between gender groups (W = 14592, p-value = 0.027e-12), with Women having higher PC1
values; importantly, the PC1 values are inversely correlated with scores on the psychometric
inventories, implying that women had lower scores on the inventories than people of other
genders. Mann-Whitney U tests also demonstrated that there were significant differences in PC1
values between race categories (w=8064, p=0.018)., and orientation categories (w=11408,
p=0.026e-6
).
Finally, we examined the composition of PC1 to determine which of the psychometric
variables it is composed of was driving the interaction between PC1 and Category for Rating
behavior. We once again found, however, that there was a fairly equivalent contribution to the
overall composition from SDO, BIAJW, and Ambivalent Sexism. Notably, there was much less
factor loading on Mental Health and Narcissism in the composition as compared to SDO,
BIAJW, and Ambivalent Sexism. Thus, we determined that it would be beneficial to create
individual models with each psychometric variable added as a predictor independently to better
investigate what traits were driving differences in Rating.

127
Psychometric Variables
We created a new baseline model of the effect of Category, with all participants included.
In the models with principal components, participants who had incomplete psychometric
inventory responses were excluded, as their data could not be included in the calculation of the
principal components themselves. In this new Baseline LME model, Category once again had a
significant main effect on the model, F(3,238.4) = 1728.0; Category also had significant main
effects in the individual contrasts between ratings of the Baseline stimulus category, and the
General Hate, b=4.79, t(238.40)=56.79, Misogynist Hate, b= 4.23, t(238.40)=50.13, and Threat,
b=0.50, t(238.67)=5.85, categories. This is notably different from the baseline model created for
the PCA analyses, since there was a significant main effect of Category on the contrast between
the Baseline and Threat stimuli.
An LME model with Category+Ambivalent Sexism (BIC= -296.63) scores was found to
be a significantly better model than the baseline model with only
Category as a predictor (BIC= -308.77), χ
2
(4)=24.29, via BIC values. The model also
demonstrated a significant interaction between Ambivalent Sexism and Category,
F(3,238.78)=8.50. Ambivalent sexism had significant main effects in the contrasts between the
Baseline category, and the General Hate, b= -0.01, t(238.72)= -2.33 , and Misogynist Hate, b= -
0.02, t(238.72)= -3.09, categories. The mean ambivalent sexism score for all participants in
Experiment 1 + Experiment 2 was 33.43, with a standard deviation of 13.83.
Ambivalent Sexism

128
Figure 46
An LME model with Category + Social Dominance Orientation (SDO) (BIC=632.27)
scores was found to be a significantly better model than the baseline model with only Category
as a predictor (BIC=652.12), χ
2
(4)=42.90, via BIC values. The LME also demonstrated a
significant interaction between SDO and Category, F(3,238.96)=13.58, and that SDO had a main
effect on the model, F(1,80.71)= 5.17. SDO had significant main effects on the contrasts between
the Baseline category and the General Hate,b= -0.03, t(239.14)= -3.84 , and Misogynist Hate, b=
-0.04, t(239.14)= -5.13 , categories. The mean SDO score observed for all participants in
Experiment 1+ Experiment 2 was 30.67, with a standard deviation of 11.56.
Figure 47
Distributions of Experiment 1 + Experiment 2
participants’ Ambivalent Sexism scores.
Social Dominance Orientation
Distribution of Experiment 1 + Experiment 2
participants’ social dominance orientation
scores.

129
An LME model with Category + Belief in a Just World (BIAJW) scores as predictors
(BIC=649.01) was found to be a significantly better model than the Baseline model with only
Category as a predictor (BIC=652.12), χ
2
(4)=26.16, via BIC values. The model of
Category+BIAJW demonstrated that there was a significant interaction between BIAJW and
Category, F(3, 238.20)=8.16. BIAJW had significant main effects in the contrasts between the
Baseline category and the General Hate,b= -0.04, t(238.15)= -3.10 , and Misogynist Hate, b= -
0.05, t(238.15)= -4.06, categories. The mean score on the Belief in a Just World inventory
among Experiment 1+ Experiment 2 participants was 19.90 with a standard deviation of 6.11.
Figure 48
An LME model with Category + Narcissism (BIC=564.60) was found to be a
significantly better model than the Baseline model (BIC=567.03), χ
2
(4)=24.84, according to BIC
values. The LME also demonstrated a significant interaction between Narcissism and Category,
F(3, 203.59)=5.99, and that Narcissism had a main effect on the overall model, F(1,68.47)=7.97.
Narcissism had significant main effects on the contrasts between the Baseline category and the
General Hate, b= -0.01, t(203.14)= -3.72, and Misogynist Hate, b= -0.01, t(203.14)= -3.18,
categories. The mean score on the Grandiose Narcissism inventory among Experiment 1
+Experiment 2 participants was 72.80, with a standard deviation of 33.20.
Belief in a Just World
Distributions of Experiment 1 + Experiment 2
participants’ Belief in a Just World scores.

130
Figure 49
An LME model with Category + Mental Health (BIC=672.39) scores was not
significantly different from the Baseline model, and was found to be a worse model than the
Baseline model (BIC=652.12), χ
2
(4)=2.28, via BIC values. The LME model also demonstrated
that there was not a significant interaction between Mental Heath and Category, F(3,
238.15)=0.93, in the model, or a significant main effect of Mental health on the model,
F(1,79.87)=0.0008. The mean score on the CPLC Cross Cutting Symptoms Measure-Adult
among participants in Experiment 1+ Experiment 2 was 13.52, with a standard deviation of
9.14.
Figure 50
Global Grandiose Narcissism
Distributions of Experiment 1 + Experiment 2
participants’ distributions of Grandiose Narcissism
scores.
Composite Mental Health Scores
Distribution of Experiment 1 + Experiment 2
participants’ mental health scores.

131
Next, we determined the effect sizes of the psychometric variables that improved the
Baseline model: SDO, Ambivalent Sexism, BIAJW, and Narcissism. We calculated the Eta2
values for each variable. For Ambivalent Sexism, the value was 0.10, for SDO, the value was
0.15, for BIAJW, the value was 0.09, and for Narcissism, the value was 0.08. Thus, the variables
were added to the model in the order of SDO, Ambivalent Sexism, BIAJW, and then
Narcissism.
Since we previously established that Category+SDO was a better model than Category
alone, we proceeded directly to determining whether Category+SDO
or Category+SDO+Ambivalent Sexism was the better model. An ANOVA revealed that the
model of Category + SDO (BIC=632.27) was a better model than Category+SDO+Ambivalent
Sexism (BIC=664.05), χ
2
(8)=14.32. An ANOVA also revealed that the model of Category+SDO
(BIC=632.27) was a better model than Category+SDO+BIAJW (BIC=669.52), χ
2
(8)=8.85. An
ANOVA also revealed that the model of Category+SDO (BIC=546.42) was not a better model
than Category+SDO+Narcissism (BIC=560.99), χ
2
(8)= 30.25. Thus, Category+SDO were
determined to be the best predictors for a psychometric model.
The following figure illustrates the relationship between SDO and Rating in the different
stimulus categories. There is a notably steep negative correlation between SDO scores and
ratings of offensiveness/upsettingness in the General and Misogynist Hate categories, implying
that having high endorsement of SDO is associated with being less upset/offended by Hate Based
Rhetoric.

132
Figure 51
Number of Skin Conductance Responses
Effect of Category
To address the primary hypothesis, we again investigated the effect of stimulus category
on the number of skin conductance responses (SCRs) produced. We found that Category did not
have a significant effect on the model of SCRs produced, F(3,186)=2.60. We removed overly
influential participants from the model to determine whether that had an impact of the effect of
Category on the model. Two participants were removed. After their removal, there was a
significant effect of Category on the model, F(3,180)=4.10. Category had a significant main
effect on the contrast between the baseline category, and the misogynist hate category, b=0.15,
t(180)=1.70. Since the effect of Category is an integral part of producing meaningful results, we
will continue with the data from the model with the overly influential participants removed.
Relationship between social dominance orientation
scores vs Rating for each category of stimuli.

133
Figure 52
Effect of Demographic Variables
We next addressed the secondary hypotheses: whether demographics had an impact on
the number of SCRs produced by each stimulus category. We first explored the impact of selfrelevance by adding Gender as a predictor to the model. However, we found that the model with
Category+Gender (BIC=141.09) was not significantly different from the model with only
Category as a predictor (BIC=123.12), χ
2
(4)=3.95. Gender also did not have an interaction with
Category, F(3,180)=0.47, or a main effect, F(1,60)=2.62, in the larger model. We conducted
Mann-Whitney U tests to determine if women had a difference in the number of skin
conductance responses to misogynist hate vs general hate stimuli, but there was no significant
difference w=724.5, p=0.47. We ran Mann-Whitney U tests to determine if non-women had a
difference in the number of skin conductance responses to misogynist hate vs general hate
stimuli, but there was no significant difference, w=405,p=0.51.
We then investigated whether being a member of any marginalized group, even if it is not
the group being targeted, had an effect on the number of SCRs produced in each stimulus
category. We created models with Category+Race, and Category+Sexual Orientation as
predictors to answer this question. We found that the model with Category+Race (BIC=138.53)
Distributions of mean numbers of skin conductance
responses between stimulus categories.

134
as predictors was not significantly different from the baseline model, (BIC=123.12) with only
Category as a predictor, χ
2
(4)=6.51. Race also did not have an interaction with Category,
F(3,180)=1.48, or a main effect, F(1,60)=2.18, in the larger model. The model with
Category+Orientation (BIC=140.08) as predictors was also not significantly different from the
baseline model (BIC=123.12), with only Category as a predictor, χ
2
(4)=4.96. Orientation did not
have an interaction with Category, F(3,180)=1.25, or a main effect, F(1,60)=1.27, in the larger
model.
We performed Mann-Whitney U tests to determine if there were significant differences in
the number of skin conductance responses for each stimulus category between white vs nonwhite, straight vs queer, and women vs non-women participants. There were no significant
differences between any demographic pairing in any category.
We next investigated whether psychometric variables had an effect on the number of
SCRs produced in each category. We used the Principal Components calculated for
Experiment1+Experiment2 previously, which were composed of the results from the
psychometric assessments, to do so.
Figure 53
Distribution of women’s mean number of skin
conductance responses to General Hate vs Misogynist
Hate stimuli.

135
Figure 54
Principal Components Analyses
A model with PC1+Category (BIC=141.44) as predictors was not significantly different
than the baseline model (BIC=123.12) with only Category as a predictor, χ
2
(4)=3.60. PC1 did
not have an interaction with Category, F(3,180)=1.10, or a main effect, F(1,60)=0.34 in the
larger model.
A model with PC2+Category (BIC=140.54) as predictors was not significantly different
than the baseline model (BIC=123.12) with only Category as a predictor, χ
2
(4)=4.50. PC2 did
not have an interaction with Category, F(3,180)=1.29, or a main effect, F(1,60)=0.68 in the
larger model.
A model with PC3+Category (BIC=140.82) as predictors was not significantly different
than the baseline model (BIC=123.12) with only Category as a predictor, χ
2
(4)=4.22. PC3 did
not have an interaction with Category, F(3,180)=0.80, or a main effect, F(1,60)=1.88, in the
larger model.
A model with PC4+Category (BIC=142.73) as predictors was not significantly different
than the baseline model (BIC=123.12) with only Category as a predictor, χ
2
(4)=2.31. PC4 did
Distribution of non-women’s mean number of skin
conductance responses to General Hate vs Misogynist
Hate stimuli.

136
not have an interaction with Category, F(3,180)=0.77, or a main effect, F(1,60)=0.03, in the
larger model.
A model with PC5+Category (BIC=143.40) as predictors was not significantly different
than the baseline model (BIC=123.12) with only Category as a predictor, χ
2
(4)=1.64. PC5 did
not have an interaction with Category, F(3,180)=0.10, or a main effect, F(1,60)=1.36, in the
larger model.
We thus determined that the best model of the number of SCRs produced was the
baseline model, where only Category was used as a predictor.
Psychometric Variables
We next investigated whether the psychometric variables would have an effect on the
models of the number of SCR responses to stimuli when added to the model as predictors
individually. First, we created a new Baseline model, from which only overly influential
participants and participants who were SCR non-responders were removed. In the previous
models, if a participant did not have a complete data for all psychometric inventories, their data
could not be included in the calculation of the principal components, and thus could not be
included in the models with the principal components as predictors. In these analyses,
participants only needed to be excluded from individual analyses they did not have complete
datasets for. For analyses with some participants removed, corresponding additional baseline
models were created.
The new Baseline model demonstrated that there was a significant main effect of
Category on the model, F(3,180.82)=8.48. Category had a significant main effect on the contrast
between the Baseline stimulus category, and the Misogynist Hate stimulus category, b=0.15,
t(181.7)=5.00, the Baseline stimulus category and the General Hate stimulus category, b=0.10,

137
t(180.72) = 3.21, and the Baseline stimulus category and the Threat stimulus category, b=0.09,
t(180.96)=2.99. This notably differs from the Baseline model created for the principal
components analyses, in that there is a significant main effect of Category on the contrast
between the Baseline stimuli and the Threat stimuli in this model.
Figure 55
Next, we created a model with Category+Ambivalent Sexism as predictors (BIC= 18.56),
and compared it to the Baseline model (BIC= -1.79). The models were not significantly different,
χ
2
(4)=1.66. Ambivalent Sexism did not have a main effect on the larger model, F(1,64.00)=0.04,
or a significant interaction with Category, F(3,180.41)=0.54.
Next, we created a model with Category + Social Dominance Orientation (SDO) as
predictors (BIC=17.01), and compared it to the Baseline model (BIC= -1.79). The models were
not significantly different from each other, χ
2
(4)=3.21. SDO did not have a main effect on the
larger model, F(1, 64.08)=0.41 or a significant interaction with Category, F(3,180.50)=0.94.
Next, we created a model with Category + Belief in a Just World (BIAJW) as predictors
(BIC= 18.71), and compared it to the Baseline model (BIC= -1.79). The models were not
Number of skin conductance responses between stimulus
categories for the psychometric Baseline model of
Exp1+Exp2

138
significantly different, χ
2
(4)=1.50. BIAJW did not have a significant main effect on the larger
model, F(1,63.99)=0.13, or a significant interaction with Category, F(3,180.39)=0.46.
Next, we created a model with Category+Narcissism as predictor (BIC=31.04), and
compared it to the Baseline model (BIC= 16.06). The models were not significantly different,
χ
2
(4)=1.90. Narcissism did not have a significant main effect on the larger model,
F(1,17.33)=0.01, or a significant interaction with Category, F(3,50.05)=0.65.
Next, we created a model with Category + Mental Health (measured via scores on the
CPLC Cross-Cutting Symptoms Measure-Adult) as predictors (BIC= 17.23), and compared it to
the Baseline model (BIC= -1.79). The models were not significantly different, χ
2
(4)=2.99.
Mental Health did not have a significant main effect on the larger model, F(1,64.30)=1.06, or a
significant interaction with Category, F(3,180.75)=0.67.
We thus concluded that the best model of the number of SCRs produced in response to
stimuli was the psychometric Baseline model, with only Category as a predictor.
Area Under the Curve
Effect of Category
We investigated whether there was a difference between stimulus categories not just in
the number of SCRs produced, but in their magnitude, via Area Under the Curve (AUC). We
first examined our primary hypothesis: whether the category of stimulus impacts the magnitude
of the SCR produced. We found that Category did indeed have a significant main effect on the
model, F(3,177)=3.70. Category has specific interactions in the contrasts between the baseline
stimuli and the General Hate, b=0.21, t(177)= 2.68, and the Misogynist Hate, b=0.23,
t(177)=2.97, categories.

139
Figure 56
Effect of Demographic Variables
We next investigated the potential role of demographic factors on AUC of SCRs between
categories. We examined the role of Self-Relevancy of statements of hate via the effect of gender
on the models of AUC. We created an LME model with Category+Gender (BIC= -49.20) as
predictors of AUC, and found that it was not significantly different from the baseline model
(BIC= -63.84), with only Category as a predictor, χ
2
(4)=5.65. Gender did not have an interaction
with Category, F(3,120)=1.35, or a main effect, F(1,40)=1.69, in the larger model.
We performed a Mann-Whitney U test to determine if women produced significantly
different areas under the curve for Misogynist Hate vs General Hate stimuli, but there was no
significant difference, w=1149, p=0.31. We performed a Mann-Whitney U test to determine if
non-women produced significantly different areas under the curve for Misogynist Hate vs
General Hate stimuli, but they were not significantly different, w=527, p=0.83.
Distribution of mean area under the curve of
skin conductance responses between stimulus
categories.

140
We performed a series of Mann-Whitney U tests to determine if there were significant
differences in the areas under the curve for women vs non-women for each stimulus category.
There were significant differences between women and non-women’s areas under the curve in
the Threat stimuli category, w=618, p=0.04, women’s mean =0.22 microsiemens2
, non-women’s
mean microsiemens2 =0.37, and in the General Hate category, w=575, p=0.01, women’s mean
=0.24 microsiemens2
, non-women’s mean =0.62 microsiemens2
.
We investigated the effect of empathy from marginalized people, even if they are not the
target of a given hateful statement, by examining how Race and Orientation affect models of
AUC. We created an LME with Category+Race as predictors (BIC= -46.15), and found that it
was not significantly different from the baseline model (BIC= -63.84), with only Category as a
predictor, χ
2
(4)=2.60. Race did not have an interaction with Category, F(3,120)=0.40, or a main
effect, F(1,40)=1.42, in the larger model. We performed a series of Mann-Whitney U tests to
determine if there were significant differences between the areas under the curve produced by
White and Non-White participants, for each category of stimuli, but there were no significant
differences.
We created a model with Category+Orientation (BIC= -52.21) as predictors, and found
that it was not significantly different from the baseline mode (BIC=-63.84), with only Category
as a predictor, χ
2
(4)=8.67. Orientation did have a significant interaction with Category in the
larger model, F(3,120)=2.98, but did not have a main effect, F(1,40)=0.03, in the larger model.
We thus concluded that demographic factors did not have an impact on the AUC of SCRs
produced. We performed a series of Mann-Whitney U tests to determine if there were significant
differences between the areas under the curve produced by queer vs straight participants in
response to each category of stimuli, but there were no significant differences.

141
Principal Components Analyses
We next investigated whether psychometric variables had an impact on the AUC of
SCRs. We added the principal components calculated with the psychometric data for Experiment
1 + Experiment 2 previously (see Table 4). A model with Category+PC1 as predictors (BIC=
420.90) was not significantly different from the baseline model (BIC=405.22), with only
Category as a predictor, χ
2
(4)=6.18. PC1 did not have an interaction with Category,
F(3,177)=2.01, or a main effect, F(1,59)=0.23, in the larger model.
A model with Category+PC2 as predictors (BIC=421.43) was not significantly different
from the baseline model (BIC=405.22), with only Category as a predictor, χ
2
(4)=5.65. PC2 did
not interact with Category in the larger model, F(3,177)=0.02, but did have a main effect in the
larger model, F(1,59)=5.86.
A model with Category+PC3 (BIC=422.50) as predictors was not significantly different
from the baseline model (BIC=405.22), with only Category as a predictor, χ
2
(4)=4.57. PC3 did
not have a significant interaction with Category, F(3,177)=1.53, or a main effect, F(1,59)=0.03,
in the larger model.
A model with Category+PC4 (BIC=424.54) as predictors was not significantly different
from the baseline model (BIC=405.22), with only Category as a predictor, χ
2
(4)=2.53. PC4 did
not have a significant interaction with Category, F(3,177)=0.77, or a main effect, F(1,59)=0.23,
in the larger model.
A model with Category+PC5 (BIC=425.33) as predictors was not significantly different
from the baseline model (BIC=405.22), with only Category as a predictor, χ
2
(4)=1.75. PC5 did
not have an interaction with Category, F(3,177)=0.38, or a main effect, F(159)=0.62, in the
larger model.

142
Next, we identified overly influential participants to investigate how their removal from
the dataset would affect the results of adding principal components as predictors to the model.
Two overly influential participants were removed. A new baseline LME model was created with
this smaller dataset, with only Category as a predictor of AUC. As before, there was a main
effect of Category on the model, F(3,120)=6.56. There was a significant main effect of Category
on the contrast between the Misogynist Hate category and the Baseline stimuli category, b=0.11,
t(120)=4.17.
We then added the PCs back to the model. A model with Category+PC1 (BIC= -45.36) as
predictors of AUC was not significantly different from the baseline model (BIC= -63.84), with
only Category as a predictor, χ
2
(4)=3.60. PC1 did not have an interaction with Category,
F(3,120)=0.55, or a main effect, F(1,40)=0.17, in the larger model.
A model with Category+PC2 (BIC= -54.68) as predictors was not significantly different
from the baseline model (BIC= -63.84), χ
2
(4)=11.14. PC2 did not interact with Category,
F(3,120)=2.01, but did have a main effect, F(1,40)=5.60, in the larger model. However, since the
models were not significantly different, and the baseline model had a lower BIC than the model
with PC2 added as a predictor, PC2 was not kept in the model.
A model with Category+PC3 (BIC= -56.03) as predictors was significantly different from
the Baseline model (BIC= -63.84), with only Category as a predictor, χ
2
(4)=12.53. PC3 did not
have a significant interaction with Category, F(3,120)=2.56, in the larger model, but did have a
main effect, F(1,40)=5.43, in the larger model. PC3 had a significant main effect on the contrast
between the Baseline category and the Misogynist Hate category, b=0.07, t(120)=2.72. Since the
model with PC3 added as a predictor has a larger BIC than the model without PC3, however, it is
not kept in the larger model.

143
A model with Category+PC4 (BIC= -45.09) as predictors was not significantly different
from the baseline model with only Category (BIC= -63.44) as a predictor, χ
2
(4)=1.55. PC4 did
not have a significant interaction with Category, F(3,120)=0.12, or a main effect, F(1,40)=1.19,
in the larger model.
A model with Category+PC5 (BIC= -44.47) as predictors was not significantly different
from the baseline model (BIC= -63.44), with only Category as a predictor, χ
2
(4)= 0.92. PC5 did
not have a significant interaction with Category, F(3,120)= 0.20, or a main effect, F(1,40)=0.33,
in the larger model.
Thus, the final model for predicting the AUC associated with SCRs produced used only
Category as a predictor. However, the fact that there were still interactions of some of the PCs
with the models suggests that it may be worthwhile to investigate whether the individual
psychometric variables that were composited into the PCs had more meaningful effects on the
models on their own.
Psychometric Variables
We next investigated whether any of the individual psychometric variables would affect
the Baseline model when added as predictors. First, we created a new Baseline model, from
which the only participants removed were overly-influential participants, and SCR nonresponders. In the PCA analyses, any participants who did not have complete data for all the
psychometric variables could not be included in the calculation of the PCs, or in the analyses that
used the PCs afterwards. However, since models with the psychometric variables added directly
will not use the PCs, all participants can be included in all the models that they have complete
data for.

144
In this new Baseline model, the main effect of Category did not achieve significance,
F(3,156.9)=2.59. However, it is worth noting that the main effect of Category was very close to
achieving statistical significance, with a p-value of 0.0549. Category had a significant main
effect on the contrast between the Baseline stimulus category, and the Misogynist Hate category,
b=0.17, t(157.97)=1.99. Given that this Baseline model had a larger sample size than the model
used for the PCA, this could suggest that with a larger sample size, statistical significance would
be achieved, as this new model may have been underpowered.
We next created a model with Category + Ambivalent Sexism as predictors
(BIC=426.30), and compared it to the Baseline model (BIC=406.92). The models were not
significantly different from each other, χ
2
(4)=2.08. BIC values indicate that the Baseline model
was the better model. Ambivalent Sexism did not have a significant main effect on the larger
model, F(1,54.25)=0.64, or a significant interaction with Category, F(3,156.19)=0.49.
We next created a model with Category+Social Dominance Orientation (SDO) as
predictors (BIC=423.35), and compared it to the Baseline model (BIC=406.92). The models
were not significantly different, χ
2
(4)=5.03. BIC values indicate that the Baseline model was the
better model. SDO did not have a significant main effect on the larger model, F(1,54.51)=0.80,
or an interaction with Category, F(3,156.42)=1.43.
We next created a model with Category+Belief in a Just World (BIAJW) as predictors
(BIC=425.00), and compared it to the Baseline model (BIC=406.92). The models were not
significantly different, χ
2
(4)=3.38. BIC values indicate that the Baseline model was the better
model. BIAJW did not have a significant main effect on the larger model, F(1,54.43)=1.54, or an
interaction with Category, F(3,156.36)=0.63.

145
We next created a model with Category+Narcissism as predictors (BIC=147.18), and
compared it to the Baseline model (BIC=147.18). The models were not significantly different,
χ
2
(4)=0.32. BIC values indicate that the Baseline model is the better model. Narcissism did not
have a significant main effect on the larger model, F(1,16.70)=0.04, or a significant interaction
with Category, F(3,48.30)=0.09.
We created a model with Category+Mental Health (Measured via the CPLC CrossCutting Symptoms Measure-Adult) (BIC=419.89), and compared it to the Baseline model
(BIC=406.92). The models were not significantly different, χ
2
(4)=8.49, and BIC values indicate
that the Baseline model is the better model. Mental health did have a significant main effect on
the larger model, F(1, 53.89)=8.45, but did not have a significant interaction with Category,
F(3,156.07)=0.21.
Heart Rate
Effect of Category
We again addressed the first primary hypothesis, whether there was an effect of Category
of stimuli on changes in heart rate, with the larger combined datasets of experiment 1 +
experiment 2. We created a baseline LME model with only Category as a predictor of heart rate.
Category did not have a significant main effect on the larger model, F(3,264)=1.33. We
identified overly influential participants, and removed them from the dataset. We found that after
doing so, Category did have a significant main effect on the model, F(3,252)=3.33. Category also
had a significant main effect on the contrast between the Baseline category and the General Hate
category, b=2.96, t(252)=2.23. As such, we will discuss the rest of the results here with regards
to the model with the overly influential participants already removed.

146
Figure 57
Effect of Demographic Variables
We next addressed the secondary hypotheses: whether self-relevance or outgroup
marginalized identity impacted responses to the different categories of stimuli, by investigating
the effects of demographics on these interactions.
We first created a model with both Category+Gender as predictors (BIC=1781.1) of
changes in heart rate. We found that this new model was not significantly different from the
baseline model, with only Category as a predictor (BIC=1759.9), χ
2
(4)=0.88. Gender did not
have a significant interaction with Category, F(3,252)=0.29, or a main effect, F(1,252)=0.00, in
the larger model.
A Mann-Whitney U test demonstrated that women’s changes in heart rate in response to
Misogynist Hate vs General Hate stimuli were significantly different, w=1237, p=0.04. A MannInverse distributions of mean changes in
interbeat interval between stimulus categories.

147
Whitney U test demonstrated that non-women’s changes in heart rate in response to Misogynist
Hate vs General Hate stimuli was significant, w=655, p=0.01.
Figure 58

Figure 59
We next created a model with both Category+Race as predictors of changes in heart rate.
We found that the model with Category+Race as predictors (BIC=1779.7) was not significantly
different from the baseline model, with only Category as a predictor (BIC=1759.9), χ
2
(4)=2.30.
Inverse distributions of women’s changes in
interbeat interval in response to General Hate
vs Misogynist Hate stimuli.
Inverse distributions of non-women’s changes
in interbeat intervals in response to General
Hate and Misogynist Hate stimuli.

148
Race did not have a significant interaction with Category, F(3,252)=0.75, or a main effect,
F(1,252)=0.08, in the larger model.
We finally created a model with Category+Sexual Orientation as predictors of heart rate.
We found that the model with Category + Orientation as predictors (BIC=1798.3) was not
significantly different from the Baseline model, with only Category as a predictor (BIC=1759.9),
χ
2
(8)=5.81. Orientation did not have a main effect in the larger model, F(6,252)=0.83, but did
have a significant interaction with Category, F(2,252)=0.44.
A series of Mann-Whitney U tests were performed to determine if there were significant
differences between Women vs Non-Women, White vs Non-White, and queer vs straight
participants’ changes in heart rate in response to each category of stimuli, but there were no
significant differences.
Principal Components Analyses
We next examined the roles of psychometric variables on the models, by adding the
principal components calculated previously to the models as predictors. We found that a model
with Category +PC1 (BIC=1779.1) as predictors was not significantly different from the
Baseline model (BIC=1759.9), with only Category as a predictor, χ
2
(4)=2.91. PC1 did not have
a significant interaction with Category, F(3,252)=0.94, or a main effect, F(1,252)=0.10, in the
larger model.
We found that a model with Category+PC2 (BIC=1781.5) as predictors of heart rate was
not significantly different from the Baseline model (BIC=1759.9), with only Category as a
predictor, χ
2
(4)=0.57. PC2 did not have a significant interaction with Category, F(3,252)=0.12,
or a main effect, F(1,252)=0.20, on the larger model.

149
We found that a model with Category +PC3 (BIC=1777.3) as predictors was not
significantly different from the Baseline mode (BIC=1759.9), with only Category as a predictor,
χ
2
(4)=4.77. PC3 did not have a significant interaction with Category, F(3,252)=0.12, or a main
effect, F(1,252)=0.20, in the larger model.
We found that a model with Category +PC4 (BIC=177.9) as predictors was not
significantly different from the Baseline mode (BIC=1759.9), with only Category as a predictor,
χ
2
(4)=4.16. PC4 did not have a significant interaction with Category, F(3,252)=1.40, or a main
effect, F(1,252)=0.001, in the larger model.
We found that a model with Category +PC5 (BIC=1781.0) as predictors was not
significantly different from the Baseline mode (BIC=1759.9), with only Category as a predictor,
χ
2
(4)=1.03. PC5 did not have a significant interaction with Category, F(3,252)=0.34, or a main
effect, F(1,252)=0.001, in the larger model.
Given that none of the models with principal components added as predictors were
significantly different from the Baseline model, or had lower BIC values than the Baseline
model, we concluded that the best model was the Baseline model, with only Category as a
predictor.
Psychometric Variables
Given that there were no significant interactions between any of the principal components
and Category, and the addition of principal components did not improve the Baseline model with
only Category as a predictor, there was not sufficient motivation to investigate the impact of each
individual psychometric variable on the Baseline model, as there was for previous analyses
where principal components with equally loaded factors did improve the Baseline model, in

150
order to determine if all of the psychometric factors contributed equally to explaining variance
within the model.
Emotional Response Types
To assess whether there were significantly different distributions of emotional response types
between stimulus categories, we performed Pearson’s Chi-squared tests. For experiment 1, there were not
significantly different emotion distributions between stimuli categories, χ
2=14.98, df=15, p=0.45. For
Experiment 2, there were not significantly different emotion distributions between stimuli categories,
χ
2=28.00, df=20, p=0.11.
For Experiment 1+2 datasets combined, however, the Pearson’s Chi-Squared test revealed that
there was a significantly different distribution of emotion responses between stimulus categories,
χ
2=25.48, df=15, p=0.04. Further Pearson’s Chi-Squared tests were performed to determine if there were
gender differences in the number of each emotion response type in each stimulus category; however, there
were no significant differences in distributions between stimuli categories.
A Durbin-Conover nonparametric pairwise comparisons test was run to determine if there were
significant differences in the number of occurrences of each emotion response type between stimulus
categories/across the distributions, χ
2=1.78, df=3, p=0.62. There were no significant pairwise
comparisons identified, indicating that it is the distributions by stimulus category that differ significantly,
and not differences in the number of each type of emotion response independently. As such, we will
discuss the implications of these results in terms of differences in the context of the overall distributions,
not in differences between the individual response types.

151
Figure 60 Figure 61
Distributions of emotional response types
between stimulus categories for Experiment 1
+ Experiment 2 combined data.
Individual outputs of pairwise comparisons
analysis of emotional response types between
stimulus categories for Experiment
1+Experminet 2 combined data.

152
Chapter 4- Experiment 3
Methods
Participants
500 participants took part in Experiment 3. Experiment 3 consisted of a participant pool
of 307 women, 182 men, 1 genderfluid person, 3 non-binary people, and 7 people who declined
to provide their gender. There were 394 straight, 2 asexual, 39 bisexual, 6 questioning, 1 binary,
39 gay/lesbian, 5 queer, 5 pansexual, and 1 “straight-ish,” participants, and 17 participants who
declined to provide their sexual orientation. There were 129 White, 353 Non-White + Mixed
Race participants, and 18 participants who declined to provide their race. The mean age of
participants was 20.07 years old.
Stimuli
Experiment 3 was a direct replication of Experiment 2 with regards to which stimuli were
used, and how participants were asked to rate them. As such, Experiment 3 used the same
stimuli, and presentation orders, as Experiment 2. All stimuli can be found in the appendix.
Psychometric Inventories
To assess how variables affecting participants’ world view and mental health, we used 5
psychometric inventories: The Ambivalent Sexism Inventory (Glick & Fiske 1996) the Social
Dominance Inventory (Sidanius et al 1994), The Grandiose Narcissism Scale (Foster et al 2015),
The Global Belief in a Just World Scale (Lipkus 1991), and the DSM-5 Self-Rated Level 1
Cross-Cutting Symptom Measure-Adult (American Psychiatric Association 2013).
The Ambivalent Sexism Inventory consists of 22 items, half of which are reverse-coded,
measuring both benign and hostile sexism (Sidanius et al 1994). The scores from all 22 questions
were summed (accounting for reverse-coding) and the single score used for analyses.

153
The Grandiose Narcissism Scale is a 33-Item inventory measuring separate aspects of
narcissism: authority, self-sufficiency, superiority, vanity, exhibitionism, entitlement, and
exploitativeness (Foster et al 2015). The scores from these subsections were aggregated into a
single “Narcissism,” score for analyses.
The Social Dominance Orientation Inventory consists of 16 questions (Sidanius et al
1994), and scores from all questions were aggregated to create a single Social Dominance
Orientation score for analyses.
The Global Belief in a Just World Scale (Lipkus 1991) consists of 7 questions measuring
how fair participants perceive society to be. The scores from these questions were aggregated
into a single “Belief in a Just World,” score for analyses.
The DSM-5 Self-Rated Level 1 Cross-Cutting Symptom Measure—Adult (American
Psychiatric Association 2013) consists of 23 questions, which assess symptomology of mental
health issues across the following domains: Depression, Anger, Mania, Anxiety, Somatic
Symptoms, Suicidal Ideation, Psychosis, Sleep Problems, Memory, Repetitive Thoughts and
Behaviors, Dissociation, Personality Functioning, and Substance Use. However, we did not
include scores from questions measuring “Personality Functioning,” which ask participants to
rate how much they feel a sense of “Not knowing who you really are or what you want out of
life?” and “Not feeling close to other people or enjoying your relationships with them?” when
calculating a total “Mental Health” score for analyses. A strong majority of participants had very
high scores on these questions, indicating that they strongly related to these sentiments. Because
the participants for these experiments consisted entirely of college students, however, we
suspected that this was due to their life stage, and not their mental health. As such, we worried

154
that these scores were artificially heightening the “Mental Health,” scores, and omitted them
from the analyses.
Equipment
For Experiment 3, participants completed the experiment remotely, and physiological
measurements were not recorded. Participants completed the experiment using a computer of
their choosing, not provided by the research team, and in a location of their own choosing. The
stimuli were presented, and responses recorded, using Qualtrics, a free online survey-maker
often used by researchers. Responses to psychometric and demographic inventories were also
recorded in Qualtrics, using the same sel-provided equipment used to collect responses to
stimuli.
Procedure
Data Collection
For Experiment 3, participants were read a description of what their participation would
entail, that that they would be reading posts that were taken from social media, that some of them
would contain hate speech, and some would not, that they would be asked to rate the posts they
were shown, and that they would complete psychometric measures after they finished rating all
of the posts. They were provided with the researchers’ contact information to reach out with any
questions they had while deciding whether or not to participate. If they decided to continue, they
signed an online consent form. They were shown the same practice stimuli that the participants
in Experiments 1 and 2 were, so that they had the same opportunity to get a sense of how the
rating system worked. Participants then read and responded to all of the stimuli at their own
pace, and completed the psychometric and demographic inventories.

155
In terms of analyses of rating, demographic, and psychometric data, Experiment 3 was a
replication of Experiments 1 and 2. Each participant’s data was organized and compiled into a
dataset in the same way that the rating data from previous experiments were.
Data Analyses
All response data from Qualtrics was downloaded, organized by participant, and
formatted into a dataframe in Excel that was compatible with R statistical software for further
analyses.
Statistical Analyses
With regards to assessing the rating data collected for Experiment 3, the same analyses
used to investigate the effects of stimulus category, demographics, and psychometrics for the
previous experiments were used again. However, as there was no physiological data collected,
these were the only analyses performed on Experiment 3 data.
Results
Rating
Effect of Category
Experiment 3 was similar to the previous experiments in that it asked participants to rate
the same stimuli, from all 5 categories, in terms of upsettingness on a scale of 1-7. Experiment 3
differed, however, in that participants did not have any physiological responses recorded, and
data was collected fully remotely, in locations of the participants’ choosing. Because of this, the
sample size was much larger, allowing us to gain more insights into the analyses of this Rating
data.
A Baseline model of the effects of variables on ratings was created via an LME model,
where Category of stimuli was the only predictor. There was a significant main effect of

156
Category in the larger model, F(4,1444)=769.12. There were also significant main effects of
Category in the contrasts between each stimulus category and the baseline condition (“Neutral
Speech”): Criticisms of Systemic Racism vs baseline, b= 1.17, t(1444.0)=15.94, General Hate vs
Baseline, b = 3.31, t(1440)=45.11, Misogynist Hate vs baseline, b=2.90, t(1440)=39.52, and
Threat vs baseline, b=0.64, t(1440)=8.71.
Figure 62
Effects of Demographics
With a relationship between Category and Rating confirmed, we again investigated the
effects of self-relevancy on Rating behaviors, via the effect of Gender. An ANOVA comparing
the baseline model (BIC= 5541.2) to the model of Category + Gender (BIC= 5484.0),
χ
2
(5)=94.57 was performed, and found that the addition of Gender as a predictor did not create a
better model than the Baseline model, according to BIC values. However, Gender did have a
main effect on the larger model, F(4,354)=10.84, as well as an interaction with Category in the
larger model, F(4,1416)=21.61. Gender also had main effects in the contrasts between the
Baseline category and the Criticisms of Systemic Racism, b=0.38, t(1416)=2.59, General Hate,
Distributions of mean rating responses to stimulus
categories.

157
b=0.99, t(1416)=6.70, Misogynist Hate, b=1.17, t(1416)=7.92, and Threat, b=0.35, t(1416)=2.38
categories.
We again investigated whether being a member of another marginalized group increased
the likelihood of someone rating hate speech as offensive, even if they weren’t necessarily a
member of the targeted group themselves, by examining how Race and Sexual Orientation
affected the baseline model of the effect of Category. Race was added as a predictor along with
Category in an LME, and an ANOVA was performed comparing it to the Baseline model. It was
determined that the model with only Category as a predictor (BIC=5390.1) was a better model
than Race+Category (BIC=5421.9), χ
2
(5)=5.46. Race did not interact with Category. Another
LME was created with Category+Orientation as predictors, and an ANOVA was performed to
compare it to the baseline Category model. The BIC values produced by the ANOVA
demonstrated that the baseline model (BIC=5398.3) was a significantly better model than the
model including Orientation as a predictor (BIC=5411.9), χ
2
(5)=23.73. Category and
Orientation did have a significant interaction in the larger model, F(4,1380)=5.73. Orientation
also had a main effect in the contrast between the Baseline category and the Criticisms of
Systemic Racism category, b=0.55, t(1380)=2.93.
To further investigate the impacts that demographics played on Rating behaviors, we
compared the average ratings of Women vs Other gender, White vs Non-White, and Queer vs
Straight participants for each stimulus category. We found that there were significant differences
in rating between Women and Other gender participants in the General Hate category (w=37578,
p=0.06695e-07), Women’s mean rating = 5.5, Non-Women’s mean rating = 4.76, and
Misogynist Hate (w=17446, p=0.04592e-08) categories, Women’s mean rating = 5.04, NonWomen’s mean rating = 4.11, and between Queer and Straight participants in the Criticisms of

158
Systemic Racism (w=13379, p=0.022e-2) category, Queer mean rating = 2.53, Straight mean
rating =3.10, via Mann-Whitney U tests.
Figure 63
Figure 64
Distributions of Women’s vs Non-Women’s
Ratings of General Hate Stimuli.
Distributions of women’s vs non-women’s
mean ratings of Misogynist Hate stimuli.

159
Figure 65
Figure 66
Distributions of women’s mean ratings of
Criticisms of Systemic Racism, General Hate,
and Misogynist Hate stimuli.
Distributions of queer vs straight participants’
mean ratings of Criticisms of Systemic Racism
stimuli.

160
Figure 67
Principal Components Analyses
Next, a Principal Components Analysis (PCA) was done with participant scores on the
psychometric inventories. The Principal Component (PC) loadings are shown in the table below.
Table 5
Factor PC1 PC2 PC3
Ambivalent Sexism 0.5030541 -0.1613645 0.47728290
Social Dominance Orientation 0.5026363 -0.2705023 0.32790791
Belief in a Just World 0.4694799 0.5004522 0.02944778
Narcissism 0.4205076 0.4522851 -0.55149426
Mental Health 0.3115371 -0.6676644 -0.59972021
Distributions of non-women’s mean ratings of
Criticisms of Systemic Racism, General Hate,
and Misogynist Hate stimuli.

161
Figure 68
A model with Category + PC1 as predictors created a model (BIC=5489.2) that was
significantly better than the Baseline model (BIC=5667.5), χ
2
(5)=215.76. There was a
significant interaction between Category and PC1 F(4,103.20)= 16.26 in the larger model. There
were significant main effects of PC1 in the contrasts between the Baseline category and the
General Hate, b= 0.48, t(1444) = 10.55, and Misogynist Hate, b=0.44, t(1444) = 0.89,
categories.
A model with Category + PC2 as predictors created a model (BIC=5614.3) that was
significantly better than the Baseline model (BIC=5667.5), χ
2
(5)=90.67. PC2 had a significant
interaction with Category in the larger model, F(4,1444)= 21.01, as well as a significant main
effect on the model, F(1,361)=9.07. PC2 had significant main effects in the contrasts between the
Baseline category and the Criticisms of Systemic Racism, b= -0.18, t(1444)= -2.52, General
Hate, b= -0.51, t(1444) = -7.10, and Misogynist Hate, b= -0.48, t(1444) = -6.64, categories.
Proportion of variance explained by each
principal component.

162
A model with PC3 added as a predictor to Category created a model (BIC=5364.2) that
improved the Baseline model (BIC=5667.5), χ
2
(5)=70.75. PC3 had a significant interaction with
Category in the larger model, F(4,1444)=16.59, as well as a significant main effect on the larger
model, F(1,361)=5.92. PC3 also had significant main effects in the contrasts between the
Baseline category and the General Hate, b= 0.42, t(1444)=5.30, and Misogynist Hate, b= 0.43,
t(1444)=5.40, categories.
Next, ANOVAs were used to determine how many PCs should be added to the final
model. The model with PC1+Category (BIC= 5489.2) as predictors was already shown to be a
better model than the baseline model((BIC=5667.5), χ
2
(5)=215.76. The model with
PC1+PC2+Category as predictors, (BIC=5417.2) was then determined to be a better model than
the model with only PC1+Category (5489.2), χ
2
(10)=147 as predictors. The model with
PC1+PC2+PC3+Category (BIC=5461.4) as predictors was not a better model than the model
with PC1+PC2+Category (BIC=5417.2), as predictors, χ
2
(20)=105.79, according to BIC values.
We checked for overly influential participants to remove, and then reassess this model, but found
that there were no overly influential participants in this dataset. Thus, the final best model was
determined to have Category, PC1, and PC2 added as predictors of Rating.
Shape of Interactions between PC1, PC2, and Rating
The figures below illustrate the relationships between Rating and PC1, Rating and PC2,
and PC1 and PC2. Very interestingly, we find that the directions of the relationship between PC1
and Rating, and the directions of the relationship between PC2 and Rating, are opposite of each
other in every stimulus category. We also find that there is no relationship between PC1 and
PC2, which is very unexpected. Remembering that PC values are sign reversed, we thus observe
that as PC1 values increase, Ratings of the upsettingness of stimuli from the Criticisms of

163
Systemic Racism, Baseline, and Threat category increase, while Ratings of the upsettingness of
stimuli from the General Hate and Misogynist Hate categories decreases. We observe the
opposite to be true for PC2- as PC2 values increase, ratings of the upsettingness of stimuli from
the Criticisms of Systemic Racism, Baseline, and Threat categories decrease, while ratings of the
upsettingness of items from the Misogynist Hate and General Hate categories increase.

Figure 69
Figure 70
Inverse relationships between PC1 and Rating
values for all stimulus categories.
Inverse relationships between PC2 and Rating
values for all stimulus categories.

164
Figure 71
PC Values Between Demographic Groups
We next explored potential differences in PC 1 and PC2 values between demographic
groups. A series of Mann-Whitney U Tests demonstrated that there was a significant difference
in PC1 values between Women and Other gender participants, w= 276225, p<0.022e-14, and
Straight and Queer participants, w=265713, p=0.023e-4, but no significant difference in PC1
values between White and Non-White Participants, w=287138, p=0.063e-2. Women had higher
PC1 values than Other gender participants. White participants had higher PC1 values than NonWhite participants (but not significantly so), and Queer participants had higher PC1 values than
Straight participants. Given the inverse relationship between PC values and inventory scores, this
demonstrates that women, white participants, and queer participants had lower scores on the
inventories contributing to PC1. It is worth noting that in all cases, the means between
demographic categories are very close together.
We found that for PC2, there was a significant difference in values between Women and
Other gender participants, w=392300, p=0.017, but no difference in P2 values between White
and Non-White participants, w=335238, p=0.11, or Straight and Queer participants, w=234838,
Relationships between PC1 and PC2 values for
all stimulus categories.

165
p=0.36. Other gender participants had higher PC2 values than women, where ambivalent sexism
and belief in a just world scores had an indirect relationship with Rating, and mental health had a
direct relationship with rating. Other gender participants thus had lower ambivalent sexism and
belief in a just world scores, but higher mental health scores than women. The mean values
between genders were very close, however, with women having a larger spread. It is worth
noting that in all cases, the means between demographic categories are very close together.
Figure 72 Figure 73

Other Women Other Women
Differences in distributions of PC1 values
between gender categories.
Differences in distributions of PC2 values
between gender categories.

166
Figure 74 Figure 75
Figure 76 Figure 77
Non-White White Non-White White
Distribution of PC1 values between race
categories.
Distribution of PC2 values between race
categories.
Queer Straight Queer Straight
Distributions of PC1 between orientation categories. Distributions of PC2 between orientation categories.

167
Effects of Psychometric Variables
We next examined the individual effects of the psychometric variables on the rating
models. As in the previous experiments, we created a new Baseline model of the effect of
Category, with all participants included. In the models with principal components, participants
who had incomplete psychometric inventory responses were excluded, as their data could not be
included in the calculation of the principal components themselves. In this new baseline LME
model, Category once again had a significant main effect on Rating, F(4,1840.3) = 1109.1.
Category also had a main effect in the contrasts between the Baseline category(“Neutral
Speech”), and the Criticisms of Systemic Racism, b= 1.18, t(1825.03)=19.62, General Hate,
b=3.40, t(1825.03)=56.37, Misogynist Hate, b=2.92, t(1869.58)=42.20, and Threat, b= 0.56,
t(1825.03)= 9.37, categories.
An LME model with Category+Ambivalent Sexism (BIC=6915.0) scores was found to
be a significantly better model than the baseline model with only Category as a predictor
(BIC=7183.9), χ
2
(5)=307.64, via BIC values. The LME also demonstrated a significant
interaction between Ambivalent Sexism and Category, F(4,1839.56)=83.69. Ambivalent sexism
had significant main effects in the contrasts between the Baseline category and the Criticisms of
Systemic Racism, b=0.01, t(1825.18)= 2.19, General Hate, b= -0.04, t(1825.18)= -11.94, and
Misogynist Hate, b= -0.04, t(1867.70)= -10.47, categories.
An LME model with Category + Social Dominance Orientation (SDO) (BIC=6737.6)
scores as predictors was found to be a significantly better model than the baseline model with
only Category as a predictor (BIC=7183.9), χ
2
(5)=485.01, via BIC values. SDO had significant
main effects in the contrasts between the Baseline category, and the Criticisms of Systemic
Racism, b= -0.01, t(1825.44)= -2.81, General Hate, b= -0.05, t(1825.44)= -18.08, and

168
Misogynist Hate, b= -0.05, t(1863.13)= -14.66, categories. SDO was also found to have a
significant interaction with Category in the overall model, F(4,1838.28) = 138.52.
An LME model with Category + Belief in a Just World (BIAJW) (BIC=7172.9) scores as
predictors was found to be a significantly better model than the Baseline model with only
Category (BIC=7183.9), χ
2
(5)=49.77, as a predictor of Rating via BIC values. BIAJW had
significant main effects on the contrasts between the Baseline category and the Criticisms of
Systemic Racism, b=0.02, t(1825.00)=2.40, General Hate, b= -0.03, t(1825.00)= -3.61, and
Misogynist Hate, b= -0.03, t(1875.10)= -3.34, categories. BIAJW had a significant interaction
with Category in the overall model, F(4, 1841.85) = 12.45.
An LME model with Category + Grandiose Narcissism (BIC = 5253.4) scores as
predictors did not create a better model than the baseline model with only Category
(BIC=5287.4), χ
2
(5)=3.18, as a predictor of Rating, via BIC values.
An LME model with Category + Mental Health (CPLC Cross Cutting Symptoms
Measure-Adult) as predictors (BIC=7120.00) was a better model than the baseline model with
only Category as a predictor of Rating (BIC=7183.9), χ
2
(5)=102.67, via BIC values.
We next determined the effect size for each of the psychometric variables that improved
the baseline model. For Ambivalent Sexism, eta2= 0.15, for SDO, eta2=0.23, for BIAJW,
eta2=0.03, and for Mental Health, eta2= 0.04. Thus, we proceeded to add these variables to the
baseline model in the order of effect size, SDO first, Ambivalent Sexism second, Mental Health
third, and BIAJW fourth.
Since we had already shown that the model with both Category and Social Dominance
Orientation was a better model than the baseline model, we next proceeded to determine if
adding Ambivalent Sexism, Mental Health, and/or BIAJW improved the model of

169
Category+Social Dominance Orientation, via a series of ANOVAs. We determined that the
model of Category+SDO+ Ambivalent Sexism (BIC=6717.7) was a significantly better model
than one with only Category+SDO (BIC=6737.6), χ
2
(10) = 97.47, as predictors.
We next determined that the model of Category+SDO + Ambivalent Sexism + Mental
Health (BIC=6789.5) was not a better model than the model of Category+SDO+Ambivalent
Sexism (BIC=6717.7), χ
2
(20) = 83.26.
The final psychometric model, then, is the model with Category+SDO+Ambivalent
Sexism as predictors of Rating. We can thus conclude that while they are certainly not the only
driving factors behind rating behaviors, Social Dominance Orientation and Ambivalent Sexism
are the strongest drivers, and are the best predictors for future rating behaviors. SDO especially
seems to be important, as even in the smaller sample sizes of the previous experiments, it
significantly improves the baseline models.

170
Discussion
Does Hate Speech Elicit Greater Responses than Non-Hate Speech?
Does Category of Speech Impact Ratings of Offensiveness/Upsettingness?
These studies strongly demonstrated that stimuli from both the misogynist hate category
(targeting women), and the general hate category (targeting various other marginalized groups),
were rated as more offensive/upsetting than non-hate speech, with results that were consistent
across all experiments. While not surprising, these results build on previous research, whose
resultst determined that hate speech was more “harmful,” than hate speech (Leets 1999, Leets
2002, Cowan & Khatchadourian 2003). Our studies indicate that individuals do not just think
that hate speech has a negative impact on the collective culture, but on their individual internal
states as well.
We also found that criticisms of systemic racism, and stimuli containing words known to
elicit a subconscious physiological threat response (“Threat” stimuli), were rated as more
offensive/upsetting than baseline/neutral stimuli, but less offensive/upsetting than hate speech.
While neither of these categories of stimuli contain hateful content, they do either contain either
words known to elicit a physiological threat response, or discuss topics that are emotionally and
politically charged within American culture. These stimulus categories were also found to elicit
more physiological responses than the baseline stimuli in some experiments. Future avenues of
research should investigate whether the presence of a physiological response may have a
mediating effect on how offensive/upsetting someone perceives a statement to be, even if the
effect is subconscious.

171
Does Hate Speech Elicit More Skin Conductance Responses than Non-Hate?
These experiments demonstrated that both misogynist and general hate speech elicited
more skin conductance responses than the baseline stimuli. Interestingly, misogynist hate stimuli
had a significantly larger number of SCRs elicited compared to baseline even when smaller
datasets were used, whereas the general hate stimuli only achieved significance over the baseline
stimuli in the aggregated Exp1+Exp2 dataset. That the misogynist hate stimuli elicited more
responses even in smaller sample sizes is somewhat unexpected, considering that the general
hate stimuli were consistently rated as being more offensive than the misogynist hate stimuli.
Given that gender did not have a significant impact as a predictor in any of the statistical models,
self-relevancy does not appear to mediate the increased response rate of misogynist hate.
Additionally, the threat stimuli were found to elicit more skin conductance responses than the
baseline stimuli in the combined Experiment 1 + Experiment 2 dataset, replicating results from
previous studies (Isenberg 1999).
Does Hate Speech Elicit Skin Conductance Measurements of Greater Magnitudes Than
Non-Hate?
Using the area under the curve of each skin conductance response as a measure to
quantify the “size,” of the responses, we found that stimuli from the misogynist hate category
elicited larger skin conductance responses than the baseline/neutral stimuli in Experiment 1, and
in the combined dataset of Experiment 1+Experiment 2. The general hate speech stimuli,
however, only elicited larger skin conductance responses when the Experiment 1 + Experiment 2
datasets were combined, mirroring the pattern observed in the number of skin conductance
responses elicited. Contrastingly, the stimuli from the threat category did not produce larger skin
conductance response than the baseline stimuli.

172
Both previous research (Isenberg et al 1999), and these experiments, have demonstrated
that certain innocuous-seeming words elicit subconscious fear responses, and, in turn, skin
conductance responses, even when devoid of hateful or threatening contexts. Hate speech stimuli
eliciting skin conductance responses that have a greater magnitude than these responses to both
neutral stimuli and to “threat words,” however, demonstrates that hate speech is not eliciting the
same type of response as “threat words,” and should not be considered to be equivalent to them.
These larger responses demonstrate that exposure to hate speech causes notable levels of
physiological distress that cannot be explained solely by cognitive load or attention orientation.
Further, the confirmation that not only does hate speech elicit more skin conductance responses,
but greater ones, leads us to conclude that hate speech is more arousing than non-hate speech.
It is also worth noting that the criticisms of systemic racism stimuli did not produce skin
conductance responses with larger areas under the curve than the baseline/neutral stimuli, but
that in Experiment 2, the only dataset where physiological responses to criticisms of systemic
racism were analyzed, none of the stimuli elicited skin conductance responses with greater area
under the curve than the baseline stimuli, so we cannot make any conclusions about how
criticisms of systemic racism are processed in comparison to hate speech with regards to
magnitude at this time- the experiment will need to be repeated, with a larger sample size.
Does Hate Elicit Bigger Changes in Heart Rate Than Non-Hate?
The results regarding changes in heart rate were somewhat inconclusive. In Experiment
2, misogynist hate stimuli and criticisms of systemic racism stimuli elicited significantly
different changes in heart rate than baseline stimuli, demonstrating increases in heart rate. For
Experiment 1+2, however, only general hate stimuli elicited a significant change (increase in
heart rate) compared to baseline. This does not necessarily imply that these decreases correspond

173
directly to increases in relaxation, however, given that decreases in heart rate can be associated
with attention orientation responses, and different emotional responses (Graham & Clifton 1966,
Zimmer & Richter 2023). However, increases in heart rate indicate negatively valent responses,
so this does, minimally, demonstrate that hate speech elicits stress responses.
However, the heart rate data from Experiment 1 did not reveal any relationships between
category and changes in heart rate, but this may have been because data for Experiment 1 was
collected in a noisier area than Experiment 2. Given that attention orientations result in decreases
in heart rate (Graham & Clifton 1966, Zimmer & Richter 2023), this may have masked any
effect the stimuli would have actually elicited. Further testing is needed to determine how hate
speech affects heart rate more conclusively.
However, given that misogynist hate was found to elicit both more, and larger, skin
conductance responses, the findings from Experiment 2 that misogynist hate stimuli elicited
greater increases in heart rate seem to agree with these other results. The elicitation of more skin
conductance responses, larger skin conductance responses, and increases in heart rate, align to
reveal a picture of hate speech, especially misogynist hate speech, eliciting physiological stress
and arousal responses within this dataset. That misogynist hate stimuli did not elicit significantly
different changes in heart rate when the data sets were combined, however, while the general
hate stimuli did, may be pointing to the lack of significant difference in heart rate changes being
the results of different specific emotions “cancelling out,” the significant observable effects, as
some response types elicit increases in heart rate, while others elicit decreases. Further research
should be conducted with a larger sample size to further investigate the potential similarities in
trends in emotional response patterns, and trends in the direction of heart rate change, in
response to each type of stimulus.

174
Does Self-Relevance Affect Hate Responses?
Does Self-Relevance Affect Ratings of Hate Speech vs. Non-Hate?
Contrary to what we hypothesized, we found that self-relevance did not have an impact
on ratings of, or physiological responses to, hate speech. Gender did not have an effect on the
models of the number of skin conductance responses, area under the curve of skin conductance
responses, or changes in heart rate.
While women did rate both categories of hate speech as more offensive/upsetting than
participants of other genders, they did not rate misogynist hate speech as more offensive than
general hate speech. In fact, in Experiment 1 and Experiment 3, they actually rated general hate
stimuli as more offensive than the misogynist hate stimuli. This mirrored the rating pattern of
non-women, who also rated general hate speech as more offensive than misogynist hate speech.
In the other experiments, there was no difference between ratings of misogynist vs general hate
speech for either gender group. Taken together, this suggests that women’s greater
offensiveness/upsettingness ratings for misogynist hate compared to the ratings from participants
of other genders was due to the general tendency for women to be less tolerant of hate speech,
(Cowan and Khatchadourian 2003, Cowan et al 2008, Lo Cricchio & Stefanelli 2023, LeMaire
2014), not due to self-relevance.
Does Self-Relevance Affect Physiological Responses?
Complementing these results, we found that self-relevance did not have an effect on
physiological responses, either. We found that self-relevance consistently had no impact on skin
conductance responses, either in terms of number of responses elicited, or area under the curve.
Further, gender did not have a significant effect as a predictor in models of skin conductance
responses.

175
We found that self-relevance did not have an impact on changes in heart rate. While in
some experiments, women were observed to have greater increases in heart rate in response to
misogynist hate speech than to general hate speech, this pattern was also observed in the
corresponding non-women participants. Thus, it seems likely that these differences were due to
an inherent property of the misogynist hate stimuli, not due to self-relevance. These results
continue to support the observation that misogynist hate speech was more arousing than general
hate speech, for reasons we do not understand, and in direct contradiction with
offensiveness/upsettingness ratings.
Does Inter-Group Empathy Affect Responses to Hate Speech?
These experiments predominantly did not demonstrate that members of a non-targeted
marginalized group respond differently to hate speech than non-marginalized people, either via
rating behavior or physiological responses, with only a few exceptions. Women were found to
consistently rate hate speech as more offensive than their counterparts of other genders, which
replicates many previous studies’ findings (Cowan and Khatchadourian 2003, Cowan et al 2008,
Lo Cricchio & Stefanelli 2023, LeMaire 2014). However, our data offer interesting context to
this conversation. It has been somewhat unclear in the literature whether women’s differential
assessments of hate speech were due to an inter-marginalized group “empathy,” effect, or due to
how women tend to be socialized. Across cultures, women are often taught to be more
empathetic and less tolerant of perceived aggression in general, not necessarily only in regards to
specific groups (Cowan and Khatchadourian 2003, Cowan et al 2008, Lo Cricchio & Stefanelli
2023, LeMaire 2014). This is also demonstrated by women’s collective tendency to be less
supportive of ideologies associated with authoritarianism, such as social dominance orientation,

176
belief in a just world, etc. (Dambrun et al 2010, Pratto et al 2011, O’Connor et al 1996, Russell
& Trigg 2004, Cowie et al 2019).
Our results appear to demonstrate that this gender-effect is more likely to be due to
women’s socialization, since neither race nor sexual orientation showed similar effects. If an
inter-marginalized group effect was driving women’s behaviors, we would not expect them to be
the only demographic to display this behavior.
This was true for both physiological responses and offensiveness/upsettingness ratings.
Further research should explore whether these results are replicated with larger data sets, and
with more varied age demographics sampled, to verify that there is not a generation effect on
these results, considering Boeckmann & Liew 2002 and Sanchez et al 2017 did seem to observe
an “empathy,” effect between marginalized groups. However, within the context of only our own
experiments, this finding is consistent with the demonstration that demographics are not good
predictors of responses to hate speech, but psychometric variables are.
Further, sexual orientation did have some effects on responses in Experiment 2. Queer
participants were found to rate criticisms of systemic racism as significantly less offensive than
straight participants, which will be discussed in more detail later, in the Criticisms of Systemic
Racism section of this discussion.
Additionally, queer participants were found to have significantly greater areas under the
curve in response to general hate stimuli than straight participants. In fact, queer participants’
mean areas under the curve in response to general hate stimuli was approximately twice the size
of the mean areas under the curve produced by straight participants.

177
Finally, there was a significant difference in area under the curve of responses to
misogynist hate between white and non-white participants in Experiment 1, with non-white
participants having significantly greater areas under the curve than white participants.
However, in all three of these cases, sexual orientation/race did not improve the
predictive ability of the linear mixed effects models. Further research should be done to
determine if these results were simply due to chance, or to an effect that is not visible to us at this
sample size. Based only on these results, we do not have enough evidence to conclude that
intergroup empathy was either occurring, or responsible for these differences.
Do Psychometrics Affect Responses to Hate Speech?
Rating Responses
Our results demonstrate that psychometric variables are the best predictors for how
someone will rate the offensiveness/upsettingness of hate speech and criticisms of systemic
racism, but not the non-hate speech stimuli, as hypothesized. The relationships between
psychometrics and ratings are multi-layered between individual and aggregated interactions.
Individually, the best predictor of rating behavior is social dominance orientation. The
effect of social dominance orientation is strong enough that it was observed as a significant
predictor in each individual experimental model, not just the combined Experiment
1+Experiment 2 dataset, or the larger Experiment 3 dataset. Further, in all of these models, social
dominance orientation explained more variance than any of the other psychometric variables.
Achieving this level of significance even when sample sizes for some experiments were fairly
small underscores the strength of its effect. These results are not surprising, though, as social
dominance orientation has been demonstrated previously by the literature to be a predictor of

178
prejudice and engagement in violence against marginalized groups (Castellanos et al 2023, Pratto
et al 1994, Gordon 2021, Schrader et al 2024).
Following social dominance orientation, belief in a just world and ambivalent sexism also
had a significant, but more limited degree of predictive ability. In the smaller datasets, they
demonstrated an effect on rating behavior, but did not explain enough variance that was not
already explained by social dominance orientation to merit being added to the final models.
Since social dominance orientation, belief in a just world, and ambivalent sexism are known to
have strong positive correlations with each other (Todorovic et al 2020), having overlapping
areas of explained variance seems logical. In the much larger Experiment 3 dataset, however,
ambivalent sexism did explain enough separate variance to merit being added to the final model
along with social dominance orientation, implying that it was explaining variance that was not
accounted for by social dominance orientation when a larger n was achieved. Further research
should be conducted to further investigate this subtler effect of sexism on rating behaviors.
When considering aggregated predictive effects of psychometrics, however, the story gets
a bit more complicated. Using principal components analyses revealed that all experiments’ PC1
values were fairly evenly loaded across all the psychometric factors, especially social dominance
orientation, ambivalent sexism, belief in a just world, and narcissism. For all of the smaller
models, only PC1 was added to the final models. In experiment 3, however, both PC1 and PC2
were added to the final model. PC2 loaded fairly evenly across belief in a just world, narcissism,
and mental health scores, where mental health had an opposite-direction loading than belief in a
just world and narcissism, implying an opposite direction relationship on rating behaviors.
Despite sharing loading on multiple factors, PC1 and PC2 had no relationship with each
other, and had opposite directional interactions with each stimulus category. Thus, they each

179
independently explain significant parts of the total variance observed in rating behaviors. We
posit that each of these principal components represents an independent “profile,” of respondent.
Given that PC1 loads fairly evenly across all of the psychometric variables, with all variables
loading in the same direction in relation to each other, PC1 appears to measure the spectrum
along which someone endorses or rejects an authoritarian mindset, and has greater symptoms of
mental health issues, such as depression and anxiety (Note that higher scores on the mental
health inventory = higher scores on measurements of presence of symptoms). Participants with
high PC1 scores rated hate speech as less offensive, and criticisms of systemic racism as more
offensive, than participants with low PC1 scores, who did not endorse an authoritarian mindset.
PC2 appears to create a profile of whether or not one believes that all people are being
recognized appropriately by society, and how that impacts their mental health. Participants who
had high PC2 values had high scores on measures of belief in a just world and narcissism, and
low scores on measures of mental health symptomology. These scores suggest that these
participants view themselves as more important or worthy than other people, while
simultaneously believing that those who have social privilege earned that privilege through
systems that bolster the hard-working and just, and punish those who do not work hard enough,
and the unjust. They may believe that they have earned all of their success and social privilege
individually, and deserve increased respect and power, while those who have not attained the
same degree of success are being justly penalized by social paradigms. These participants also
exhibit good mental health, while those with low PC2 values do not. We speculate that perhaps
mental health is stressed by the lack of belief in a just world. These individuals rate hate speech
as less offensive, and criticisms of systemic racism as more offensive, than people with low PC2
scores.

180
Conversely, people with low individual PC2 scores seem to respond in a way that
suggests they may view the world as an unjust place, and that many people, perhaps themselves
included, have to work harder to find success under current social structures that favor some
groups over others. They do not seem to endorse the idea that they are better than other people,
or are more deserving of respect and power. The stress of being aware of some of the world’s
injustice may also be taking a toll on their mental health- resulting in greater presentation of
symptoms of mental health issues, perhaps in a way similar to compassion fatigue, though we do
not have data to support this claim as more than speculation. These individuals rate hate speech
as more offensive, and criticisms of systemic racism as less offensive, than people with high PC2
scores.
All of these profile descriptions need further investigation before they can conclusively
be accepted as accurate descriptions of participants.
Physiological Responses
Physiological responses to all categories of stimuli were not related to scores on
psychometric inventories. However, relationships between the category of stimuli and
physiological arousal did exist. This importantly denotes that, regardless of individual beliefs, or
how offensive/upsetting someone considers hate speech to be, exposure to hate speech results
increased stress levels, at least in the short term. This could suggest that, over time, exposure to
hate speech would have a negative impact on the physical health of the entire population exposed
to it, not just those targeted by or upset by hate speech. More research would clearly need to be
done to conclude that that is an accurate assessment of the long-term impacts of exposure to hate
speech, and to examine how this phenomenon would interact with the desensitization that

181
typically occurs after recurrent exposure to hate speech, but our results do provide motivation for
such studies to be necessary. (Soral et al 2017, Pluta et al 2023).
Criticisms of Systemic Racism
Rating Responses
As previously stated, criticisms of systemic racism were rated as being more
offensive/upsetting than neutral stimuli, but less offensive/upsetting than either category of hate
speech stimuli. This conclusively demonstrates that people do not cognitively conflate the two as
being equally upsetting or offensive.
These experiments expand on the existing literature surrounding “reverse racism,” the
belief that white people in white-dominated societies experience discrimination based on their
race, and that anti-racist measures are harmful to white people, and reduce their rights and
freedoms (Norton et al 2011, Ballinger 2021, Nelson et al 2018, Hawkins & Saleem
2022). Discussions of “reverse racism,” are importantly distinct from discussions criticizing
systemic racism. Criticisms of systemic racism, as we have defined its parameters, address the
fact that institutional racism exists, and has a negative effect on both individuals and our society.
While untested, we expect that there would be strong overlaps between people who believe that
“reverse racism,” exists, and people who would find criticisms of systemic to be offensive.
Notably, our definition of criticisms of systemic racism does not consider “reverse racism,” to be
a legitimate form of racism, and thus, criticisms of the concept of “reverse racism,” are not
considered to be “criticisms of systemic racism.” Future research should further investigate this
relationship directly.
Previous research often describes viewpoints from the perspective of those who claim,
“reverse racism,” exists as “zero-sum game,” - where, in order for someone to gain social rights,

182
they must be taken away from someone else. Increasing rights for one group of people
objectively does not take them away from privileged groups, but this perception remains
persistent among members of white populations (Norton et al 2011, Ballinger 2021, Nelson et al
2018, Hawkins & Saleem 2022). These previous studies do not demonstrate that this is a
perception that all white people share, but do demonstrate that it is mediated most strongly by the
zero-sum belief, even when psychometric characteristics such as social dominance orientation
are accounted for, as well as by political ideology, in tandem with engagement with political
news on social media (Ballinger 2021, Hawkins & Saleem 2022). In the US citizens studied,
participants who self-identified as Conservative and Republican showed a positively correlated
belief in “reverse racism,” with increased political engagement, while those who self-identified
as Liberal and Democrats showed a negative correlation between belief in “reverse-racism,” and
increased political engagement (Hawkins & Saleem 2022).
Surprisingly, in our experiments, race did not mediate the ratings of this category of
stimuli. The only demographic variable we measured that had an effect on these ratings was
sexual orientation, where queer participants rated the stimuli as less offensive than straight
participants. Queer people are known to be more likely to identify as Liberal, as well as being
more supportive of equality movements for all marginalized groups than their straight
counterparts (Golebiowski 2019), so these results appear to correspond with the findings that
Liberal ideology is a strong predictor of how someone will respond to “reverse racism,”
(Hawkins & Saleem 2022). Given that there were no effects of gender or race on the rating
models, it seems more likely that differences in ideologies between straight and queer
participants mediated this difference in rating, as opposed to the “empathy” hypothesis we

183
proposed, where we expected members of one marginalized group to be more sympathetic to the
treatments of other marginalized groups, out of a sense of shared experiences of discrimination.
While more research would certainly need to be done to confirm this hypothesis, the fact
that queer participants produced significantly greater areas under the curve in response to the
general hate stimuli in this same Experiment 2 dataset could potentially be related to their rating
criticisms of systemic racism as less offensive than their straight counterparts. It would follow an
internal logic for participants who rated criticisms of systemic racism as less offensive to also
have significantly greater magnitudes of responses to the general hate stimuli, which contained
large amounts of racism. Further research should consider investigating both whether these
results can be replicated in other queer participants, but also whether the source of these
differences in responses is in fact due to more Liberal ideologies, regardless of sexual
orientation.
Further supporting the idea that ideology, not intergroup empathy, mediates rating
responses, our results demonstrate that ambivalent sexism, social dominance orientation, and
belief in a just world measurements were predictors for offensiveness/upsettingness ratings of
criticisms of systemic racism. These results support existing literature regarding the relationships
between social dominance orientation, belief in a just world, and perceptions of “reverse racism,”
as a concept. Measures of sexism, whether hostile, benign, or ambivalent, however, do not
appear to have been previously directly investigated in relation to attitudes surrounding “reverse
racism,” so this constitutes what appears to be a novel finding.
“Status legitimizing beliefs” refer to beliefs that support the idea that power structures in
society are good, and just. Social dominance orientation and belief in a just world are both status
legitimizing beliefs (Wilkins et al 2013, Bahamondes 2021). Status legitimizing beliefs were

184
found to have an inverse relationship with how negatively people claiming to be the victims of
anti-white bias were perceived (Wilkins et al 2013). Different aspects of status legitimizing
beliefs, especially relating to social dominance orientation, have been shown to have a
relationship to both how majority and minority group members perceive themselves as victims of
systemic discrimination. White people with high scores in the aspects of social dominance
related to endorsement of group-based dominance, or the idea that some groups are better than
others, were found to more strongly endorse the idea that they had been victims of anti-white
discrimination. Conversely, ethnic minorities who scored high on measures of the aspects of
social dominance orientation that measure system justification, or the idea that social hierarchies
are justified, were found to be less likely to perceive themselves as victims of racist
discrimination (Bahamondes 2021).
Thus, in parallel to what we found in regards to perceptions of hate speech, we conclude
that one’s beliefs are what predict response behaviors, not one’s demographic identity. Our
results, like Bahamondes 2021, demonstrate that being a racial minority is not a sure-fire
predictor that someone will not endorse or defend structures of systemic discrimination, even
when they themselves are the victims of that discrimination. Conversely, these results also
demonstrate that, at least among American college students, benefitting from these structures
does not necessarily predict endorsement of or investment in them. These experiments should be
repeated with a larger sample size, and with a wider age demographic, to determine if these
results are still true when a more age-diverse population is studied, and to identify potential
differences in response patterns between age cohorts to better understand the causes of any
differences.

185
Physiological Responses
The physiological responses to criticisms of systemic racism are somewhat difficult to
interpret, especially since there is only one dataset that has physiological measurements for this
category. In experiment 2, criticisms of systemic racism led to greater increases in heart rate
than the baseline/neutral stimuli, as did misogynist hate speech and threat stimuli. Both the
criticisms of systemic racism, and misogynist hate, stimuli also elicited a greater number of skin
conductance responses than the baseline stimuli in the psychometric baseline model. This model
had a larger n than the baseline model created for the principal components analyses. By
contrast, in the analyses of area under the curve, there were no significant differences between
the baseline stimuli and any other category of stimuli.
Understanding how these results fit among the outputs from the other experimental
datasets is somewhat challenging. Undoubtedly, these experiments should be repeated with a
larger dataset before any strong conclusions are made. However, our results do seem to reflect
that criticisms of systemic racism elicit physiological stress responses, while also raising more
questions about whether the types of stress responses elicited are the same, or different, between
criticisms of systemic racism and hate speech. Given that none of the stimulus categories
produced had significantly different areas under the curve than the baseline stimuli, it is possible
that all of these responses were more reflective of the sorts of subconscious threat responses than
of the emotionally-driven responses we observed in Experiment 1, and in the combined
Experiment 1+Experiment 2 dataset, especially when considering the misogynist hate stimuli. In
both the Experiment 1 and Experiment 1+2 datasets, the misogynist hate stimuli had
significantly larger areas under the curve than the baseline stimuli. It is thus somewhat puzzling

186
as to why that was not also observed in the Experiment 2 dataset, though perhaps a smaller n in
Experiment 2 is the reason behind these results.
Conversely, in Experiment 1, there were no observed differences in heart rate between
any of the stimulus categories, and in Experiment 1+2, there were only significant changes in
heart rate in response to the general hate stimuli. But for Experiment 2, threat stimuli, misogynist
hate stimuli, and criticisms of systemic racism ALL elicited greater changes in heart rate than the
baseline stimuli, indicating stress responses.
One large aspect that makes interpreting these results challenging is that heart rate can
both increase and decrease as parts of different response patterns (Petry & Desdirato 1978,
Wager et al 2009, Arza et al 2019, Eisenbarth et al 2016, Isenberg et al 1999, Porges et al 1969).
One of the limitations we discussed earlier in this dissertation describes how we hypothesized
that the lack of heart rate changes observed in Experiment 1 were due to participants completing
the experiment in a noisier, and potentially more distracting, environment, which could lead to
greater numbers of attention orientations throughout the experiment, and, in turn, a greater
number of heart rate decreases potentially “counteracting,” the increases in heart rate that the
variable stimuli may be eliciting. However, after examining how all three datasets’ results are
similar, and different, we wonder if perhaps there were, instead, significant differences in the
emotions participants in Experiment 1 vs Experiment 2 experienced in response to stimuli. We
will discuss this further in the next section of the dissertation.
Additional research is needed, but we hypothesize that the stress responses that were
observed via heart rate in response to both the criticisms of systemic racism and misogynist hate
stimuli may be related to the political nature of their content. While politics were not explicitly
addressed in these stimuli (which can be found in the Appendix of this dissertation), they do

187
address topics surrounding equality and civil rights, which are heavily enmeshed in the current
political landscape.
Discussing both political topics, as well as discrimination and systemic injustice can
often be inherently stressful for people (Lucas et al 2023, Peacock 2019, Carlson et al 2019). In
fact, Ford & Feinberg 2020 demonstrate that discussing political topics has become a widespread
source of chronic stress in the population, to the extent that it requires active emotional
regulation to cope with. In conjunction with the previous explanation, anticipation of a stressful
event or situation results in increased physiological arousal (Behnke & Sawyer 2009,
Schumacher et al 2015, Abel et al 1990, Poli et al 2007). It seems possible, then, that upon
reading statements discussing what is currently a divisive political issue, such as systemic
racism, or women’s rights, (Hawkins & Saleem 2022) participants may experience a sense of
anticipatory stress.
Exemplifying this, Carlson et al 2019 found that people who actively avoid exposure to
conflicting political opinions due to anxiety surrounding conflict specifically had an increase in
heart rate, but not skin conductance, when they knew a political discussion was about to take
place.
This could, perhaps, be applied to the results of Experiment 2. We know that, overall,
general hate stimuli were eliciting responses that required a larger n to be observed, because in
the combined Experiment1+2 dataset, the general hate stimuli resulted in changes in heart rate,
and the number and magnitude of skin conductance responses, despite not being observed in
either Experiment 1 or Experiment 2 individually. So, we are left with the question of what we
ARE observing in the Experiment 2 results. It seems possible that we are actually capturing two
distinct major response patterns, and that the majority of the responses in Experiment 2

188
potentially fall into this area of anticipatory stress. While Carlson et al 2019 may not have
observed changes in the number of skin conductance responses, and we did, this could have been
due to differences in protocols regarding data analyses, and the sensitivity of equipment used.
Though again, we cannot conclude this without further testing. But, given that the skin
conductance responses in Experiment 2 did not vary in magnitude actress categories, it seems
reasonable to hypothesize that these were minimal-to-subconscious responses, as we did earlier.
This would then align more closely with Carlson et al 2019’s results.
By contrast, the results obtained from Experiment 1, and observed in the combined
Experiment 1+Experiment 2 dataset, could reflect the presence of more emotionally driven
responses of being upset. Given that the general hate stimuli are the only stimuli to elicit
significantly greater increases, specifically, in heart rate, the number of skin conductance
responses, and the magnitude of skin conductance responses, it would seem like these stimuli are
eliciting responses as a result of being truly upset. The loss of Experiment 2’s observed increases
in heart rate for the misogynist hate and threat stimuli in the combined dataset could reflect that
these stimuli are more sharply salient than the general hate stimuli, and that the decrease in heart
rate as a result of attention orientation (Graham & Clifton 1966, Zimmer & Richter 2022) is
counteracting the stress-related increase in heart rate (Ducharme et al 2021).
Thus, we may be observing that the general hate stimuli elicit emotional distress, while
the misogynist hate stimuli elicit attention orientations and anticipation anxiety. This would also
align with the rating data that was constant through all of the experiments, demonstrating that the
participants consistently found the general hate stimuli to be more offensive/upsetting than the
misogynist hate stimuli, and potentially explain why the misogynist hate stimuli seemed to elicit
more physiological responses throughout the experiments. This would also suggest that it is

189
easier to capture the effects of saliency than emotional distress. Given that Isenberg et al 1999
demonstrated that the saliency of some stimuli can be observed without the participant even
being aware of their response, this does not seem like an impossible hypothesis to propose.
This, of course, leaves us with the question of how the responses to criticisms of systemic
racism fit into this paradigm. We seem to have determined that criticisms of systemic racism
elicit some type of stress-related response. Unfortunately, at this stage, we can only offer
speculations as to what type, since further replications would be necessary to provide more
conclusive answers. But, considering how closely the patterns of physiological responses of the
misogynist hate stimuli and the criticisms of systemic racism pair with each other in Experiment
2, we hypothesize that these two types of stimuli are eliciting the same type of response. Within
the hypothetical paradigm we have created, then, the criticisms of systemic racism would be
eliciting anticipation anxiety, and/or salience-related arousal.
While the results of the investigations related to the criticisms of systemic racism are
largely inconclusive, they provide us with several interesting avenues of future research, and
contribute novel insights to a largely understudied subcategory of hate speech and physiology.
Does Hate Speech Vary Emotional Physiology Pattern?
Building on our discussion of the criticisms of systemic racism, we found that for the
combined Experiment 1 + Experiment 2 dataset, there were observable differences in the
distributions of the emotional response types represented by our approach-avoidance x positivenegative valence paradigm. Notably, we found that the general hate stimuli elicited the most
anger responses, while the misogynist hate stimuli elicited the most attention orientations of any
of the stimulus categories, and more threat/fear responses than the general hate category. The
threat stimuli and the misogynist hate stimuli also appear to have elicited equal amounts of

190
threat/fear responses. Taken together, these results both seem to support the paradigm discussed
in the last section, where we propose that general hate stimuli elicit more emotionally upset
responses, and misogynist hate stimuli elicit more salience related responses. These trends in
emotional response patterns also provide evidence in support of hate speech eliciting more
negatively valent emotions than non-hate speech. We speculate that it may be possible, even, that
the threat/fear responses observed in response to the threat stimuli are the result of subconscious
amygdala responses to certain words, that the threat/fear responses exhibited in response to the
misogynist hate stimuli represent fear/threat related to anticipation anxiety.
However, this approach-avoidance paradigm is not built for distinguishing between
emotions that would fall within the same “quadrant,” of approach-avoidance responses, so this
would need to be investigated further with a more purpose-built experimental design. Our results
do seem to provide solid evidence that such a series of experiments would be well-motivated by
these data.
Providing further evidence that our paradigm did achieve a degree of accuracy, we also
found that the neutral/baseline stimuli elicited the fewest anger and threat responses, and the
most positive (positive approach, and positively valenced) responses of any of the stimulus
categories.
However, it should also be noted that these types of emotion detection models do not
function at 100% accuracy. These types of models are used to identify trends in data, not to be
able to determine definitively what the response type of a given individual datapoint is (Dechert
et al 2005, Giakoumis et al 2010, Vachiratamporn et al 2015). Our results are no exception. For
all categories of stimuli, the negatively valent responses- anger and threat- were by far the most
frequently elicited response patterns. This may have been due to the length of inter-stimulus

191
intervals. Drosschot & Thayer 2003 found that after experiencing negatively valent emotions,
heart rates take longer to return to baseline than after experiencing positively valent emotions.
These experiments should be repeated, then, with greater interstimulus intervals, in order to
determine if some of the (likely) misattributions of anger and threat responses were due to
carryover effects between stimuli. However, this does not automatically invalidate the accuracy
of the trends observed- this is the level at which it is most appropriate to consider these data.
Thus, while our models do provide legitimate evidence that the stimuli are actually
eliciting some of the predicted emotional responses, we cannot conclusively say that hate speech
elicits more negative emotions and attention orienting responses, as this work should be repeated
with a much larger n, and longer interstimulus intervals.

192
Conclusions
In these experiments, we expanded our understanding of how people respond to hate
speech, and criticisms of systemic racism, what types of responses were elicited, and what traits
drive these responses.
We demonstrated that hate speech is more upsetting to observers than neutral speech.
Additionally, hate speech elicited more, and larger, physiological stress responses than both
neutral speech and statements that contain words that elicit subconscious threat responses.
Psychometric traits, especially social dominance orientation and ambivalent sexism, were shown
to be strong predictors of rating behaviors. We also found two distinct trait “profiles,” consisting
of combinations of endorsements of social dominance orientation, belief in a just world,
ambivalent sexism, mental health, and grandiose narcissism in different proportions, which
predicted rating responses. Endorsement of psychometric traits, and/or adherence to these
profiles, however, had no impact on physiological responses.
We found that, in terms of both offensiveness/upsettingness ratings and physiological
responses, self-relevance did not predict rating behavior, nor did demographics, largely.
Demographics may predict endorsement of traits associated with authoritarian world views, but
it is the endorsement of those views that predicts actual ratings. Self-relevance not being a good
predictor for ratings may seem counterintuitive at first, but makes more sense when we consider
that people can, and frequently do, endorse ideologies that marginalize themselves. Women can
be sexist, non-white people can support systemic racism, etc. This highlights, however, that
future discussions and research surrounding hate speech, prejudice, and their endorsers, should
not begin by considering demographic factors, as many often do, but by considering
psychometric variables, and their effects in different cultural contexts.

193
We found that, for all people, exposure to hate speech elicited more, and larger, physical
stress responses, than non-hate speech, regardless of demographics or beliefs. This has potential
implications for hate speech being a source of chronic stress in all observers, and thus a chronic
public health hazard.
We also provided preliminary evidence that exposure to hate speech may elicit not just
non-specific stress responses, but negatively valent emotions such as anger and feelings of threat.
More research needs to be done, but our results also suggest that the misogynist hate and the
general hate elicit different types of responses, with the misogynist hate stimuli eliciting more
threat and saliency related responses, potentially as a result of anticipation anxiety, while general
hate stimuli elicited more emotionally derived responses, especially with regards to anger. This
coincides with observed differences in other response data, where the general hate stimuli were
consistently rated as being more offensive/upsetting than the misogynist hate stimuli, while
misogynist hate frequently elicited more physiological responses.
We also found that criticisms of systemic racism elicited more physiological responses
than the baseline stimuli. It is difficult to elaborate upon this finding without repetitions of the
study with a larger n, but it does appear that these stimuli elicit the same types of responses that
the misogynist hate stimuli do, relating to threat, salience, and potentially anticipation anxiety.
The criticisms of systemic racism stimuli were not rated as being as offensive/upsetting as hate
speech, and ratings of these stimuli were also predicted by the same psychometric variables that
predicted ratings of hate speech stimuli, albeit in the opposite direction. Social dominance
orientation and ambivalent sexism had a strong inverse relationship with
offensiveness/upsettingness ratings of hate speech, and a strong direct relationship with
offensiveness/upsettingness ratings of criticisms of systemic racism.

194
Thus, it appears that responses to criticisms of systemic racism share similarities with responses
to hate speech, but should still be considered distinct from them, until further research can
investigate the differences between the two more specifically.

195
Limitations
These studies did have a few notable limitations, which should be addressed with further
testing. First, the n for experiments that collected physiological data were somewhat small, due
to the increased time and effort it takes to collect physiological data. Thus, further testing to
replicate these results should be conducted. Second, when data was being collected for
Experiment 1, the room that was available for use was located in a high traffic hallway, while
data for Experiment 2 was collected in a quieter area. As a result, differences between heart rate
responses between Experiment 1 and Experiment 2 may have been due to attention orienting
responses to noise from the hallway, and further investigations should be done to determine if
this was the cause, or if another underlying variable was responsible. Third, due to budget, time,
and labor constraints, participants for these studies were all university students. This greatly
limited the diversity of participant ages and educational backgrounds. Further testing should be
conducted to determine if these results are replicated in study populations outside of college
students.

196
References
Abel, J. L., & Larkin, K. T. (1990). Anticipation of performance among musicians: Physiological
arousal, confidence, and state-anxiety. Psychology of music, 18(2), 171-182.
Adams, J. M., Florell, D., Burton, K. A., & Hart, W. (2014). Why do narcissists disregard socialetiquette norms? A test of two explanations for why narcissism relates to offensive-language use.
Personality and Individual Differences, 58, 26-30.
Allport, G. W., Clark, K., & Pettigrew, T. (1954). The nature of prejudice.
Anttonen, J., & Surakka, V. (2005, April). Emotions and heart rate while sitting on a chair. In
Proceedings of the SIGCHI conference on Human factors in computing systems (pp. 491-499).
Ataee, P., Hahn, J. O., Dumont, G. A., & Boyce, W. T. (2014). Noninvasive subject-specific
monitoring of autonomic-cardiac regulation. IEEE Transactions on Biomedical Engineering,
61(4), 1196-1207.
Bach, D. R., Friston, K. J., & Dolan, R. J. (2010). Analytic measures for quantification of arousal
from spontaneous skin conductance fluctuations. International journal of psychophysiology,
76(1), 52-55.
Bahamondes, J., Sibley, C. G., & Osborne, D. (2022). System justification or social dominance?
A multilevel test of the ideological motivators of perceived discrimination. Personality and
Social Psychology Bulletin, 48(7), 1134-1148.
Ballinger, T. (2021). Zero-Sum Beliefs Underlie Whites’ Perceptions of “Reverse
Discrimination”. The Ohio State University.
Behnke, R. R., & Sawyer, C. R. (2001). Public speaking arousal as a function of anticipatory
activation and autonomic reactivity. Communication Reports, 14(2), 73-85.
Beissner, F., Meissner, K., Bär, K. J., & Napadow, V. (2013). The autonomic brain: an activation
likelihood estimation meta-analysis for central processing of autonomic function. Journal of
neuroscience, 33(25), 10503-10511.
Ben‐Shakhar, G. (1985). Standardization within individuals: A simple method to neutralize
individual differences in skin conductance. Psychophysiology, 22(3), 292-299.
Betancourt, M. A., Dethorne, L. S., Karahalios, K., & Kim, J. G. (2017). Skin conductance as an
in situ marker for emotional arousal in children with neurodevelopmental communication
impairments: Methodological considerations and clinical implications. ACM Transactions on
Accessible Computing (TACCESS), 9(3), 1-29.
Bică, A. (2023). Peer pressure: Conformity outweighs reciprocity in social anxiety. Current
Psychology, 42(21), 18142-18149.

197
Brosschot, J. F., & Thayer, J. F. (2003). Heart rate response is longer after negative emotions
than after positive emotions. International journal of psychophysiology, 50(3), 181-187.
Bradley, M. M., Sambuco, N., & Lang, P. J. (2022). Affective perception: The power is in the
picture. Human perception of visual information: psychological and computational perspectives,
59-83.
Braithwaite, J. J., Watson, D. G., Jones, R., & Rowe, M. (2013). A guide for analysing
electrodermal activity (EDA) & skin conductance responses (SCRs) for psychological
experiments. Psychophysiology, 49(1), 1017-1034.
Brand, N., Hanson, E., & Godaert, G. (2000). Chronic stress affects blood pressure and speed of
short-term memory. Perceptual and Motor Skills, 91(1), 291-298.
Brown, C. L., Van Doren, N., Ford, B. Q., Mauss, I. B., Sze, J. W., & Levenson, R. W. (2020).
Coherence between subjective experience and physiology in emotion: Individual differences and
implications for well-being. Emotion, 20(5), 818.
Calvo, R., Nourbakhsh, N., Chen, F., & Wang, F. (2017). Detecting Users' Cognitive Load by
Galvanic Skin Response with Affective Interference.
Cameron, C. D., & Payne, B. K. (2011). Escaping affect: how motivated emotion regulation
creates insensitivity to mass suffering. Journal of personality and social psychology, 100(1), 1.
Carlson, T. N., McClean, C. T., & Settle, J. E. (2020). Follow your heart: Could
psychophysiology be associated with political discussion network homogeneity?. Political
Psychology, 41(1), 165-187.
Carlton, S., Harrison, A., Honoré, S., & Goodmon, L. B. (2020). Conceal, don't feel: Gender
differences in implicit and explicit expressions of emotions. Modern Psychological Studies,
25(1), 10.
Carnagey, N. L., Anderson, C. A., & Bushman, B. J. (2007). The effect of video game violence
on physiological desensitization to real-life violence. Journal of experimental social psychology,
43(3), 489-496.
Carver, C. S., & Harmon-Jones, E. (2009). Anger is an approach-related affect: evidence and
implications. Psychological bulletin, 135(2), 183.
Castellanos, M., Wettstein, A., Wachs, S., & Bilz, L. (2024). Direct and indirect effects of social
dominance orientation on hate speech perpetration via empathy and moral disengagement among
adolescents: A multilevel mediation model. Aggressive Behavior, 50(1), e22100.
Chalmers, J. A., Quintana, D. S., Abbott, M. J. A., & Kemp, A. H. (2014). Anxiety disorders are
associated with reduced heart rate variability: a meta-analysis. Frontiers in psychiatry, 5, 80.

198
Chen, C. M., & Wu, C. H. (2015). Effects of different video lecture types on sustained attention,
emotion, cognitive load, and learning performance. Computers & Education, 80, 108-121.
Christopher, A. N., Zabel, K. L., & Miller, D. E. (2013). Personality, authoritarianism, social
dominance, and ambivalent sexism: A mediational model. Individual Differences Research,
11(2).
Cichocka, A. (2016). Understanding defensive and secure in-group positivity: The role of
collective narcissism. European Review of Social Psychology, 27(1), 283-317.
Cichocka, A., & Cislak, A. (2020). Nationalism as collective narcissism. Current Opinion in
Behavioral Sciences, 34, 69-74.
Cicourel, A. V. (1982). Interviews, surveys, and the problem of ecological validity. The
American Sociologist, 11-20.
Clementson, D. E. (2018). Truth bias and partisan bias in political deception detection. Journal
of Language and Social Psychology, 37(4), 407-430.
Collet, C., Vernet-Maury, E., Delhomme, G., & Dittmar, A. (1997). Autonomic nervous system
response patterns specificity to basic emotions. Journal of the autonomic nervous system, 62(1-
2), 45-57.
Cowan, G., & Hodge, C. (1996). Judgments of hate speech: The effects of target group,
publicness, and behavioral responses of the target. Journal of Applied Social Psychology, 26(4),
355-374.
Cowan, G., & Khatchadourian, D. (2003). Empathy, ways of knowing, and interdependence as
mediators of gender differences in attitudes toward hate speech and freedom of speech.
Psychology of women quarterly, 27(4), 300-308.
Critchley, H. D. (2009). Psychophysiology of neural, cognitive and affective integration: fMRI
and autonomic indicants. International journal of psychophysiology, 73(2), 88-94.
Critchley, H. D. (2002). Electrodermal responses: what happens in the brain. The Neuroscientist,
8(2), 132-142.
Cuthbert, B. N., Schupp, H. T., Bradley, M. M., Birbaumer, N., & Lang, P. J. (2000). Brain
potentials in affective picture processing: covariation with autonomic arousal and affective
report. Biological psychology, 52(2), 95-111.
Dalbert, C. (2013). The justice motive as a personal resource: Dealing with challenges and
critical life events. Springer Science & Business Media.
Dambrun, M., Duarte, S., & Guimond, S. (2004). Why are men more likely to support group‐
based dominance than women? The mediating role of gender identification. British journal of
social psychology, 43(2), 287-297.

199
De Zavala, A. G., Cichocka, A., Eidelson, R., & Jayawickreme, N. (2009). Collective narcissism
and its social consequences. Journal of personality and social psychology, 97(6), 1074.
De Zavala, A. G., & Bierwiaczonek, K. (2021). Male, national, and religious collective
narcissism predict sexism. Sex Roles, 84(11), 680-700.
Dechert, N. T., Flack Jr, W., & Craig Jr, F. (2005). Patterns of cardiovascular responses during
angry, sad, and happy emotional recall tasks. Cognition and Emotion, 19(6), 941-951.
Deits-Lebehn, C., Smith, T. W., Williams, P. G., & Uchino, B. N. (2023). Heart rate variability
during social interaction: Effects of valence and emotion regulation. International Journal of
Psychophysiology, 190, 20-29.
Dharmapala, D., Garoupa, N., & McAdams, R. H. (2009). Belief in a just world, blaming the
victim, and hate crime statutes. Review of Law & Economics, 5(1), 311-345.
Döring, N., & Mohseni, M. R. (2020). Gendered hate speech in YouTube and YouNow
comments: Results of two content analyses. SCM Studies in Communication and Media, 9(1),
62-88.
Ducharme, P., Kahn, J., Vaudreuil, C., Gusman, M., Waber, D., Ross, A., ... & GonzalezHeydrich, J. (2021). A “proof of concept” randomized controlled trial of a video game requiring
emotional regulation to augment anger control training. Frontiers in Psychiatry, 12, 591906.
Duriez, B., Klimstra, T. A., Luyckx, K., Beyers, W., & Soenens, B. (2012). Right–wing
authoritarianism: Protective factor against or risk factor for depression?. European Journal of
Personality, 26(5), 536-549.
Dvash, J., & Shamay-Tsoory, S. G. (2014). Theory of mind and empathy as multidimensional
constructs: Neurological foundations. Topics in Language Disorders, 34(4), 282-295.
Eisenbarth, H., Chang, L. J., & Wager, T. D. (2016). Multivariate brain prediction of heart rate
and skin conductance responses to social threat. Journal of Neuroscience, 36(47), 11987-11998.
Elliot, A. J., Eder, A. B., & Harmon-Jones, E. (2013). Approach–avoidance motivation and
emotion: Convergence and divergence. Emotion Review, 5(3), 308-311.
Ekman, P., Levenson, R. W., & Friesen, W. V. (1983). Autonomic nervous system activity
distinguishes among emotions. Science, 221(4616), 1208-1210.
Feather, N. T., & McKee, I. R. (2012). Values, right‐wing authoritarianism, social dominance
orientation, and ambivalent attitudes toward women. Journal of Applied Social Psychology,
42(10), 2479-2504.
Ford, B. Q., & Feinberg, M. (2020). Coping with politics: The benefits and costs of emotion
regulation. Current Opinion in Behavioral Sciences, 34, 123-128.

200
Foster, J. D., McCain, J. L., Hibberts, M. F., Brunell, A. B., & Johnson, R. B. (2015). The
Grandiose Narcissism Scale: A global and facet-level measure of grandiose narcissism.
Personality and individual differences, 73, 12-16.
Fowles, D. C. (1980). The three arousal model: Implications of Gray's two‐factor learning theory
for heart rate, electrodermal activity, and psychopathy. Psychophysiology, 17(2), 87-104.
Funk, J. B., Baldacci, H. B., Pasold, T., & Baumgardner, J. (2004). Violence exposure in reallife, video games, television, movies, and the internet: is there desensitization?. Journal of
adolescence, 27(1), 23-39.
Ghiasi, S., Greco, A., Barbieri, R., Scilingo, E. P., & Valenza, G. (2020). Assessing autonomic
function from electrodermal activity and heart rate variability during cold-pressor test and
emotional challenge. Scientific reports, 10(1), 5406.
Giakoumis, D., Vogiannou, A., Kosunen, I., Moustakas, K., Tzovaras, D., & Hassapis, G. (2010,
January). Identifying Psychophysiological Correlates of Boredom and Negative Mood Induced
during HCI. In B-Interface (pp. 3-12).
Golebiowska, E. A. (2019). Sexual orientation, gender identity, and political decision making. In
Oxford research encyclopedia of politics.
Gordon, S. L. (2022). Can social dominance orientation predict intergroup hostility and
aggression in South Africa? Testing the applicability of the SDO7 (s) Scale. South African
Journal of Psychology, 52(1), 23-35.
Gorman, J. M., & Sloan, R. P. (2000). Heart rate variability in depressive and anxiety disorders.
American heart journal, 140(4), S77-S83.
Graham, F. K., & Clifton, R. K. (1966). Heart-rate change as a component of the orienting
response. Psychological bulletin, 65(5), 305.
Gray, AL (2004). Parasympathetic control of the heart II: A novel intergangionic intrinsic
cardiac circuit mediates neural control of heart rate. J Appl Physiol., 96, 2273-2278.
Greaves-Lord, K., Ferdinand, R. F., Sondeijker, F. E., Dietrich, A., Oldehinkel, A. J., Rosmalen,
J. G., ... & Verhulst, F. C. (2007). Testing the tripartite model in young adolescents: is
hyperarousal specific for anxiety and not depression?. Journal of Affective Disorders, 102(1-3),
55-63.
Green, J. H., & Green, J. H. (1990). Physiology of the autonomic nervous system. The
Autonomic Nervous System and Exercise, 32-55.
Grigoropoulos, I. (2024). Relations Between Authoritarianism, Just World Beliefs, and the Dark
Triad Life Philosophy. Interpersona: An International Journal on Personal Relationships, 18(1),
104-123.

201
Gruss, L. F., & Keil, A. (2019). Sympathetic responding to unconditioned stimuli predicts
subsequent threat expectancy, orienting, and visuocortical bias in human aversive Pavlovian
conditioning. Biological psychology, 140, 64-74.
Hawkins, I., & Saleem, M. (2022). How social media use, political identity, and racial
resentment affect perceptions of reverse racism in the United States. Computers in Human
Behavior, 134, 107337.
Hangartner, D., Gennaro, G., Alasiri, S., Bahrich, N., Bornhoft, A., Boucher, J., ... & Donnay, K.
(2021). Empathy-based counterspeech can reduce racist hate speech in a social media field
experiment. Proceedings of the National Academy of Sciences, 118(50), e2116310118.
Heerdink, M. W., Van Kleef, G. A., Homan, A. C., & Fischer, A. H. (2013). On the social
influence of emotions in groups: interpersonal effects of anger and happiness on conformity
versus deviance. Journal of Personality and Social Psychology, 105(2), 262.
Himichi, T., & Ohtsubo, Y. (2020). An inverted U-shaped relationship between heart rate
deceleration and empathic emotions. Biological Psychology, 150, 107828.
Hjortskov, N., Rissén, D., Blangsted, A. K., Fallentin, N., Lundberg, U., & Søgaard, K. (2004).
The effect of mental stress on heart rate variability and blood pressure during computer work.
European journal of applied physiology, 92, 84-89.
Isenberg, N., Silbersweig, D., Engelien, A., Emmerich, S., Malavade, K., Beattie, B., ... & Stern,
E. (1999). Linguistic threat activates the human amygdala. Proceedings of the National Academy
of Sciences, 96(18), 10456-10459.
Iversen, N. K., Dupont-Prinet, A., Findorf, I., McKenzie, D. J., & Wang, T. (2010). Autonomic
regulation of the heart during digestion and aerobic swimming in the European sea bass
(Dicentrarchus labrax). Comparative Biochemistry and Physiology Part A: Molecular &
Integrative Physiology, 156(4), 463-468.
Kennedy, B., Atari, M., Davani, A. M., Yeh, L., Omrani, A., Kim, Y., ... & Dehghani, M. (2018).
The gab hate corpus: A collection of 27k posts annotated for hate speech. PsyArXiv. July, 18.
Kennedy, B., Atari, M., Davani, A. M., Yeh, L., Omrani, A., Kim, Y., ... & Dehghani, M. (2022).
Introducing the Gab Hate Corpus: defining and applying hate-based rhetoric to social media
posts at scale. Language Resources and Evaluation, 1-30.
Kinnick, K. N., Krugman, D. M., & Cameron, G. T. (1996). Compassion fatigue:
Communication and burnout toward social problems. Journalism & Mass Communication
Quarterly, 73(3), 687-707.
Lane, R. D., Weidenbacher, H., Smith, R., Fort, C., Thayer, J. F., & Allen, J. J. (2013).
Subgenual anterior cingulate cortex activity covariation with cardiac vagal control is altered in
depression. Journal of Affective Disorders, 150(2), 565-570.

202
Lang, P. J., Levin, D. N., Miller, G. A., & Kozak, M. J. (1983). Fear behavior, fear imagery, and
the psychophysiology of emotion: the problem of affective response integration. Journal of
abnormal psychology, 92(3), 276.
Lang, P. J., Bradley, M. M., & Cuthbert, B. N. (1988). International affective picture system.
Lazarus, R. S., Speisman, J. C., & Mordkoff, A. M. (1963). The relationship between autonomic
indicators of psychological stress: Heart rate and skin conductance. Psychosomatic medicine,
25(1), 19-30.
Lee, S., Hwang, H. B., Park, S., Kim, S., Ha, J. H., Jang, Y., ... & Kim, I. Y. (2022). Mental
stress assessment using ultra short term HRV analysis based on non-linear method. Biosensors,
12(7), 465.
Leets, L., Giles, H., & Noels, K. (1999). Attributing harm to racist speech. Journal of
Multilingual and Multicultural Development, 20(3), 209-215.
Leets, L. (2002). Experiencing hate speech: Perceptions and responses to anti‐semitism and
antigay speech. Journal of social issues, 58(2), 341-361.
LeMaire, K. L. (2014). Who confronts sexual prejudice? How gender and idelogies are related
to heterosexual allies challenging hate speech (Doctoral thesis, Marquette University).
Lipkus, I. (1991). The construction and preliminary validation of a global belief in a just world
scale and the exploratory analysis of the multidimensional belief in a just world scale.
Personality and Individual differences, 12(11), 1171-1178.
Liu, K., Jiao, Y., Du, C., Zhang, X., Chen, X., Xu, F., & Jiang, C. (2023). Driver stress detection
using ultra-short-term HRV analysis under real world driving conditions. Entropy, 25(2), 194.
Lo Cricchio, M. G., Stefanelli, F., Menesini, E., & Nocentini, A. (2023). Adolescents' Ethnic
Hate Speech Exposure and Ethnic Bullying Perpetration: The Moderating Role of Tolerance
Towards Diversity and Gender. Cyberpsychology, 17(5).
Lobel, A., Gotsis, M., Reynolds, E., Annetta, M., Engels, R. C., & Granic, I. (2016, May).
Designing and utilizing biofeedback games for emotion regulation: The case of nevermind. In
Proceedings of the 2016 CHI Conference Extended Abstracts on Human Factors in Computing
Systems (pp. 1945-1951).
Lohani, M., Payne, B. R., & Strayer, D. L. (2019). A review of psychophysiological measures to
assess cognitive states in real-world driving. Frontiers in human neuroscience, 13, 57.
Lucas, T., Yamin, J. B., Krohner, S., Goetz, S. M., Kopetz, C., & Lumley, M. A. (2023). Writing
about justice and injustice: Complex effects on affect, performance, threat, and biological
responses to acute social stress among African American women and men. Social Science &
Medicine, 316, 115019.

203
Mai, X., Zhang, W., Hu, X., Zhen, Z., Xu, Z., Zhang, J., & Liu, C. (2016). Using tDCS to
explore the role of the right temporo-parietal junction in theory of mind and cognitive empathy.
Frontiers in psychology, 7, 380.
Masaoka, Y., Hirasawa, K., Yamane, F., Hori, T., & Homma, I. (2003). Effects of left amygdala
lesions on respiration, skin conductance, heart rate, anxiety, and activity of the right amygdala
during anticipation of negative stimulus. Behavior modification, 27(5), 607-619.
Mayer, S. J., Berning, C. C., & Johann, D. (2020). The two dimensions of narcissistic personality
and support for the radical right: the role of right–wing authoritarianism, social dominance
orientation and anti–immigrant sentiment. European Journal of Personality, 34(1), 60-76.
McDuff, D., Gontarek, S., & Picard, R. (2014, August). Remote measurement of cognitive stress
via heart rate variability. In 2014 36th annual international conference of the IEEE engineering
in medicine and biology society (pp. 2957-2960). IEEE.
Mehler, B., Reimer, B., & Coughlin, J. F. (2012). Sensitivity of physiological measures for
detecting systematic variations in cognitive demand from a working memory task: an on-road
study across three age groups. Human factors, 54(3), 396-412.
Moriguchi, A., Otsuka, A., Kohara, K., Mikami, H., Katahira, K., Tsunetoshi, T., ... & Ogihara,
T. (1992). Spectral change in heart rate variability in response to mental arithmetic before and
after the beta-adrenoceptor blocker, carteolol. Clinical Autonomic Research, 2, 267-270.
Moses, Z. B., Luecken, L. J., & Eason, J. C. (2007, August). Measuring task-related changes in
heart rate variability. In 2007 29th annual international conference of the IEEE engineering in
medicine and biology society (pp. 644-647). IEEE.
Moya-Garófano, A., Moya, M., Megías, J. L., & Rodríguez-Bailón, R. (2022). Ambivalent
sexism and women’s reactions to stranger harassment: The case of piropos in Spain. Psychology
of Women Quarterly, 46(4), 454-467.
Murray, R. (2011). Do words harm? The perceptions and attitudes of african American college
students to hate speech (Doctoral dissertation, Capella University).
Nagai, Y., Critchley, H. D., Featherstone, E., Trimble, M. R., & Dolan, R. J. (2004). Activity in
ventromedial prefrontal cortex covaries with sympathetic skin conductance level: a physiological
account of a “default mode” of brain function. Neuroimage, 22(1), 243-251.
Nelson, J. K., Hynes, M., Sharpe, S., Paradies, Y., & Dunn, K. (2018). Witnessing anti-white
‘racism’: White victimhood and ‘reverse racism’ in Australia. Journal of Intercultural Studies,
39(3), 339-358.
Norton, M. I., & Sommers, S. R. (2011). Whites see racism as a zero-sum game that they are
now losing. Perspectives on Psychological science, 6(3), 215-218.

204
Nourbakhsh, N., Wang, Y., Chen, F., & Calvo, R. A. (2012, November). Using galvanic skin
response for cognitive load measurement in arithmetic and reading tasks. In Proceedings of the
24th Australian computer-human interaction conference (pp. 420-423).
Nurik, C. (2019). “Men are scum”: Self-regulation, hate speech, and gender-based censorship on
Facebook. International Journal of Communication, 13, 21.
O'Connor, W. E., Morrison, T. G., McLeod, L. D., & Anderson, D. (1996). A meta-analytic
review of the relationship between gender and belief in a just world. Journal of Social Behavior
and Personality, 11(1), 141-148.
Peacock, C. (2019). (Not) talking politics: Motivations and strategies for avoiding the expression
of political opinions. Western Journal of Communication, 83(5), 581-599.
Pedersen, S. H., & Strömwall, L. A. (2013). Victim blame, sexism and just-world beliefs: A
cross-cultural comparison. Psychiatry, Psychology and Law, 20(6), 932-941.
Peng, S., Kuang, B., & Hu, P. (2020). Right temporoparietal junction modulates in-group bias in
facial emotional mimicry: A tDCS study. Frontiers in Behavioral Neuroscience, 14, 143.
Petry, H. M., & Desiderato, O. (1978). Changes in heart rate, muscle activity, and anxiety level
following shock threat. Psychophysiology, 15(5), 398-402.
Pittig, A., Arch, J. J., Lam, C. W., & Craske, M. G. (2013). Heart rate and heart rate variability in
panic, social anxiety, obsessive–compulsive, and generalized anxiety disorders at baseline and in
response to relaxation and hyperventilation. International journal of psychophysiology, 87(1),
19-27.
Pluta, A., Mazurek, J., Wojciechowski, J., Wolak, T., Soral, W., & Bilewicz, M. (2023).
Exposure to hate speech deteriorates neurocognitive mechanisms of the ability to understand
others’ pain. Scientific Reports, 13(1), 4127.
Poli, S., Sarlo, M., Bortoletto, M., Buodo, G., & Palomba, D. (2007). Stimulus-preceding
negativity and heart rate changes in anticipation of affective pictures. International Journal of
Psychophysiology, 65(1), 32-39.
Porges, S. W. (2007). The polyvagal perspective. Biological psychology, 74(2), 116-143.
Porges, S. W., & Raskin, D. C. (1969). Respiratory and heart rate components of attention.
Journal of experimental psychology, 81(3), 497.
Porter, A. M., & Goolkasian, P. (2019). Video games and stress: How stress appraisals and game
content affect cardiovascular and emotion outcomes. Frontiers in psychology, 10, 967.
Potter, P., Deshields, T., Divanbeigi, J., Berger, J., Cipriano, D., Norris, L., & Olsen, S. (2010).
Compassion Fatigue and Burnout. Clinical journal of oncology nursing, 14(5).

205
Pratto, F., Sidanius, J., Stallworth, L. M., & Malle, B. F. (1994). Social dominance orientation: A
personality variable predicting social and political attitudes. Journal of personality and social
psychology, 67(4), 741.
Pratto, F., Stallworth, L. M., & Sidanius, J. (1997). The gender gap: Differences in political
attitudes and social dominance orientation. British journal of social psychology, 36(1), 49-68.
Ravaja, N., Saari, T., Salminen, M., Laarni, J., & Kallinen, K. (2006). Phasic emotional reactions
to video game events: A psychophysiological investigation. Media psychology, 8(4), 343-367.
Richter, J., Pietzner, A., Koenig, J., Thayer, J. F., Pané-Farré, C. A., Gerlach, A. L., ... & Hamm,
A. O. (2021). Vagal control of the heart decreases during increasing imminence of interoceptive
threat in patients with panic disorder and agoraphobia. Scientific Reports, 11(1), 7960.
Robino, A. E. (2019). Global Compassion Fatigue: A New Perspective in Counselor Wellness.
Professional Counselor, 9(4), 272-284.
Roccato, M., & Ricolfi, L. (2005). On the correlation between right-wing authoritarianism and
social dominance orientation. Basic and applied social psychology, 27(3), 187-200.
Rollero, C., Bergagna, E., & Tartaglia, S. (2021). What is violence? The role of sexism and
social dominance orientation in recognizing violence against women. Journal of interpersonal
violence, 36(21-22), NP11349-NP11366.
Roussos, G., & Dovidio, J. F. (2018). Hate speech is in the eye of the beholder: The influence of
racial attitudes and freedom of speech beliefs on perceptions of racially motivated threats of
violence. Social psychological and personality science, 9(2), 176-185.
Ruiz Vargas, E., Sörös, P., Shoemaker, J. K., & Hachinski, V. (2016). Human cerebral circuitry
related to cardiac control: a neuroimaging meta‐analysis. Annals of Neurology, 79(5), 709-716.
Russell, B. L., & Trigg, K. Y. (2004). Tolerance of sexual harassment: An examination of gender
differences, ambivalent sexism, social dominance, and gender roles. Sex Roles, 50, 565-573.
Sanchez, D. T., Chaney, K. E., Manuel, S. K., Wilton, L. S., & Remedios, J. D. (2017). Stigma
by prejudice transfer: Racism threatens White women and sexism threatens men of color.
Psychological science, 28(4), 445-461.
Sakallı-Uğurlu, N., Yalçın, Z. S., & Glick, P. (2007). Ambivalent sexism, belief in a just world,
and empathy as predictors of Turkish students’ attitudes toward rape victims. Sex Roles, 57, 889-
895.
Sang, Q., Kang, Q., Zhang, K., Shu, S., & Quan, L. (2023). The Effect of Just-World Beliefs on
Cyberaggression: A Moderated Mediation Model. Behavioral Sciences, 13(6), 500.

206
Schaefer, M., Kühnel, A., Rumpel, F., & Gärtner, M. (2021). Do empathic individuals behave
more Prosocially? Neural correlates for altruistic behavior in the dictator game and the dark side
of empathy. Brain Sciences, 11(7), 863.
Scheff, T. J. (1988). Shame and conformity: The deference-emotion system. American
sociological review, 395-406.
Schmuckler, M. A. (2001). What is ecological validity? A dimensional analysis. Infancy, 2(4),
419-436.
Schrader, T., Jolley, D., Jolley, R. P., & Krahenbuhl, S. (2024). Upholding social hierarchies:
Social dominance orientation moderates the link between (intergroup) conspiracy exposure and
violent extremism. Group Processes & Intergroup Relations, 13684302241247985.
Schumacher, S., Herwig, U., Baur, V., Mueller-Pfeiffer, C., Martin-Soelch, C., Rufer, M., &
Brühl, A. B. (2015). Psychophysiological responses during the anticipation of emotional
pictures. Journal of Psychophysiology.
Setz, C., Arnrich, B., Schumm, J., La Marca, R., Tröster, G., & Ehlert, U. (2009). Discriminating
stress from cognitive load using a wearable EDA device. IEEE Transactions on information
technology in biomedicine, 14(2), 410-417.
Shaffer, F., & Ginsberg, J. P. (2017). An overview of heart rate variability metrics and norms.
Frontiers in public health, 5, 258.
Shehu, H. A., Oxner, M., Browne, W. N., & Eisenbarth, H. (2023). Prediction of moment‐by‐
moment heart rate and skin conductance changes in the context of varying emotional arousal.
Psychophysiology, 60(9), e14303.
Shi, H., Yang, L., Zhao, L., Su, Z., Mao, X., Zhang, L., & Liu, C. (2017). Differences of heart
rate variability between happiness and sadness emotion states: a pilot study. Journal of Medical
and Biological Engineering, 37, 527-539.
Sidanius, J., Pratto, F., & Mitchell, M. (1994). In-group identification, social dominance
orientation, and differential intergroup social allocation. The Journal of social psychology,
134(2), 151-167.
Sjouwerman, R., & Lonsdorf, T. B. (2019). Latency of skin conductance responses across
stimulus modalities. Psychophysiology, 56(4), e13307.
Smets, E., De Raedt, W., & Van Hoof, C. (2018). Into the wild: the challenges of physiological
stress detection in laboratory and ambulatory settings. IEEE journal of biomedical and health
informatics, 23(2), 463-473.
Society for Psychophysiological Research Ad Hoc Committee on Electrodermal Measures,
Boucsein, W., Fowles, D. C., Grimnes, S., Ben‐Shakhar, G., Roth, W. T., ... & Filion, D. L.
(2012). Publication recommendations for electrodermal measurements. Psychophysiology, 49(8),
1017-1034.

207
Sokolov, E.N. (1960). Neuronal models and the orienting reflex. The central nervous system and
behaviour.
Sokolov, E. N. (1963). Perception and the Conditioned Reflex. Pergamon Press. Original, 1958:
Vospriyatiye i Uslovnyi Refleks. Moscow University Press, Moscow.
Sokolov, E. N. (1975). The neuronal mechanisms of the orienting reflex. In E. N. Sokolov & O.
S. Vinogradova (Eds.), Neuronal Mechanisms of the Orienting Reflex (pp. 217–235). Lawrence
Erlbaum Associates.
Soral, W., Bilewicz, M., & Winiewski, M. (2018). Exposure to hate speech increases prejudice
through desensitization. Aggressive behavior, 44(2), 136-146.
Soral, W., Malinowska, K., & Bilewicz, M. (2022). The role of empathy in reducing hate speech
proliferation. Two contact-based interventions in online and off-line settings. Peace and
Conflict: Journal of Peace Psychology, 28(3), 361.
Sztajzel, J. (2004). Heart rate variability: a noninvasive electrocardiographic method to measure
the autonomic nervous system. Swiss medical weekly, 134(3536), 514-522.
Todorović, Z., Stojanović, K., Trajković1, T., & Grozdanović, A. (2020). BELIEF IN AN
UNJUST WORLD AND PERSONALITY TRAITS AS PREDICTORS OF SEXIST
ATTITUDES. International Thematic Proceedia, 93.
Vachiratamporn, V., Legaspi, R., Moriyama, K., Fukui, K. I., & Numao, M. (2015). An analysis
of player affect transitions in survival horror games. Journal on Multimodal User Interfaces, 9,
43-54.
Valor-Segura, I., Expósito, F., & Moya, M. (2011). Victim blaming and exoneration of the
perpetrator in domestic violence: The role of beliefs in a just world and ambivalent sexism. The
Spanish journal of psychology, 14(1), 195-206.
Van Hiel, A., & Kossowska, M. (2006). Having few positive emotions, or too many negative
feelings? Emotions as moderating variables of authoritarianism effects on racism. Personality
and Individual Differences, 40(5), 919-930.
Vanman, E. J., Dawson, M. E., & Brennan, P. A. (1998). Affective reactions in the blink of an
eye: Individual differences in subjective experience and physiological responses to emotional
stimuli. Personality and Social Psychology Bulletin, 24(9), 994-1005.
Voss, D., & Newman, L. S. (2021, June). Confronted with bullying when you believe in a just
world. In Frontiers in Education (Vol. 6, p. 634517). Frontiers Media SA.
Vrana, S. R., Cuthbert, B. N., & Lang, P. J. (1989). Processing fearful and neutral sentences:
Memory and heart rate change. Cognition & Emotion, 3(3), 179-195.

208
Wascher, C. A. (2021). Heart rate as a measure of emotional arousal in evolutionary biology.
Philosophical Transactions of the Royal Society B, 376(1831), 20200479.
Weathers, F. W., Litz, B. T., Herman, D. S., Huska, J. A., & Keane, T. M. (1993, October). The
PTSD Checklist (PCL): Reliability, validity, and diagnostic utility. In annual convention of the
international society for traumatic stress studies, San Antonio, TX (Vol. 462).
Weber, M., Viehmann, C., Ziegele, M., & Schemer, C. (2020). Online hate does not stay online–
How implicit and explicit attitudes mediate the effect of civil negativity and hate in user
comments on prosocial behavior. Computers in human behavior, 104, 106192.
Wilkins, C. L., Wellman, J. D., & Kaiser, C. R. (2013). Status legitimizing beliefs predict
positivity toward Whites who claim anti-White bias. Journal of Experimental Social Psychology,
49(6), 1114-1119.
Withers, K. L., Parrish, J. L., Terrell, S., & Ellis, T. J. (2017). The relationship between the “dark
triad” personality traits and deviant behavior on social networking sites.
Wood, K. H., Ver Hoef, L. W., & Knight, D. C. (2014). The amygdala mediates the emotional
modulation of threat-elicited skin conductance response. Emotion, 14(4), 693.
Wolpe, J. (1968). Behavior therapy in complex neurotic states. In Research in Psychotherapy
Conference, 3rd, May-Jun, 1966, Chicago, IL, US. American Psychological Association.
Wolpe, J. (1968). Psychotherapy by reciprocal inhibition. Conditional reflex: a Pavlovian
journal of research & therapy, 3, 234-240.
Wu, Y., Gu, R., Yang, Q., & Luo, Y. J. (2019). How do amusement, anger and fear influence
heart rate and heart rate variability?. Frontiers in neuroscience, 13, 490153.
Xu, X., Zuo, X., Wang, X., & Han, S. (2009). Do you feel my pain? Racial group membership
modulates empathic neural responses. Journal of Neuroscience, 29(26), 8525-8529.
Zannettou, S., Bradlyn, B., De Cristofaro, E., Kwak, H., Sirivianos, M., Stringini, G., &
Blackburn, J. (2018, April). What is gab: A bastion of free speech or an alt-right echo chamber.
In Companion Proceedings of the The Web Conference 2018 (pp. 1007-1014).
Zhu, J., Ji, L., & Liu, C. (2019). Heart rate variability monitoring for emotion and disorders of
emotion. Physiological measurement, 40(6), 064004.
Zimmer, H., & Richter, F. (2023). Novelty detection and orienting: effects on skin conductance
and heart rate. Psychological Research, 87(4), 1101-1113.

209
Appendix-All Stimuli Used
Included are all of the stimuli presented to participants throughout the 3 experiments.
Stimuli are arranged by category.
Content Warning: Many of the following stimuli contain graphic content surrounding
topics including sexual assault, physical assault, racism, homophobia, anti-semitism, antiMuslim sentiments, and sexism.
Misogynist Hate Stimuli:
The sexual revolution brought us- the Pill- the disconnect between Sex & Responsibility- a
rampant hookup and divorce culture- millions of aborted children- the single-mother home
sponsored by the government- removal of the importance of the male role in the family- women
that absolutely resent men- weakened, feminized men THIS. IS. NOT. PROGRESS.
The more the feminist screams for equality, the more her subconscious screams to be
dominated.
The problem is women say silly shit, and the meaning of the words is irrelevant to them. Words
are just a medium for manipulating feels and relationships. If you kill a male communist, great,
now there's one less communist. If a stupid cunt dies, have you accomplished anything? Mollie
should have been grounded by her patriarch.
going to rape the absolute fuck out of my stupid wife when I see her, bitch has no idea what's
coming to her (me)
Stefan Molyneux said a while back that the best possible scenario for the improvement of
gender relations is a massive civil war/economic collapse/catastophe. Make women realize how

210
full of shit they are and how helpless they are without Men/The State to save them. Then they'll
come running back with their tail between their legs begging to return to patriarchy and
traditional gender norms.
Well... at least a man• was still first. By the time all of this makes it to Globalist textbooks in
the future it will be a female astronaut who went to the moon just to abort her baby there
because she wanted to do it first... before using her size A bra to sling the spacecraft back.
Sure my slut daughter was raped and murdered but hey at least I get easy access to authentic
burritos! Checkmate NAHTZEES. @SCALE
#Feminism is an infection!
Tell this fat ugly ape that if she is getting threatened at her job, it's her employer's responsibility
to pay for her protection. And if they refuse, it just shows they don't care about her, just like the
rest of us don't.
Now: 'We need more women engineers!' 30 years from now: 'We need more work visas for
engineers because we can't find enough good domestic ones!'
International womens day today eh, hope the ladies are doing their bit to level up the 95% of
workplace deaths and 66% suicide rate. I worry for boys growing up now, told they are toxic,
brought up with few male role models/teachers and assumed to be potential rapists. Never mind,
we are disposable.
That's why you should never trust a FEMINAZI.
If you are right-wing and pay to send your daughter to college, you are retarded. Woman are
submissive and much more prone to peer pressure. They are not equipped (NAWALT) to

211
handle life in a re-education center. By sending her to college, you are destroying her future and
increasing the risk she'll die in childless misery.
Given the different birth rates based on education levels: Any women who has the brains and
talent to be a good scientist or doctor is too valuable as a mother to be wasted on science or
medicine. #Femininity #Mothers #FamilyValues #TraditionalValues #RandomThoughts
#PoliticallyIncorrect
Toxic masculinity is a made up term used to denigrate men. Some men are toxic; masculinity is
natural & healthy. Women murder millions of babies every year. Where are the toxic femininity
or toxic hypergamy labels? I refuse to be drawn into a war of words. I speak truth.
Male Feminists believe in trickle down pussy.
Because nothing says "respect me"• like running around dressed up as syphillis and chanting
about how cool abortions are. #WomensMarch2018
The latest trend in Feminism: Dyeing your butthole for equality. That's right, ladies, say no to
bleaching your butthole, that's just a simple way to say 'I'm white-washing my butthole!' Dye
your butthole a fun color for Feminism! If you aren't dyeing it, you're not a real Feminist, so get
out there and represent! #ButtRainbows
People with 'large secondary sexual characteristics”big muscles, large breasts' are 'in personality
terms...more animal-like.' This is the only way in which porn is an accurate representation of
real life.
I like that she talks about Great Comics and includes herself, 'our job'. That miserable twat
wouldn't know great comedy if it slapped her on the ass.

212
General Hate:
The fact is that these will never stop happening until autistics and mental retards are sterilized
and removed from the population and institutionalized. This kid snapped because he lost a
game. Only autistics behave this way.
David Hogg is a massive fucking faggot and I wish I could kick the shit out of his stupid pussy
face
Absolutely agree with everything...except perhaps compensating them. Muslims are raised to
have strong hatred for non-Muslims and seek to subjugate them and force them to live as
persecuted second class citizens - people as poisonous as this do not deserve the slightest bit of
recompense or civility. Burma has the right idea.
Turning the frogs #Gay or children #Trans all part of the #Marxist #Communist take over, turn
all white men into weak little twigs so they would have to become a bundle of faggots #Leftism
#Agenda
Listen, faggot, you're just a brainwashed lazy minded drone who has never looked into it. Tell
me: what proof is there that any sort of genocide happened against the Jews.
It's time to purge the entire landmass of the United States of the plague that is the Jewish
pedophile population They hate our right to bear arms more than SEX WITH CHILDREN I
wouldn't mind seeing these vermin Hunger Gamed. Put them in a pit, tell them to hunt each
other down
Holy shit, the man who shouted 'Shut the fuck up, you fucking loudmouthed monkey
motherfucker,I can't listen to your fucking black ass no more' to loudmouthed negress charged
with felony hate crime. Burn it all down, US is fucked beyond repair

213
What's the worst thing, beginning with N and ending in R, that you can call a black man?
Neighbour.
THE THIRD COLOSSUS The Lady of Liberty is not a French whore, We have endured
enough; we don't want any more. Don't give us your tired, your poor huddled masses, Your
refugee refuse of conflict and lack. They may be the finest of your foreign classes, But
nevertheless, they have to go back!
If most blacks want to laze around smoking weed and having babies with three different
daddies, drinking water from a river because they can't manage a municipal water system, I
Don't Care. It's no reason to hate them. Let them live in their own communities where they can
raise a ruckus without getting arrested.
Now that I think about it, these tranny freaks who pump their bodies full of hormones/chemicals
that destroy their actual, biological reproductive system... this is essentially eugenics.
@Cantwell What better way to eliminate degenerate DNA than to trick mentally insane
undesirable males to medically sterilize themselves via hormones for years, make it impossible
for them to reproduce.
Maybe some people think my language on this is crass, but it's just more political correctness.
It's yet another example of getting demonized because you notice patterns. Every mass shooter
of every major incident has been autistic or somewhere on the spectrum.
Irish people are scum
Once you allow Islam in it starts spreading fast like a disease.
She's not going to help you. You're white. She's black. She wants to see you dead.

214
all black women look like bull mastiffs
Is there anything whites AREN'T better at? Anything we take from blacks is improved by us.
I hope this county becomes 90% black and @pnehlen's wife gets gang raped by a bunch of
niggers.
Wow you're a fucking Clown. You must be a Jew
People who say 'nobody's born a racist, you have to be taught' are also the ones who tell their
children to be gay degenerates and killing white people is right.
Threat Word:
"This is part of a parable in which Jesus was telling a story intended to teach a lesson, not a
command by him to kill anybody. The direct words there are what Jesus was quoting a King to
have said within the story. This all becomes clear when you look at the entire passage in full. "
"Whatever your absurd dreams might be, this little girl's going to tear them down."
"For the enemy pursues my soul. He has struck my life down to the ground. He has made me
live in dark places, as those who have been long dead. Therefore my spirit is overwhelmed
within me. My heart within me is desolate. Ps 143:3-4 NHEB #Prayer"
In Thailand, I encountered a Spyderco knife on sale for a mere five hundred baht. But if it's too
cheap to be true, it always is.#Steemit
Freeway of Love! 130 pink Cadillacs owners come from all over the country to line Detroit's
streets as part of Aretha Franklin's funeral procession130 pink Cadillacs lined up in a procession
for the Queen of Soul's Friday funeral in tribute to her 1985 hit Freeway of Love

215
"So starting tomorrow, I'll have 11 days of peace and quiet in the house while the family goes
on vacation. Could use any and all cheers, thoughts, prayers, whatever, cause I'd like to enjoy
myself without getting plastered and hangover-sleeping the whole time away."
"If you ever hear me order a decaf coffee, I've been kidnapped and am trying to signal you."
Just remember it was not beauty that killed King Kong it was Hollywood
"Politicians prefer to use the power and reach of tech giants for their own ends. Politicians aren't
at all concerned about the outsized influence of centralized tech, they just want to be in control
of how this power is abused. "
I hate the mobile app's colorization customization. I wish I had a simple dark theme similar to
the website's.
"I don't endorse any of the links on this site, so you can follow them at your own risk, but the
story is one which should be branded upon the ass of everyone who runs for office, and upon
the conscience of everyone who votes. "
Damn, France beating Belgium was only the second result I didn't pick correcly since the
knockout stage started.
Rasputin Treatment' - when a tv / movie villain is so evil the good guy can't just shoot them.
The bad guy has to fall out of a plane, land on a sharp post, have a gasoline truck crash into
him, which then explodes.
"@pitenana I like most of your arguments with other people. maybe you are just playing devil's
advocate with me, dunno. But I don't enjoy debating with you. Let's just say our most basic
world views are too different to debate anything

216
"'An amazingly good coming of age story, complete with an alternate historical backstory, an
evil cabal, interesting characters and an engrossing narrative.' Now on sale for only
$0.99/¬£0.99 #scifi US - UK - "
Talk about a sense of entitlement! The groom sure did dodge this bullet.
Cities need to stop being held hostage by pro sports teams who threaten to leave town if
taxpayers aren't fleeced to the tune of hundreds of millions or billions of dollars for a new
stadium.
"World War 1 was horrific enough with vast artillery barrages, massed machine guns, and
poison gas. How do you make it even more horrifying? Add zombies, apparently."
@Sargonofakkad100 Have you seen Bob's latest bit of character development? It's like the
writers just decided they hated the character and decided to abandon any hint of subtlety
"I've seen more than a few other instances where a founder opted for a lawyer he liked rather
than one who understood the relevant area. In the case of one (now) household name, it very
nearly killed the company."
Baseline/Neutral:
Regarding exercise - I have started doing light yoga before and after exercise. That has made a
huge difference in recovery and just feeling better all around. It has made all the difference in
regards to my back pain as well. Going to be 52 in a couple of months and feel better than ever.
We are having steak tonight, so I wanted a side salad. I went out to the garden and picked
cherry tomatoes, cucumbers, and basil to make my Cucumber Caprese Salad! (ok, it is Hubby's

217
recipe, but I am the one who makes it.) I use fresh basil when available!If you want the
cucumber caprese salad recipe, go here --->
I'm thinking about making a youtube channel where I just read fiction that's in the public
domain. It will be good practice for seriously doing audiobooks: all my voice work so far has
been shorter than books and I haven't had to edit it myself.
I moved to a rural area a yr ago in July. I've spoken to my neighbors more in 1 yr than in 10 yrs
on Long Island. He helps me mow my land. You rely on each other more than in a city or
suburb.
Would you like to log into the WiFi of this building you're driving by? No? How 'bout this one?'
-My phone
@calcusa The best explanation I have ever read!'
Not so sure that 'Cats are cute even when they are angry.'
This should be interesting.
Anamorph currently #1 Family game on the US AppStore! Thanks for everyone that's playing.
Download todayAppStore: Play: #indiedev #indiegame #iOS #iPhone #Apple #gamedev
A beard was the sign of a soldier in the Roman Republic and Principate. One was expected to
be clean shaven as a civilian, but soldiers on campaign were allowed to grow beards as long as
it was kept close-cut.
More Friday night treats! The kids and I watched some shows on Pureflix and ate homemade
peanut butter cookies! â•¤ï¸•

218
I'm thankful for juul as it got me off chewing tobacco but I do feel it makes me kind of dumb
and dull.
The weather broke Thursday and we are back to reasonable temps (high 70s, low 80s). The
garden likes that more than the lousy 90s of the 10 days prior (blech).I had lunch with my son
and his SO today, and I dumped off this round of veggies on them. She is too nice to not take
them. In the photo (possibly covered): lettuce, green beans, peas, banana peppers, and
cucumbers.
Loading the trailer to go camping... the dog is not guarding it, she's just afraid it might leave
without her. [which has never happened]
This is a good read!
Telegram just rewrote its user agreement to hand over some data to governments.
Notice how some people are obsessed with how many followers they have? Quantity more
important that quality? Just a thought.
I can never get a non-blury picture of Uther! He's always moving!! The kids put a blanket on
him when he was in the hallway chewing on one of his toys. I thought he looked darling.
Veggie pizza night!! Yellow squash and tomatoes are from our garden and the crust is
homemade, hand tossed.
September already huh? Fall is my favorite time of year. Only one drawback...Before you know
it the snow will be flying. But then, there is this...Hockey starts in a couple weeks

219
Criticisms of Systemic Racism:
Remember that you're white before you're anything else and this impacts every single way you
interact with the world compared to poc [people of color]
whiteness is a shield for you as a person and its part of your privilege, ignorance on said
privilege is complacency and support to white supremacy. A poc [person of color] saying you
have privilege is not an insult but a fact and a lot of you seriously have to think critically for
once and question as to why you feel insulted.
Yo don't automatically assume that white people who happen to be a part of marginalized
groups will care about your experiences of racism by default, or be able to relate to you.
It is genuinely impossible for white people to fully understand the trauma that racism causes. I
will never be able to connect to a white person's experiences of oppression with my experiences
of racism.
My friend has had the police called on him for sitting in a parking spot eating a sandwich...at a
park....in his own neighborhood. I hope one day white people can't wave "pre-crime" around as
a weapon any longer.
White liberals and conservatives use Asians in their political party to attack minorities of the
other party.
It still amazes me how America is so racist that an entire culture of police worship rose in
response to black people asking not to be killed over minor infractions or even doing nothing
wrong at all.
Perhaps the biggest white privilege ever, is the freedom as a white person to be ignorant of
racial discrimination and their responsibility in it.

220
The mechanism at which white fragility works is simple: the racist system created and
maintained by white people, bestows upon them benefits at the expense of marginalized groups
over decades, moving them at the top of the power hierarchy. This leads to white people feeling
accustomed to this type of preferrential treatment on a societal level and shelters them from any
real discrimination and prejudice that marginalized races face on a daily basis.
Upon confrontation on issues regarding racism and how they [white people] benefit from such a
racist system, they will often deny, ignore, defend or victimize themselves, and throw fits
similar to that of narcissistic rage.
White people have a right to dignity and to exist. They don't have a right to be fragile and
uncomfortable whenever they're confronted with race issues, especially when they're not
harmed by them.
Actually, we do have examples of POC making their own media to get representation. An
excellent example would be Hamilton. The only problem is that people will riot over the fact
that POC are doing anything at all.
"Write your own damn musical." "Okay sure, it'll be fucking awesome." "How dare you write
your own damn musical, I am insulted by the lack of white people, you're the real racist here, go
back home."
I hate this, but what I hate more are the people (usually cishet white men) who complain about
diversity that's "not done well." To them, it's not that they hate diversity/representation, it's just
that conveniently, in almost every instance, they deem that it's been done poorly or for the
"wrong reasons" which is vague, meaningless, and clearly a shitty cover for their bigotry.
always complaining that those darn gays/pocs/etc. are just being represented for political

221
brownie points or some BS like that, as if they're the judge for when diversity is done "well" or
not.
Stop Being ignorant about white privilege. It does NOT mean you're racist. It does NOT mean
your life has been easy. It does NOT mean you don't face struggles too. It means your life isn't
made harder by your skin color. THAT'S IT. HAVE SOME EMPATHY. DON'T MAKE IT
ABOUT YOU.
The trouble with "white privilege" is that it doesn't feel like privilege when you see your own
life as the baseline.
White privilege exists as a direct result of both historic and enduring racism, biases, and
practices designed to oppress people of color.
Damn, so you telling me I can storm the Capitol building if I'm a white dude with a confederate
flag but I can't peacefully protest without being shot with tear gas and having the shit beat out of
me if I'm black or advocate for black causes? Only in mf America.
if your solidarity with other oppressed groups doesn't extend to when YOU'RE criticized, you're
a bigot and your allyship is not intersectional and is ONLY performative.
The Canadian government and Catholic Church intentionally destroyed an entire generation of
indigenous people. Please, support your Indigenous friends, and please stop pretending that
Canada is better than America when it comes to institutional racism.

Abstract (if available)

Abstract Full title: Self-reported and physiological responses to hate speech and criticisms of systemic social inequality: an investigation of response patterns and their mediation by demographic and psychometric variables. Abstract: These studies investigate how people respond to hate speech vs other types of speech in several modalities. We explored how people rated hate speech vs neutral speech in terms of offensiveness and upsettingness, and stress responses measured via physiological changes in skin conductance and heart rate. We found that hate speech was consistently rated as being significantly more offensive/upsetting than neutral speech, hate speech elicited more skin conductance responses, with greater magnitudes, measured via area under the curve, than neutral speech, and that hate speech elicited greater changes in heart rate than neutral speech.

We explored how personal characteristics affected responses to hate speech, including demographics, self-relevance, and psychometric traits. We compared women’s vs. non-women’s responses to specifically misogynist hate speech, as compared to hate speech targeting a wide range of other groups (“general hate speech,”). This investigated whether people who were members of the group targeted by hate speech (in this case, women) exhibited heightened responses compared to hate speech that was not personally relevant. We found that self-relevance did not have an effect on offensiveness/upsettingness ratings- in fact, both women and non-women consistently rated hate speech targeting a variety of other groups to be more offensive than misogynist hate speech. We also found that self-relevance did not have an effect on any of the physiological responses either. Interestingly, however, all participants’ physiological responses to misogynist hate speech were more frequent, and of greater magnitude, than their responses to general hate speech. We hypothesize that this may indicate that misogynist hate speech might be eliciting threat assessment and salience related responses, while general hate speech may be eliciting emotional responses, as supported by out emotion differentiation models.

We investigated whether being a member of a non-targeted marginalized group enhanced responses to hate speech, via an “empathy,” effect. We examined how gender, race, and sexual orientation affected response patterns. We found that, consistent with previous literature (Cowan and Khatchadourian 2003, Cowan et al 2008, Lo Cricchio & Stefanelli 2023, LeMaire 2014) women consistently rate hate speech as being more offensive/upsetting than non-women. Outside of this effect, demographics did not have an effect on rating or physiological responses to hate speech.

We investigated how social dominance orientation, belief in a just world, ambivalent sexism, grandiose narcissism, and mental health impacted responses to hate speech. These psychometric variables could be used to predict offensiveness/upsettingness ratings, but had no bearing on the increased physiological arousal experienced when reading hate speech. Social dominance orientation consistently had the greatest predictive ability, followed by ambivalent sexism. Belief in a just world also had strong predictive ability for ratings, but not enough to improve the linear mixed effects model of rating that already included social dominance orientation and ambivalent sexism as predictors. Grandiose narcissism and mental health had some predictive ability on the model, but did not have the same predictive strength, or consistency among datasets, that the other variables had.

We also directly compared how people responded to hate speech compared vs. how they responded to criticisms of systemic racism: non-hateful statements describing the differential effects of systemic racism on white vs non-white people. These statements were meant to evoke a sense of “reverse racism,” in some participants, so that the similarities and differences between how responses to these stimuli and actual hate speech varied. We found that criticisms of systemic racism were rated as being more upsetting than neutral speech, but not as upsetting as hate speech. Further, we found that criticisms of systemic racism elicited physiological responses patterns that matched the patterns observed in response to misogynist hate speech, but not general hate speech. We again hypothesize that this may be indicative of a salience-related response, as opposed to the emotional responses elicited by general hate speech.

We found that queer participants rated criticisms of systemic racism as less offensive than straight participants, but that demographics outside of this did not have an impact on rating responses. Finally, we found that the same psychometric variables that predicted ratings of hate speech also predicted ratings of criticisms of systemic racism, but in an opposite direction, since people’s ratings of hate speech and criticisms of systemic racism had an indirect relationship.

Finally, we created a framework for integrating recorded changes in heart rate, heart rate variability (via RMSSD), and skin conductance to assess what types of emotional responses the stimuli elicited, on intersecting scales of approach-avoidance, and positive-negative valence, with attention orienting responses removed and considered separately. We found that hate speech elicited more instances of negatively-valent emotional responses, both approach and avoidance motivated, than neutral speech, as well as more attention orientation responses. We found that general hate speech elicited more anger-related responses (negative-approach), while misogynist hate speech elicited more fear/threat responses (negative-avoidance).

Linked assets

University of Southern California Dissertations and Theses

Asset Metadata

Creator Ryan, Erin Grace (author)

Core Title Self-reported and physiological responses to hate speech and criticisms of systemic social inequality: an investigation of response patterns and their mediation…

School College of Letters, Arts and Sciences

Degree Doctor of Philosophy

Degree Program Neuroscience

Degree Conferral Date 2024-08

Publication Date 08/13/2024

Defense Date 07/31/2024

Publisher Los Angeles, California (original), University of Southern California (original), University of Southern California. Libraries (digital)

Tag ambivalent sexism,belief in a just world,criticisms of systemic racism,demographic,demographics,emotion,emotional responses,emotional responses determined by physiology,Empathy,Gab,gender,grandiose narcissism,hate speech,heart rate,heart rate variability,ingroup,Mental Health,OAI-PMH Harvest,outgroup,physiological responses,physiology,psychometrics,race,rating,reddit,reverse racism,RMSSD,self-assessment,self-relevance,self-relevant,sexual orientation,skin conductance,social dominance orientation,social media,threat,threat assessment,threat response,Tumblr

Format theses (aat)

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Monterosso, John (committee chair), Kaplan, Jonas (committee member), Zevin, Jason (committee member)

Creator Email e.g.ryan6@gmail.com,erinryan@usc.edu

Permanent Link (DOI) https://doi.org/10.25549/usctheses-oUC113998TFS

Unique identifier UC113998TFS

Identifier etd-RyanErinGr-13380.pdf (filename)

Legacy Identifier etd-RyanErinGr-13380

Document Type Dissertation

Format theses (aat)

Rights Ryan, Erin Grace

Internet Media Type application/pdf

Type texts

Source 20240813-usctheses-batch-1197 (batch), University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

Access Conditions The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the author, as the original true and official version of the work, but does not grant the reader permission to use the work if the desired use is covered by copyright. It is the author, as rights holder, who must provide use permission if such use is covered by copyright.

Repository Name University of Southern California Digital Library

Repository Location USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA

Repository Email cisadmin@lib.usc.edu