University of Amsterdam
MSc in Brain and Cognitive Sciences Cognitive Science
Interpersonal affect regulation and empathic
charac-teristics
in Internet trolling
Andrea Veggerby Lind
Netherlands Institute for Neuroscience Supervisor: Riccardo Paracampo
Co-assessor: Yacob Israelashvili
32 ECTS
UvA ID 11753390
09/01/2018 to 27/10/2018
What happens to a troll in
day-light?
Interpersonal affect regulation and empathic characteristics in Internet
trolling
Andrea Veggerby Lind Abstract
As the Internet becomes increasingly central in our social interaction, poten-tially harmful phenomena such as Internet trolling have arisen. This study aimed to characterise the role of empathy in trolling and online interactions across multiple levels of empathy assessment. A sample of 328 subjects (47% males) between 18-35 years of age participated in an online study, in which they 1) rated the pain intensity felt by actors in pre-recorded videos and 2) completed questionnaires on their trolling tendencies, empathy levels, and in-terpersonal affect regulation tendencies both online and offline. Results re-vealed men to be more prone to partake in or enjoy trolling. We found no sig-nificant correlation between the perception of others’ pain and trolling. Trolling tendencies showed a negative correlation with affective empathy, were posi-tively correlated with interpersonal affect worsening scores online and offline, and negatively correlated with interpersonal affect improving scores offline.
1. Introduction
Humans are social beings. Our sur-vival depends on our relations with other people, from maternal care to complex production and trading sys-tems. In immediate interactions with others, empathy helps us optimize the circumstances, not only for ourselves, but also for others involved (de Vi-gnemont & Singer, 2006). Through the last centuries, technological in-ventions have dramatically acceler-ated the efficiency, range, and com-plexity of human communication and interaction; most notably the Internet (Leiner et al., 2009). The Internet al-lows us to rapidly reach a hitherto un-precedented number of people, unre-stricted by geographical limits. But its vast potential also gives rise to prob-lematic deviance in social behaviour. The circumstances under which we in-teract have shifted, and consequently so has the way empathic mechanisms regulate our interactions. Perhaps due to a level of anonymity and distance unlike in face-to-face conversation, the Internet provides fertile grounds for otherwise socially unacceptable actions (Suler, 2004). This paper is concerned with the empathic charac-teristics underlying the social phe-nomenon of Internet trolling.
1.1. Trolling
Researchers on Internet trolling oper-ate with a large diversity in their defi-nitions of the phenomenon and report heterogeneity in motives and behav-iour (Buckels et al., 2018; Dynel, 2016; Phillips, 2011). As an overarch-ing definition, a troll is Internet slang for someone who, often anonymously
or under aliases, deceives or mocks others on online (e.g., on forums or social media) for the sake of enter-tainment (Dynel, 2016). In innocuous forms, trolling can simply entail start-ing pedantic discussions to amuse themselves and the online audience with whom they share their “achieve-ments” (Dynel, 2016). In its grimmest shape the phenomenon aims to inflict emotional harm or corrupt positive so-cial conduct by exhibiting inflamma-tory opinions of racist, misogynistic, or pornographic character (Phillips, 2015). Through a content analysis of trolling behaviour on a topical discus-sion forum, Hardaker (2010) identified four fundamental components across different trolling styles as being de-ception, aggression, disruption, and success. While the phenomenon can be aggressive and antisocial, trolling distinguishes itself from cyber-bullying as it is not fuelled by personal feuds. Additionally, the controversial and of-fensive opinions posted, are not nec-essarily the personal opinion of the troll, but rather a means to achieve successful trolling.
But why does this harmful Internet phenomenon occur? What can this phenomenon tell us about relationship between online interactions and em-pathy? Trolling may be fuelled by ap-praisal from others for crafty trolling schemes (Dynel, 2016) or by sadistic tendencies (Buckels, Trapnell & Paul-hus, 2014). Either way the antisocial and harmful aspects of the phenome-non indicate that trolling might be as-sociated with an abnormal relation-ship with the way we share the dis-comfort of others.
1.2. Empathy
Human interaction and prosocial be-haviour are thought to be largely guided by empathy, that is, the iso-morphic reaction one experiences when observing or imagining the af-fective state of someone else (de Vi-gnemont & Singer, 2006; Singer & Lamm, 2009). In a study aimed at ex-ploring the social profiles of trolls, Sest & March (2017) observed an im-portant predictive role of empathy on trolling. Here they operated with two commonly distinguished aspects of empathy. First, affective empathy refers to somewhat automatic, emo-tional mechanisms whereas cognitive empathy relates to higher order pro-cesses of understanding the experi-ence of others (Mitsopoulou & Giova-zolias, 2015). They observed that trolling was predicted by a decrease in affective empathy, perhaps implying that the harm done to others while trolling is experienced less intensely. Surprisingly, an increase in cognitive empathy also predicted trolling be-haviour, but only when participants reported high scores on trait psy-chopathy. From this finding, the au-thors interpreted that cognitive empa-thy and trait psychopaempa-thy in sym-phony provide an understanding of how to optimise the emotional impact on others, as well as a callousness to the pain induced while trolling.
A deficit in affective empathy could in-deed modify the way trolls perceive the pain of others, be it social or phys-ical (Singer et al., 2004; MacDonald & Leary, 2005). Buckels et al. (2018) showed participants 6 pictures of
peo-ple in varying degrees of pain and asked to estimate the amount of pain experienced in the video. They found a negative correlation between trolling tendencies and the amount of pain perceived while watching others in emotional or physical pain. This in-dicates that trolls tend to have a de-viant relationship with the perception of others pain. It should be noted that Buckels et al. (2018) showed still im-ages of pain. Still imim-ages, as opposed to videos or live observation, render no dynamic reactions such as hand withdrawal or facial expressions de-veloping. It is possible that the addi-tional information that video stimuli can offer would affect the empathic respond differently. The neural sub-strate of empathy for both sensory and emotional aspects of pain ap-pears to rely on vicarious neural acti-vation in the observer, both in so-matosensory and affective brain areas (Keysers & Gazolla, 2009). While it has been disputed whether somatic or purely affective networks are associ-ated with empathy for pain (Singer et al., 2004), meta-analyses indicate that the complexity of the information provided about the other’s experience is a central determinant in which em-pathic networks are involved (Lamm, Decety & Singer, 2011). The results by Buckels et al. (2018) should there-fore be considered in the context of relatively little dynamic visual infor-mation.
In fact, the visual information avail-able to trolls when inflicting emotional pain online is very low compared to face-to-face interactions, and as such the two are likely foster different
pathic responses. Decety & Lamm (2006) argued that contextual ap-praisal mediates empathic mecha-nisms in a top-down fashion, causing our empathic reactions to differ given the value we give to a situation. Un-like face-to-face conversations, online interactions are often asynchronous and with spatially distant parties, al-lowing further anonymity. This could constitute a context with dampening effect on empathy and have been speculated to diminish social inhibi-tions (Suler, 2004). This further opens the possibility that trolling rises from a disinhibition feature of online interac-tion and is not solely due to personal empathic deviance.
1.3. Pro- and antisocial behaviour
Prosocial behaviour is voluntary be-haviour intended to benefit others (Eisenberg & Fabes, 1990). Whether and how exactly prosocial behaviour is driven by empathic responses is a subject of debate (Singer & Lamm, 2009). On trait level, empathy has been positively linked to prosocial be-haviour (Lockwood et al., 2016). While empirical demonstrations of empa-thy’s exact link with prosocial behav-iour still only provide meagre insight into the relationship between the con-cepts (e.g. Lockwood, Seara-Cardoso, & Viding, 2014), prosocial behaviour is frequently assumed to be motivated partly by the aversive arousal caused by witnessing the distress of another (Decety, Ben-Ami Bartal, Uzefovsky, & Knafo-Noam, 2016). As such, prosocial behaviour could be relieving the ad-verse experience of another in an at-tempt to improve one’s own vicarious
experience. Indeed, prosocial behav-iour has been found positively associ-ated with vicarious activation of the primary somatosensory cortex during pain observation (Gallo et al., 2018), supporting the notion that the sharing of another’s affective state might be somehow related to prosocial behav-iour.
One example of prosocial behaviour is interpersonal affect regulation (IAR; Niven, Totterdell & Holman, 2009), specifically extrinsic affect improving, in which an individual improves the affective state of another by refram-ing, actively changrefram-ing, or diverting at-tention away from a bad situation. In-dividuals, especially those with high emotional intelligence, tend to use their relations with others to improve their own affect through similar strategies (Niven, Totterdell & Hol-man, 2009). Conversely, extrinsic af-fect worsening is an aspect of IAR characterised by the deliberate use of strategies such as creating, bringing attention to, or magnifying the extent of bad situations and feelings of an-other (Niven, Totterdell, Stride, & Hol-man, 2011). This can be seen as anti-social behaviour, a contrast to proso-cial behaviour, defined as actions which deliberately cause harm to or in other ways have negative conse-quences for others (Berger, 2003). Antisocial behaviour is often associ-ated with criminal acts such as vio-lence and theft (Calkins & Keane, 2009), with little concern about the wellbeing of victims. The motivations of such anti-social acts may be com-plex and plentiful, ranging from social
or monetary gains to sadistic tenden-cies (Reidy, Zeichner, Seibert, 2010). Similarly, the antisocial and disruptive behaviour seen in trolling may be fu-elled by a diverse range of motiva-tions. Some researchers have pointed to trait sadism as a driving factor for trolling (Buckels et al., 2014; 2018). Other motivational factors of trolling have been linked to a wish to gain in-group recognition from others in the trolling culture (Dynel, 2016). Just like empathy might be a catalyst for prosocial actions to improve the well-being of others and by extension one-self, it too seems to play an important inhibitory role in anti-social behaviour (Miller & Eisenberg, 1988). This is not to say, that the absence of empathy is what drives socially adverse actions, but rather that in the presence of urges to perform harmful actions, these can be impeded by empathy.
1.4. Aim and hypotheses
The current literature clearly indicates that tendencies to enjoy and perform trolling are associated with deviant empathic characteristics (Sest & March, 2017). We here address cen-tral gaps remaining in the field. Inves-tigating these can help broaden our knowledge of problematic trolling cul-ture and the people who partake. Firstly, while trolling has been linked to decreased perception of pain in others in still images, this connection should be investigated with dynamic visual material in order to gain insight into the empathic characteristics in more ecological contexts. This would help build a broader understanding of how people who troll perceive the
pain of others in different contexts. Secondly, interpersonal affect regula-tion strategies have yet to be investi-gated in trolling. Exploring how trolls use social strategies to positively and negatively alter the affective states of others may offer new perspectives on how the Internet interacts with social behaviour.
The aim of this study is to form an in-depth characterisation of the em-pathic profiles of trolls, spanning mul-tiple levels of empathy assessment. Specifically, we aimed to assess how trolling is related to empathy on an implicit (perception of others’ pain) and explicit (self-report) level, as well as through pro- and antisocial tenden-cies (IAR). Particularly we seek to pro-vide novel insight into the offline so-cial behaviour of people who enjoy trolling to find out, 1) whether trolling is associated with differences in prosocial and antisocial behaviour, specifically interpersonal affect regu-lation, and 2) whether such a differ-ence is specific to online or offline be-haviour. We tested this through an on-line behavioural study in which partic-ipants performed the following tasks: a) viewing short video clips showing hand, upper body and face of a per-son receiving varying intensities of painful electric stimulation and rating the perceived pain intensity, and b) completion of questionnaires assess-ing trait empathy, IAR with online and offline orientations, and trolling ten-dencies
We expected to find a negative asso-ciation between trolling tendencies and the perception of pain in others,
similar to the findings by Buckels et al. (2018). We hypothesised that trolling tendencies were also nega-tively correlated with self-rating mea-sures of affective empathy, thus repli-cating the results of Sest and March (2017). We expected that, due to the anonymity, distance, and asynchro-nisity of online interactions, people would generally tend to use extrinsic affect worsening strategies more on-line than offon-line. Because of the exist-ing evidence that trolls are less em-pathic, and that empathy plays a role in prosocial as well as antisocial be-haviour, we expected trolling tenden-cies to be associated with a higher use of affect worsening strategies and a lower use of affect improving.
2. Methods
2.1. Participants
400 participants were recruited through the online platform, www.prolif-ic.ac (50% male). 72 subjects were cluded due to technical issues, for ex-iting the study prematurely or for missing catch trials more frequently than two standard deviations above the sample mean of missed catch tri-als. This resulted in a final sample of 153 male subjects (mean age = 25.7, SD = 5.0) and 175 female subjects (mean age = 27.6, SD = 4.7) all screened for age (18-35) and Cau-casian origins to avoid potential race effects in empathy for pain ratings (Avenanti, Sirigu & Aglioti, 2010). All subjects gave informed consent, and the experimental design obtained eth-ical approval by the Ethics Review Board at the Faculty of Social and Be-havioral Sciences of the University of
Amsterdam. Participants who com-pleted the study received £5/h as compensation for their time.
2.2. Procedure
From Prolific, participants were di-rected via a URL to www.gorilla.sc, a platform for designing and conducting behavioural experiments online. To ensure that all participants were fa-miliar with the term trolling, all were informed from the beginning that the objective of this study was to charac-terise trolling behaviour, and the con-cept of trolling was defined to them in as neutral terms as possible. Comple-tion of the entire experiment was esti-mated to take 130 minutes and had to be concluded within 200 minutes to qualify for payment. At six occasions throughout the study, participants were encouraged to take breaks of 3 minutes, which they could shorten or extend as they wished. Furthermore, it was possible to take spontaneous breaks at any time.
2.3. Pain rating
1200 videos depicting non-profes-sional actors receiving electrical stim-ulations were pre-recorded. Of these, half of the shocks were of high (painful) intensity, delivered via elec-trodes to the right abductor digiti min-imi muscle (ADM), which is situated on the outer edge of the hand. The other half were low intensity shocks delivered to the left ADM. Half of the videos depicted a female actress and half a male actor. A Digitimer DS7A stimulator was used for the shocks. To minimise diversity of setting context the actors were dressed in a black
t-shirt in front of a black screen. Face and both hands are visible in the video. Each stimulation was cut into a 2500 ms video, using Adobe Premiere, with the stimulation (marked as initial ADM contraction) occurring at 500 ms followed by the actors’ pain reaction (figure 1). As an implicit measure of empathy for pain, a pain rating task was employed. Here, the participants were asked to watch the complete set of 600 videos of either a man or a woman receiving painful (300) and non-painful (300) electric stimulations as described in figure 1. To control for gender effects the participants were counterbalanced between the female and male actor, i.e. the first partici-pant watches the male actor, the sec-ond the female actor etc. Participants were informed that the actors in fact received painful stimulation of varying intensities and, as such, perceived varying levels of pain in recording the videos.
Four different version of the task was created per actor to counterbalance the stimuli and avoid context effects. The videos were grouped in 6 groups
consisting of 50 high and 50 low in-tensity videos. Participants were dis-tributed between the different group versions of their assigned actor. The group order within a version was then randomized, as was the order of the videos within each group.
Trials consisted of a 2s fixation cross followed by display of a video. Upon the video ending, the screen changed to a rating screen. Subjects were
in-Page 7 of 17
Figure 1 Female actor receives electric stimulation at 500ms followed by 2000ms of reaction to the
shock
Figure 2 Pain rating reference scale containing description
structed to rate how much pain they thought the filmed person experi-enced on an 11-point Likert scale, where 0 corresponds to “No pain, No sensation on the hand” and 10, “Ex-cruciating, Painful sensation” (Figure 2). No time limit was imposed for rat-ing. The pain rating task took approxi-mately 80 minutes to complete. For every 100 video trials, 8 blurred videos of the actor receiving no stimu-lation were intermixed to ensure at-tention. Subjects were instructed to respond “Blurry” to these catch trials.
2.4. Questionnaires
2.4.1. Interpersonal Reactivity Index (IRI)
In order to explicitly measure trait af-fective empathy, participants com-pleted the Interpersonal Reactivity In-dex (Davis, 1983), a widely accepted 28-item questionnaire, uses a 5-point Likert-scale, and measures cognitive and affective empathy through 4 sub-scales. Two of these are cognitive em-pathy subscales: perspective taking (spontaneously adopting others emo-tional state), fantasy (transposing oneself imaginatively into fictitious situations). This study concerns the affective empathy subscales,
em-pathic concern (other oriented
con-cern for others), and personal distress (self-oriented distress caused by the distress of others).
2.4.2. Emotion Regulation of Others and Self Scale (EROS)
Emotion Regulation of Others and Self Scale (Niven et al., 2011) is an intro-spective measure of how people con-sciously participate in interpersonal
affect regulation behaviour. We em-ployed only the extrinsic subscale of this questionnaire, which specifically concerned the emotion regulation of others. The EROS measures a be-havioural aspect of empathy by prob-ing the extent to which respondents have, in fact, recently attempted to improve (e.g. “I spent time with
someone to try to improve how they felt”) or worsen (“I told someone about their shortcomings to try to make them feel worse”) the mood of
another. Affect regulation is measured in two factors, affect worsening and
affect improvement which have been
found to be factorially distinct (Niven et al., 2011). To discern between on-line and offon-line affect regulation, we constructed two versions of the 9-item extrinsic subscale answered on a 5-point Likert-scale. In these versions all items and instructions remained unchanged, except being preceded by an initial instruction to answer each questionnaire considering specifically online or offline behaviour.
2.4.3. iTroll
The degree to which our participants can be considered trolls was assessed using the iTroll questionnaire (Buckels et al., 2018). This questionnaire mea-sures subjects’ trolling tendencies on 5 point Likert-scale as a single factor in 12 items addressing participation in trolling behaviour, identification and attitudes (e.g. “I identify with trolling
culture”; “We need more safeguards to prevent online trolling” (reversed), “My online personality is typical of a troll”) and correlated highly with
enjoy-Figure 3 Scatter plot with regression line
plotting affective empathy as a function of trolling tendencies for all participants ment and perceived trolling capability
(Buckels et al., 2018). Therefore, not only does the iTroll scale capture those who actively troll, but also those who participate as audience, on a continuous variable. Because trolls are treated as variations on a spec-trum here in order to capture the phe-nomenon as comprehensively as pos-sible, we did not discern between dif-ferent subgroups of trolling.
3. Results
Mean iTroll scores were distributed around the low end of the scale (M = 1.8, SD = 0.74) and ranged widely on the scale from 1 to 4.6. Male subjects showed higher trolling tendencies (M = 2.12, SD = 0.8) than did female subjects (M = 1.53, SD = 0.55,
t(264.43) = -7.66, p < 0.001).
Correla-tion results reported as significant in this section have been corrected for multiple comparisons.
3.1. Pain ratings and trolling tenden-cies
The implicit measure of empathy in this study was constituted of pain rat-ings of high intensity videos, which were rated with an average of 5.96 (SD = 1.22), with no significant differ-ence between scores of male and fe-male subjects. A Pearson correlation showed no support for our hypothesis of a negative association between pain ratings and trolling tendencies. To test whether there was a habitua-tion effect associated with the length of the task, a one-sided ANOVA was conducted, comparing the effect of trial number on pain rating in the first through to the fifth proportion of
tri-als. The mean pain rating on high in-tensity videos was calculated for each trial number across participants, and averages grouped into 5 groups. There was a significant effect of trial number on pain rating at the p<.05 level for the 5 groups [F(4, 725) = 3.592, p = 0.007]. Tukey HSD tests were conducted on all possible pair-wise contrasts, yielding the following groups significantly different (p < . 05): groups 1 (M = 5.899, SD = 0.161) and 2 (M = 5.992, SD = 0.140), and groups 1 and 3 (M = 6.007, SD = 0.137).
3.2. Affective empathy and trolling tendencies
The IRI subscales, personal distress (M = 13.76, SD = 5.025) and empathic concern (M = 19, SD = 5.733) consti-tute implicit measure of affective em-pathy. Female subjects scored signifi-cantly higher on PD (M = 14.469, SD = 5.175) than male subjects (M = 12.948, SD = 4.733, t(325.33) = 2.779, p = 0.006), and on EC (fe-males: M = 20.349, SD = 5.715, males: M = 17.458, SD = 5.369,
t(324.3) = 4.721, p < 0.001). A single
affective empathy score was created from an average of PD and EC for use in correlations. As hypothesised, Pear-son correlation showed a significant negative association between iTroll scores and the IRI affective empathy
Page 9 of 17
Gender differences in EROS Online Worsening
Figure 4 Violin plot of distribution of EROS
Figure 3 Scatter plot with regression line
plotting affective empathy as a function of trolling tendencies for all participants
scores (r = -.31, p < 0.001). Both gen-ders individually showed the same significant association (F: r = -.21, p = 0.004, M: r = -.26, p = 0.0013).
3.3. Interpersonal affect regulation and trolling tendencies
3.3.1. Online
Online affect worsening was signifi-cantly less frequent (M = 1.626, SD = 0.702) than online improving strate-gies (M = 3.561, SD = 1.015,
t(581.67), p < 0.001). Male subjects
reported higher frequency of online worsening strategies (M = 1.779, SD = 0.744) than females (M = 1.491, SD = 0.636, t(301.01) = -3.733, p < 0.001), but no significant gender dif-ference was found in improving strategies. iTroll scores showed a sig-nificant positive correlation with on-line affect worsening strategies (r = . 22, p < 0.001), thus supporting our hypothesis. Contrary to expectations, online affect improving was not corre-lated with iTroll scores
The results indicate that affect im-proving strategies are used more fre-quently offline (M = 3.788, SD = 0.788) than online (M = 3.561, SD = 1.015, t(616.24), p = 0.001). T-tests
indicated no significant difference be-tween online and offline use of affect worsening strategies, thus rejecting our hypothesis. Offline affect improv-ing strategies were used more fre-quently (M = 3.788, SD = 0.788) than offline affect worsening strategies (M = 1.688, SD = 0.677, t(639.49) = -36.586, p < 0.001). While females re-ported higher use of offline improving strategies (M = 3.884, SD = 0.787) than males (M = 3.677, SD = 0.778), gender difference in affect worsening strategies offline were not significant. As expected, offline worsening strate-gies and iTroll scores were signifi-cantly correlated in a positive direc-tion (r = .24, p < 0.001). Affect im-proving strategies were significantly correlated with trolling tendencies in a negative direction (r = .18, p = 0.001), supporting our hypothesis. In-terestingly, when correlating iTroll and offline worsening scores for each gen-der individually, only females show a significant association (r = .27, p < 0.001) that held up against correction for multiple comparisons, whereas for males the correlation was significant only prior to correction (r = .18, p = 0.023).
4. Discussion
The aim of this study was to provide a multilevel characterisation of the em-pathic tendencies related to Internet trolling, entailing novel insights on how trolls influence the emotions of people around them. Not only did the results indicate a negative link be-tween trolling and self-reported affec-tive empathy. They also suggest that those who troll, more frequently used
Page 11 of 17
Figure 5 Scatter plots with regression lines
plotting online affect worsening as a function of trolling tendencies for all participants (top), females (middle), and males (bottom). EROS worsening scores are averaged over 3 item scores, resulting in groupings along the y-axis.
Figure 6 Scatter plots with regression lines
plotting offline affect worsening as a function of trolling tendencies for all participants (top), females (middle), and males (bottom).
strategies to worsen the affective states of others both online and off-line. Finally, gender differences were found to be highly relevant in trolling culture and its relationship with empa-thy and affect regulation. Subjects in this study generally scored low on the iTroll questionnaire, potentially be-cause we screened for representative sample of the general population rather than for trolls specifically. Still, the sample included enough people, who enjoy, partake in, and or identify with trolling, to span most of the scale and to yield significant correlations.
4.1. Empathic characteristics in trolling
To characterise the relationship be-tween trolling and empathy for pain on an implicit level, one aim of the study was to replicate the pain rating task by Buckels et al. (2018), using dynamic stimuli to provide more infor-mation about the pain felt by the other. Contrarily to our hypothesis, the data current did not suggest a re-lationship between trolling and ratings of others’ pain.
The length of the task could have been associated with a habituation effect through prolonged exposure to pain in others (Carrillo et al., 2015). A one-sided ANOVA was conducted, comparing the effect of trial number on pain rating in the first through to the fifth proportion of trials. The mean pain rating on high intensity videos was calculated for each trial number across participants, and averages grouped into 5 groups. There was a significant effect of trial number on pain rating [F(4, 725) = 3.592, p =
0.007]. Tukey HSD tests were ducted on all possible pairwise con-trasts, yielding the following groups significantly different (p < .05): groups 1 (M = 5.899, SD = 0.161) and 2 (M = 5.992, SD = 0.140), and groups 1 and 3 (M = 6.007, SD = 0.137). As this means a slight in-crease over time, there is no indica-tion that the length of the task caused desensitisation to the stimuli.
Displaying videos rather than still im-ages may account for the discrepancy between the current results and those of Buckels and colleagues. Potentially, trolling is associated less with abnor-malities in empathy for pain, when the available visual information is more detailed (Lamm, Decety, & Singer, 2011). Conversely, previous research (Buckels et al., 2018; Sest & March, 2017) as well as current re-sults suggest an association between trolling and affective empathy mea-sured in similar tasks and via self-re-port. The pain rating task may not have been appropriate for capturing such an association. On a putative note, rating videos of only two pain in-tensities (high and low) may have elicited heterogeneous high and low responses, which depended less on an individuals’ empathic assessment of the pain than if the intensities had varied more.
Affective empathy was further as-sessed on an explicit self-report level, the results of which replicated the findings by Sest & March (2017) by showing a negative association be-tween trolling and affective empathy. It has been argued that empathy is an
important inhibitor of anti-social be-haviour as the personal discomfort in-duced by experiencing others in a bad affective state would make people avoid inducing those states in others (Miller & Eisenberg, 1988). Harmful, disruptive, and anti-social forms of trolling may therefore for some be fa-cilitated by an attenuated perception of the pain they inflict on others on-line.
4.2. Interpersonal affect regulation in trolling
As a novel point of investigation, we sought to examine whether the anti-social properties of people who troll are restricted to trolling situations, or whether they extend into interactions in the personal sphere. With this we further wished to shed light on poten-tial interactions between social behav-iour in online versus offline behavbehav-iour. As hypothesised, the results revealed a negative correlation between trolling tendencies and offline affect improving strategies. That is, the more people identify with or perform trolling, the less they appear to inten-tionally improve the mood of others in their personal sphere. Based on the notion that empathy promotes proso-cial behaviour (Decety et al., 2016), a speculative explanation is that an em-pathic decrease in people who troll, leads to less effort in improving the affective state of the people around them, because the vicarious benefits are less salient.
Interestingly, trolling tendencies were associated with higher use of affect worsening strategies both online and
offline, suggesting that antisocial ten-dencies in trolls indeed do exceed the boundaries of trolling contexts. Con-ceptually, trolling largely overlaps with online affect worsening, as the intention of trolling often is to inflict some form of emotional pain to others via an online media. Crucially, the concepts differ in that trolling primar-ily is anonymous interactions with strangers, whereas IAR also allows for measuring interaction between per-sonal acquaintances (Niven et al., 2011). As a result of this overlap, however, a high score on online affect worsening may be explained, not only by the participant worsening the af-fect of friends and family through on-line media, but also simply by the subject trolling a lot. On the other hand, what clarifies the link between trolling tendencies, and tendencies to worsen the feelings of others in the private sphere, is the positive correla-tion between trolling and offline affect worsening. As internet trolling is not captured in offline affect worsening, it becomes clear the anti-social tenden-cies of trolls are reflected in the way they interact with people offline as well.
This finding is important for the com-prehension of how online social con-duct is affected by an interaction of human empathic dispositions and characteristics of the internet such as anonymity, asynchronicity and mini-mization of authority, which may dis-inhibit impulses otherwise impeded by social norms (Suler, 2004). That con-sequences of deviant social behaviour online are minimized through the aforementioned characteristics may,
argues Suler, cause certain personal-ity types to disinhibit antisocial ten-dencies while others not. Across gen-ders our data suggests that phenom-ena like trolling are not fuelled merely by general online disinhibition, but also individual differences in empathy and social behavioural patterns.
4.3. Gender differences
From our results it becomes clear that the link between trolling and empathy cannot be comprehended without considering the salient gender differ-ences. Firstly, the results suggest that men are more likely than women to enjoy and partake in trolling. Sec-ondly, women showed higher levels of affective empathy than men. Thirdly, the way gender affected IAR differed between online and offline interac-tions. When it comes to improving the affect of others offline, women seemed to employ these strategies more frequently, while there was no gender difference in online interac-tions. Using strategies online to make others feel worse, on the other hand, was more common in men than in women, but this effect was not evi-dent in offline interactions. This may be related to males trolling more in general, and the fact that trolling and online affect worsening are overlap-ping constructs.
Similar to the association between trolling and empathy as a speculative catalyst for prosocial behaviour (De-cety et al., 2016) and inhibitor of anti-social behaviour (Miller & Eisenberg, 1988), the male decrease in affective empathy may account for their in-crease in affect worsening and
de-crease in affect improving. It does, however, not offer an immediate ex-planation for the gender asymmetric differences in online and offline be-haviour. Interestingly, gender also af-fects the relationship between trolling and offline affect worsening. When correlating the genders separately, only female offline worsening scores correlated significantly with trolling tendencies after correcting for multi-ple comparison. That is, while women do not engage in trolling as often as men, the data suggests that the ones who do, tend to also deliberately worsen the feelings of others offline more than people, who do not troll – and perhaps even more so than men who troll.
Speculatively, the gender asymmetric association between trolling and IAR might be affected by a differential ef-fect of online disinhibition between men and women. Alternatively, an ex-planation for this interaction may be found by looking to gender norms within social deviance. Fichman & Sanfilippo (2015) found identical trolling was differentially reacted to and perceived in terms of severity and motivation, depending on the gender of the observer but also the gender of the troll. They further stress that off-line gendered norms are reinforced online, such as women being per-ceived as less problematic, dangerous or malevolent than men. As docu-mented in our sample and supported by previous studies (Fichman & Sanfil-ippo, 2015; Buckels et al., 2014; Sest & March, 2017), women are a minority in trolling culture and many other on-line subcultures. Adam (2009)
high-lights that even within the computing profession there is an overemphasis on the difference between men and women. If women are less expected to engage in problematic behaviour as well as being a minority in trolling cul-tures, trolling may seem as less of a deviant behaviour for men than for women. Having to oppose the social gender norms to engage in trolling, could mean that women, who do troll, differ in empathic and interpersonal affect regulation characteristics only from people who do not troll, but also from men who troll.
4.4. Future directions
To recruit a sufficient proportion of trolls and capture a diverse range of trolling partakers, this study mea-sured trolling tendencies on a contin-uum. Consequently, we did not dis-cern between different types of trolling potentially characterised by diverse trolling behaviours, severity and motives. Since different subtypes could cause different degrees of harm, there is undoubtedly a chance that different empathic characteristics are involved. Moreover, the large gender differences found in these results merit investigation into the role of gender in trolling. Future research should investigate 1) the empathic profiles and pro- and antisocial behav-iour of different trolling subtypes to identify alarming subgroups, 2) the way social disinhibition affect men and women, respectively, and 3) the gendered norms within trolling culture and its effect on female engagement in trolling.
4.5. Implications and conclusion
The Internet now constitutes the set-ting of a substantial part of human in-teraction. As empathy is a central component of social interactions, it is of high societal relevance to under-stand the empathic characteristics of the social world online, and to be able to identify disruptive or abusive phe-nomena within it. The current study has provided further evidence for a negative relationship between Inter-net trolling and affective empathy, as well as extrinsic affect worsening strategies, which seem to pervade offline as well as online interactions. These associations between trolling, empathy, and social behaviour are strongly associated with gender differ-ences. Gaining insight into the em-pathic characteristics and social ten-dencies of Internet trolls can aid ap-propriate legislation and preventive efforts. In turn, this can attenuate negative consequences of trolling such as polarised and aggressive en-vironments on social media, or the heightened risk of self-destructive be-haviour by its targets.
References
Adam, A. E. (2009). The gender agenda in computer ethics. In K. Himma & H. T. Ta-vani (Eds.), Handbook of information & computer ethics (pp. 589–619). Hoboken, NJ: John Wiley.
Avenanti, A., Sirigu, A., & Aglioti, S. M. (2010). Racial bias reduces empathic sen-sorimotor resonance with other-race pain. Current Biology, 20(11), 1018-1022. Berger, K. S. (2003). The Developing Person Through Childhood and Adolescence (6th ed.). New York, New York: Worth Publish-ers.
Buckels, E. E., Trapnell, P. D., Andjelovic, T., & Paulhus, D. L. (2018). Internet Trolling
and Everyday Sadism: Parallel Effects on Pain Perception and Moral Judgment. Jour-nal of persoJour-nality.
Buckels, E. E., Trapnell, P. D., & Paulhus, D. L. (2014). Trolls just want to have fun. Personality and individual Differ-ences, 67, 97-102.
Calkins, S. D., & Keane, S. P. (2009). Devel-opmental origins of early antisocial be-havior. Development and psychopathol-ogy, 21(4), 1095-1109.
Carrillo, M., Migliorati, F., Bruls, R., Han, Y., Heinemans, M., Pruis, I., ... & Keysers, C. (2015). Repeated witnessing of con-specifics in pain: effects on emotional contagion. PloS one, 10(9)
Davis, M. H. (1983). Measuring individual differences in empathy: Evidence for a multidimensional approach. Journal of Personality and Social Psychology, 44(1), 113-126.
De Vignemont, F., & Singer, T. (2006). The empathic brain: how, when and why?. Trends in cognitive sci-ences, 10(10), 435-441.
Decety, J., Bartal, I. B., Uzefovsky, F., & Knafo-Noam, A. (2016). Empathy as a driver of prosocial behaviour: highly con-served neurobehavioural mechanisms across species. Philosophical transactions of the Royal Society of London. Series B,
Biological sciences, 371(1686),
20150077.
Decety, J., & Lamm, C. (2006). Human em-pathy through the lens of social neuro-science. The scientific World journal, 6, 1146-1163.
Dynel, M. (2016). “Trolling is not stupid”: In-ternet trolling as the art of deception serving entertainment. Intercultural Prag-matics, 13(3), 353-381.
Eisenberg, N., & Fabes, R. A. (1990). Empa-thy: Conceptualization, measurement, and relation to prosocial behavior. Moti-vation and Emotion, 14(2), 131-149. Fichman, P., & Sanfilippo, M. R. (2015). The
bad boys and girls of cyberspace: How gender and context impact perception of and reaction to trolling. Social science computer review, 33(2), 163-180.
Hardaker, C. (2010). Trolling in asynchro-nous computer-mediated communication: From user discussions to academic defini-tions. Journal of Politeness Research.
Lan-guage, Behaviour, Culture, 6(2), pp. 215-242.
Keysers, C., & Gazzola, V. (2009). Expanding the mirror: vicarious activity for actions, emotions, and sensations. Current opin-ion in neurobiology, 19(6), 666-671.
Lamm, C., Decety, J., & Singer, T. (2011). Meta-analytic evidence for common and distinct neural networks associated with directly experienced pain and empathy for pain. Neuroimage, 54(3), 2492-2502. Leiner, B. M., Cerf, V. G., Clark, D. D., Kahn,
R. E., Kleinrock, L., Lynch, D. C., ... & Wolff, S. (2009). A brief history of the In-ternet. ACM SIGCOMM Computer Commu-nication Review, 39(5), 22-31.
Lockwood, P. L., Apps, M. A., Valton, V., Vid-ing, E., & Roiser, J. P. (2016). Neurocom-putational mechanisms of prosocial learn-ing and links to empathy. Proceedlearn-ings of the National Academy of Sci-ences, 113(35), 9763-9768.
Lockwood, P. L., Seara-Cardoso, A., & Viding, E. (2014). Emotion regulation moderates the association between empathy and prosocial behavior. PloS one, 9(5), e96555.
MacDonald, G., & Leary, M. R. (2005). Why does social exclusion hurt? The relation-ship between social and physical pain. Psychological bulletin, 131(2), 202. Miller, P. A., & Eisenberg, N. (1988). The
re-lation of empathy to aggressive and ex-ternalizing/antisocial behavior. Psycholog-ical bulletin, 103(3), 324.
Mitsopoulou, E., & Giovazolias, T. (2015). Personality traits, empathy and bullying behaviour: A meta-analytic approach. Ag-gression and Violent Behavior, 21, 61–72. Niven, K., Totterdell, P., & Holman, D. (2009). A classification of controlled interpersonal affect regulation strategies. Emo-tion, 9(4), 498.
Niven, K., Totterdell, P., Stride, C. B., & Hol-man, D. (2011). Emotion Regulation of Others and Self (EROS): The development and validation of a new individual differ-ence measure. Current Psychology, 30(1), 53-73.
Phillips, W. (2011). LOLing at tragedy: Face-book trolls, memorial pages and resis-tance to grief online. First Monday, 16(12), 1–14.
Phillips, W. (2015). This is why we can't have nice things: Mapping the relationship be-tween online trolling and mainstream cul-ture. Cambridge, MA: MIT Press.
Reidy, D. E., Zeichner, A., & Seibert, L. A. (2011). Unprovoked aggression: Effects of psychopathic traits and sadism. Journal of personality, 79(1), 75-100.
Sest, N., & March, E. (2017). Constructing the cyber-troll: Psychopathy, sadism, and empathy. Personality and Individual Dif-ferences, 119, 69-72.
Singer, T., & Lamm, C. (2009). The social neuroscience of empathy. Annals of the New York Academy of Sciences, 1156(1), 81-96.
Singer, T., Seymour, B., O'doherty, J., Kaube, H., Dolan, R. J., & Frith, C. D. (2004). Em-pathy for pain involves the affective but not sensory components of pain. Sci-ence, 303(5661), 1157-1162.
Suler, J. (2004). The online disinhibition ef-fect. Cyberpsychology & behavior, 7(3), 321-326.
