1.1. Online trolling: Definition and relevance

A troll is defined as a person who (a) starts and/or exacerbates disruptive conflict online for their amusement; (b) is often deceptive, as they tend to have second social media accounts used for trolling; (c) is tactically aggressive to increase emotional responses; and (d) disturbs regular discussions on online platforms to seek attention [1]. Approaches to trolling other internet users with malicious intent can include veering a conversation off-topic as well as being deliberately controversial, offensive, or inflammatory [13]. Why are people becoming trolls? One explanation is that the internet can facilitate disinhibition [14, 15] which positively predicts cyberaggression [16]. According to an early study on trolls [17], users engage with trolling because they are bored, seek attention or revenge, and find it funny to create trouble for platforms and other users. To create the desired disruption, trolls may write messages that are (a) outwardly sincere, (b) deliberately designed to provoke, and (c) a waste of time through fruitless arguments [18]. At times, the media and scholars conflate trolling with any negative behavior that occurs online, e.g., cyberbullying, parody, or flaming, when the definition of trolling should be limited to social phenomena “performed individually or collectively in varying online contexts, which involves the use of antagonism, deception and vigilantism […] to provoke reactions from people or institutions” (p. 1,078) [19].

Notably, the above describes a kind of trolling behavior that aims to negatively affect other users and online discussions. However, trolling can occur in a more light-hearted or even amicable way, e.g., Sanfilippo and colleagues [20] differentiate between serious and humorous trolling. This differentiation is also highlighted by the distinction participants drew between circumstantial trolling and trolls who are committed to irritating and iterating their actions [20]. Here, we are interested in the malevolent troll as defined by Hardaker [1] because of their high relevance to society due to their anti-social behaviors [20]. In fact, being the victim of trolling is widespread: On Facebook alone, 88% of U.K. teenagers reported that they had been bullied or trolled [2], and 77% of U.S. adults reported harassment [3]. Online harassment can lead to anxiety, sleeping problems, and even suicidal thoughts [3], and 66% of U.S. adult internet users strongly agree that internet trolls are detrimental to society [4]. In a survey of British Members of Parliament, all respondents reported experiences of being trolled and women reported consequent concerns about their safety [21].

Some predictors of trolling behavior have become apparent in the literature. For example, research suggests that men are more likely to troll others than women [5, 8, 22] but this gender difference is not found in all studies [7]. Moreover, some studies found that younger people are more likely to troll others [22, 23]. The literature has also showcased that personality is a promising predictor of trolling behavior. The following section will go into more detail about personality traits that predict online trolling. First, we will introduce research on the Dark Tetrad and online trolling; then, we will discuss a link between trolling and humor. Finally, we will outline why situational factors, especially social exclusion, may lead to trolling behavior.

1.2. The dark tetrad and online trolling

Prior research on online anti-social behavior asked, ‘who trolls others?’, sparking investigations of the trolls’ personalities. Specifically, research has confirmed the link between trolling behavior and the Dark Tetrad traits (sadism, psychopathy, Machiavellianism, and narcissism). Narcissism refers to excessive self-love and a grandiose sense of self-importance [24]; Machiavellianism refers to the willingness to manipulate others [25]; subclinical psychopathy refers to fearless dominance and disinhibition [26]; and sadism refers to intentionally inflicting psychological/physical pain for enjoyment or power [27]. These four correlated, theoretically distinct traits share a core of callous manipulation [28]. Moreover, the Dark Tetrad facets are associated with self-reported, observer-reported, and behavioral aggression [28]. Indeed, research has confirmed the link between people scoring high on the Dark Tetrad traits and trolling behavior [5–8, 23, 29–31] which may be due to lower affective empathy in these individuals [32–34], a tendency for moral disengagement [35], and reduced behavioral inhibition anxiety [36]. Moreover, all four facets are positively associated with dominance [37] and social dominance orientation is also associated with past trolling and acceptance of trolling [38]. Another reason for the association between the Dark Tetrad and trolling behavior could be intrinsic enjoyment: Research indicates that sadism, psychopathy, Machiavellianism, and narcissism correlate positively with one’s enjoyment of viewing violent stimuli [36]. Sadism is also related to experiencing greater pleasure during an aggression [39]. Moreover, a recent meta-analysis suggests that the relationship between sadism and aggressive behavior is stronger in online settings, perhaps due to anonymity [40]. Psychopathy specifically may also be related to trolling behavior because of its association with impulsivity [26, 41]. In the case of narcissism, Vize and colleagues [42] showed that antagonism primarily drives aggressive behaviors, although all narcissistic dimensions are related to aggressive behavior [43].

Table 1 showcases the correlates of trolling constructs with the Dark Tetrad in previous research. Overall, a clear pattern emerges, with higher scores on the Dark Tetrad facets being related to more self-reported trolling activities. This pattern has been confirmed for sadism in a recent meta-analysis which revealed a pooled correlation between everyday sadism and online trolling of .52 [40].

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Correlations between trolling assessments and the dark tetrad facets.

https://doi.org/10.1371/journal.pone.0280271.t001

Based on this research, we aimed to confirm previous correlational findings:

H1: The Dark Tetrad is positively associated with global trolling behavior.

H1a: Machiavellianism is positively associated with global trolling behavior.

H1b: Narcissism is positively associated with global trolling behavior.

H1c: Psychopathy is positively associated with global trolling behavior.

H1d: Sadism is positively associated with global trolling behavior.

1.3. Humor styles and online trolling

This study also investigated the relationship between humor styles and trolling behavior. Specifically, we aimed to investigate humor as conceptualized in the humor styles questionnaire (HSQ) [45]. The theory of the HSQ assumes that humor can be adaptive or maladaptive for well-being and that people use humor to enhance the self and/or their relationships with others [45]. Aggressive humor refers to humor that enhances oneself while hurting others, while self-defeating humor is detrimental to oneself and is used to improve relationships [46]. Meanwhile, affiliative humor and self-enhancing humor are detrimental neither to oneself nor others; more specifically, affiliative humor is used to improve relationships while self-enhancing humor aims at enhancing oneself [46].

In two previous studies, active trolls stated that they engage in trolling behavior for their enjoyment [11] and instant entertainment as well as gratification [47]. This suggests that internet users’ humor styles may also affect whether they troll others online. Indeed, one study [48] recently showed that trolling behavior is associated with aggressive humor as well as katagelasticism (i.e., the joy of laughing at others). Moreover, self-enhancing and self-defeating humor were both related to trolling behavior in that study. Some authors suggest that katagelasticism may be a cause of trolling behavior [49]. Additionally, aggressive humor is positively associated with the readiness to be verbally aggressive [50] which, we expect, may be expressed in trolling. In fact, sarcasm and mockery (aggressive forms of humor [51]) can be tactics trolls use to disrupt discussions [20, 52].

Previous research has also shown that humor styles are associated with the Dark Tetrad. The four dark personality traits are linked with inadequate humor [32], e.g., schadenfreude in social, academic, and mourning contexts [35, 53]. Moreover, humor research indicates that the Dark Tetrad facets are linked to how people utilize and enjoy humor [54], e.g., higher Machiavellianism and subclinical psychopathy scores are associated with aggressive humor [55, 56]. Specifically, psychopathy appears to have a stronger connection with aggressive humor and katagelasticism than Machiavellianism and narcissism [54]. Sadism also uniquely explains variance in katagelasticism beyond the Dark Triad [54]. Importantly, katagelasticism not only involves enjoying laughing at others but also actively seeking out situations where one can ridicule others [54]. We believe that internet trolling may be an expression of katagelasticism in people who score highly on the Dark Tetrad facets. In sum, it is likely that trolling behavior is associated with more aggressive humor and this link may exist because people who score highly on the Dark Tetrad use humor differently than people who score lower on the Dark Tetrad. Based on the prior research, we had one further hypothesis and one research question:

H2: Aggressive humor is positively associated with trolling behavior.

RQ: How do affiliative, self-defeating, and self-enhancing humor relate to trolling behavior?

1.4. Social exclusion and online trolling

While findings of the Dark Tetrad traits and internet trolling are relatively consistent across studies, it is important to keep in mind that other variables may be of importance as well: Naturally, behaviors can be influenced by individual (e.g., a sadistic personality) and situational (e.g., being exposed to aggressive behaviors from others) factors and this should also be the case for trolling behavior. For example, in a study in which trolling behavior was assessed by rating comments the participants made, the authors found that negative mood (and exposure to troll comments) triggers trolling behavior [9]. Other studies found that anonymity in chat rooms led to more troll comments than a chat room condition where participants were identifiable [12]. Additionally, research suggests that boredom in life [57] and loneliness [58] predict trolling. These studies highlight the importance of situational factors (here negative mood, exposure to trolls, anonymity, boredom, and loneliness) in the context of internet trolls.

To the authors’ best knowledge, most studies on trolling behavior neglect the investigation of external variables; however, one experiment showed that participants who were socially excluded through a mobile phone text messaging set-up with two other people wrote more provocative messages afterward [10]. Thus, one of the situational factors that might increase trolling behavior could be social exclusion: Revenge can be a motivation for trolling behavior [11] and aggressive responses to rejection can occur, for example, to regain control [59–61]. This aligns with the meta-analytic finding that the relationship between narcissism and aggression is stronger after provocation [43]. One recent study [62] showed that social exclusion compared to social inclusion leads to significantly higher cyberaggression in narcissistic individuals. Moreover, cyberaggression has been found to relate to loneliness and being less socially accepted [63]. Additionally, in the above-mentioned study, socially excluded participants not only wrote more aggressive text messages but also reported worse mood [10], which has predicted writing trolling comments in other research [9]. Importantly, scoring highly on Dark Tetrad traits is associated with difficulties in emotion regulation [32, 64, 65]; hence, we believe that exclusion experiences may be difficult to process for trolls which then may lead to trolling behavior.

Hence, based on the prior research described above [10, 11, 62], we suggest that it might also be possible that feeling excluded motivates people to troll other internet users to avenge themselves. With quick access to social media and the potential of social exclusion occurring online, trolling posts/comments might be an easy way for people who just experienced social exclusion to regain their perceived control [59–61]. Consequently, we intend to add to the base of knowledge of ‘who trolls?’ by also asking ‘when?’. To that end, not only did we investigate global trolling behavior, i.e., a person’s general trolling behavior, but also immediate trolling motivation after an exclusion experience. We hypothesized:

H3: Participants who are socially excluded show increased immediate trolling motivation compared to people who are socially included.

As shown in Table 1, the Dark Tetrad facets correlate positively with trolling behavior. However, a new pattern emerges when looking at multiple regression results rather than biserial correlates: In prior research, only some of the Dark Tetrad facets explained a significant amount of variance in trolling behavior when looking at partial correlations or multiple regression [6, 8, 22, 23, 58]. Specifically, in some studies [23, 66], only sadism and psychopathic tendencies explained a significant amount of variance in trolling behavior when all facets of the Dark Tetrad were included in a multiple regression analysis. In contrast, earlier work revealed sadism and Machiavellianism as significant positive predictors of trolling enjoyment, while psychopathy was unrelated and narcissism showed a negative association [5]. Other research shows significant positive effects of Dark Tetrad facets except for narcissism [58]. Overall, sadistic tendencies appear to have a greater impact on trolling behavior than the other facets [32]. Hence, while all facets of the Dark Tetrad appear to correlate with trolling behavior, in multiple regression analyses, psychopathic personality traits as well as sadism appear to be more consistent predictors of internet trolling than Machiavellianism and narcissism. This highlights the need to investigate the differing roles of the Dark Tetrad facets in more detail.

H4: Social exclusion and Machiavellianism, narcissism, psychopathy, and/or sadism can predict immediate trolling motivation.

The present study aims at expanding prior findings by (a) confirming the previously established predictive role of Dark Tetrad traits, (b) investigating the role of humor styles as predictors of global trolling behavior, and (c) testing the impact of a situational variable, namely social exclusion, on immediate trolling motivation. Based on these aims, we conducted an experiment using the Cyberball paradigm and tested the effects of inclusion and exclusion on the participants’ immediate trolling motivation.

2.1. Participants

This study includes 1,026 participants (M_age = 26.46 (SD_age = 5.88); 77.2% female) recruited from four German universities and a popular science website for psychology (https://www.psychologie-heute.de/aktuelles/studienteilnahme.html). Our sample size surpasses the necessary sample size of 260 participants required to detect an effect of .18 (at the time of the pre-registration smallest reported effect of the correlations between trolling behavior and the Dark Tetrad [22]) for an alpha of .05 and a power of .90.

2.2. Materials

Demographic questions assessed participants’ age, gender, sexuality, nationality, and favorite social media platform. It was also assessed if participants had a fake account and, if they had fake accounts, on which platforms and for what purposes.

2.3. Procedure

This study was preregistered before data collection (https://osf.io/qsfe5) and adheres to Section 15 of the Professional Code of Conduct for Physicians in Bavaria; hence no vote by the Ethics Committee was necessary. All participants provided online informed consent in accordance with the Declaration of Helsinki. Before the study, participants received information about the context of the research, that all information was anonymous and that they were free to discontinue the study at any point in time. To begin the study, participants had to click on a field that read “I agree with these conditions and want to proceed.”. Minors were not allowed to participate in this research. After giving informed consent, participants answered the GAIT-Revised, SD3, CAST, and HSQ and demographic items. Then, participants were randomly allocated to either a social exclusion or inclusion condition using the Cyberball paradigm. After the Cyberball manipulation, participants were asked to rate their immediate motivation to troll others (IAIT). Following this, study subjects received an explanation of the study’s actual goal.

Throughout the survey, we placed three attention-check items. If participants gave wrong answers to these items, the experiment immediately ended and brought the subjects to an explanation page.

2.4. Analysis

Before the analysis, people were excluded when they indicated that they had not taken the questionnaire seriously (participants were asked directly if they had taken the questionnaire seriously), were already familiar with Cyberball or were unable to see the ball-tossing program, and when they were below the age of 18. Additionally, because some answers were randomly missing due to the questionnaire software, we excluded cases listwise for missing data. Finally, because only one person indicated their gender as “other than male or female,” this participant was also excluded from further analysis. This was necessary since no reliable inferences can be drawn from a sample of only one person.

In our pre-registration, we planned to use means of the measures for the analyses. However, after investigating the item loadings of our measures using confirmatory factor analysis (see S1 Appendix), we were concerned about the validity of the scales when using all items. Consequently, we conducted all analyses twice, once using the means with all items as preregistered and once using means that only included items that had a standardized loading on its factor of at least .40. This also serves as a robustness check of our analyses. We report our analyses with the traditional means in the Results section below. We report the analysis using means without low loading items in the S1 Appendix.

We preregistered an ordinary least square (OLS) multiple regression analysis to test H4 which aimed to predict immediate trolling motivation using social exclusion (yes/no) and the Dark Tetrad (Machiavellianism, narcissism, psychopathy, and sadism). Additionally, age and gender were added as control variables, as those variables have been shown to explain trolling behavior in the past [22]. Our regression assumption checks showed some violations of the homoscedasticity and normal distribution of errors assumption. Hence, we decided to deviate from our pre-registration and conducted a quantile regression analysis, which allows for residuals to have different variances [72] and does not assume parametric distributional form (here normal) of the errors [73]. Consequently, a quantile regression analysis should be better suited for our data. Moreover, prior research has compared simple linear regression with quantile regression for personality trait data and concluded that quantile regression can showcase more nuanced and heterogeneous effects [74, 75].

Unlike OLS regression, quantile regression relies on quantiles of the outcome variable which results in several coefficients for a single covariate [76]. These coefficients are interpreted based on their respective quantile of the outcome variable [76]. For example, while OLS regression may indicate an average effect of a woman’s partner’s meanness on relationship satisfaction, quantile regression can show that for the 15^th quantile of relationship satisfaction (i.e., the least satisfied women), the partner’s meanness can reduce relationship satisfaction by 0.61 points, while partner’s meanness only reduces relationship satisfaction by 0.14 points for the 85^th quantile (i.e., the most satisfied women) [77].

Finally, we decided to conduct a dominance analysis that was not pre-registered. We based our decision on criticism of the use of multivariate statistics in Dark Tetrad research [78]. Dominance analysis can provide indicators of relative predictor importance [79, 80]. Dominance analysis calls one predictor more important than another “if it would be chosen over its competitor in all possible subset models where only one predictor of the pair is to be entered” (p. 134) [79]. The dominance analysis approach should provide more definite answers to the question which of the Dark Tetrad facets are the most relevant to predict trolling than multiple regression. The approach has been used previously to answer similar questions, for example, to estimate which narcissism subcomponent is most important for predicting aggressive behavior [42]. We consider the dominance analysis here as an additional robustness check. We tested the importance of all regression predictor variables of our H4 model. The analysis can be found in the S1 Appendix.

We used the R-packages “lavaan” [81] to conduct our confirmatory factor analyses, “quantreg” [82] to conduct our quantile regression analysis, and “dominanceanalysis” [83] to conduct our dominance analysis (R version 4.1.2). For the remaining analyses, we used SPSS (version 26). For our quantile regression, our solution of the means without low loading items resulted in a non-unique solution. Restricting the tau range from 5:95 to 10:90 resulted in a unique solution and is reported in the S1 Appendix.

3.1. Sample descriptives

Overall, 1,026 people participated in this study; M_age = 26.46 (SD_age = 5.88); 77.2% were female and 22.8% were male. Table 2 describes the study’s sample concerning age and our continuous constructs.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Sample descriptives.

https://doi.org/10.1371/journal.pone.0280271.t002

3.2. Global trolling behavior and personality traits

3.2.1. Global trolling behavior and the dark tetrad.

To test our first hypothesis (H1: The Dark Tetrad is positively associated with global trolling behavior) and the respective sub-hypotheses, we looked at the correlations between Dark Tetrad personality scores and global trolling behavior; see Table 3. Our results indicate that each of the Dark Tetrad personality facets correlates positively and significantly with global trolling behavior. Thus, these findings support H1. Please note that we pre-registered tests for H1 and H2 one-sided but tested two-sided for significance.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Correlation matrix for trolling, the dark tetrad of personality, and humor styles.

https://doi.org/10.1371/journal.pone.0280271.t003

3.3. Immediate trolling motivation and social exclusion

To test our third hypothesis (H3: Participants who are socially excluded show increased immediate trolling motivation compared to people who are socially included), we conducted a t-test with exclusion (yes/no) as the independent and immediate trolling motivation as the dependent variable. Our result suggests that the experience of exclusion did not significantly impact participants’ immediate trolling motivation, t(1024) = 0.91, p = .37, CI = [-.02; .06]. Thus, findings from this analysis did not support H3.

3.4. Predicting immediate trolling motivation

The quantile regression results to test our fourth hypothesis (H4: Social exclusion and Machiavellianism, narcissism, psychopathy, and/or sadism can predict immediate trolling motivation) are graphically presented in Fig 1. For example, the third graph in the first row of Fig 1 shows how Machiavellianism is predictive of immediate trolling motivation. The red horizontal line represents the ordinary least square (OLS) coefficient for Machiavellianism, while the x-axis represents the quantiles (at 0.2, 0.4, 0.6, and 0.8) of immediate trolling motivation. Furthermore, the black broken line indicates the coefficients at the respective quantiles, here ranging from the 0.5 to the 0.95 quantile in 0.10 steps. Machiavellianism does not seem to predict immediate trolling motivation for the 0.2 quantile of immediate trolling motivation, while for the 0.6 quantile, Machiavellianism appears to have a positive (albeit non-significant, see Table 4) effect on immediate trolling motivation.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Graphs of the quantile regression coefficients for all independent variables.

Note. Simple linear regression coefficients (red line) and quantile regressions for exclusion, Machiavellianism, narcissism, psychopathy, sadism, age, and gender (male with female as the comparison group) for the dependent variable immediate trolling motivation. The x-axis represents the quantiles for immediate trolling motivation while the y-axis represents the unstandardized coefficients of the respective independent variable.

https://doi.org/10.1371/journal.pone.0280271.g001

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 4. Quantile regression coefficients of exclusion, machiavellianism, narcissism, psychopathy, sadism, age, and gender for the dependent variable “immediate trolling motivation”.

https://doi.org/10.1371/journal.pone.0280271.t004

Table 4 provides the quantile regression coefficients and their significance as well as OLS regression results. A comparison between OLS and quantile coefficients reveals some significant differences: For example, the OLS regression coefficient for sadism indicates that for every 1-point increase on the sadism scale, immediate trolling motivation also increases by 0.20. In contrast, the quantile analysis shows that there is no significant relationship between sadism and immediate trolling motivation for the quantiles 0.05 to 0.35.

Unlike the OLS coefficient for social exclusion, the exclusion coefficients for the 0.75 and the 0.85 quantiles are significant and–in contrast to our predictions–negative. This indicates that experiencing social exclusion may reduce immediate trolling motivation for people who are part of the upper immediate trolling motivation quantiles.

Thus, quantile regression allowed us to provide a more nuanced view of the relationship between our independent variables and trolling motivation. Overall, we find no significant effects of exclusion experience, Machiavellianism, and narcissism on immediate trolling motivation. Meanwhile, for the 0.45 quantile and higher quantiles of immediate trolling motivation, psychopathy and sadism appear to become increasingly relevant.

3.5. Robustness of findings

Due to concerns about construct validity, we conducted CFAs and consequently reran our analysis using means that excluded low loading items (see S1 Appendix). Overall, our robustness analysis replicates the correlational findings and the t-test finding. However, the quantile regressions differ to some degree. After excluding low-loading items, only sadism predicted immediate trolling motivation for higher quantiles of immediate trolling motivation. In other words, neither Machiavellianism nor narcissism or psychopathy uniquely predicted immediate trolling motivation when we controlled for low loading items.

This finding is mirrored by our additional dominance analyses (see S1 Appendix): When using the traditional means for our dominance analysis, both sadism and psychopathy were the most important predictors for immediate trolling motivation. However, when we used means without low loading items, sadism became the most important predictor and dominated psychopathy with its predictive power.

4.1. Limitations and further research

This study has some limitations: Firstly, since there is no instrument to measure immediate trolling motivation, we used an adapted version of the GAIT [8]. Because no official and validated German scales of the GAIT and CAST existed, we used our translation of the scales. While we aimed to capture the original meaning of the items and the concepts while maximizing the naturalness of the German formulations, this remains a limitation of the present study. Due to concerns about the scale validity, we ran robustness analyses where we excluded low loading items. This led to partially differing results for our quantile regressions and our dominance analyses. Consequently, we urge researchers to translate and validate scales to allow for more rigorous research across different cultures.

Moreover, assessing immediate trolling behaviors (observable) rather than motivation (self-report measure) could prove more fruitful in future online trolling research, since answering a questionnaire may result in socially desirable responses [84, 85] which could make it difficult to find true effects. Additionally, being asked about trolling motivation after an experimental manipulation might hint at the study’s purpose for participants. To investigate online trolling behavior further, more immersive and ecologically valid assessments (as done by Cheng and colleagues (2017) [9]) should be applied. This could even be done through field experiments in online multiplayer games.

One limitation of this study is that we did not include a manipulation check for the social exclusion manipulation. We did not include a manipulation check to avoid making the purpose of the present study apparent to participants and because a meta-analysis of 120 studies found large (d > 1.4) and generalizable effects for the Cyberball manipulation [71].

The present study only investigated malicious, serious trolling. However, another, humorous form of trolling behavior also exists [20], and to the best of our knowledge, this type of trolling has not received much attention in prior personality and humor research. Future studies may want to investigate potential differences in personality facets and humor styles between serious and humorous trolling and whether the same people use both forms of online interaction depending on different situational circumstances.

While our study has some limitations, it also provides some new avenues for further research. We suggest that future trolling research should consider criticisms of the Dark Tetrad [78, 89] and propose two strategies in future study designs. First, we suggest that researchers take the multidimensionality of the Dark Tetrad facets into account and select measures which address the complexity of the dark personality traits. This has already been done in part by Paananen and Reichl [30] when they used verbal, physical, and vicarious sadism, and by March [29] who differentiated between direct and vicarious sadism as well as between primary and secondary psychopathy. We believe that these approaches can help to provide a more nuanced perspective on trolling behavior. Second, we suggest dominance analysis [79, 80] to compare how well components and subcomponents of the Dark Tetrad explain trolling behavior. This analysis has demonstrated, for example, that the narcissism factor interpersonal antagonism explains more variance in aggressive and antisocial behaviors than the facets of extraversion and neuroticism do [42].

Concerning humor styles, an interesting finding of our study is the positive correlation between global trolling behavior and self-defeating humor. It appears that trolls (online or offline) do not only target others but also themselves, using humor. A direction to approach this further may be to include the role of self-esteem as a predictor of trolling behaviors. Moreover, the research on trolling might want to investigate antecedents of perceived funniness [91] to better understand how and why trolls attempt their aggressive versions of comedy. This may also prove fruitful in the context of investigating different kinds of trolling [20].

Finally, we suggest moving beyond correlational studies when investigating trolling research specifically and antisocial social media behavior generally. Correlations between personality facets and trolling behavior do not explain why people choose to troll others when they do. We suggest a thorough examination of (potential) trolling behavior under different situational circumstances to enhance our understanding of internet trolls. This could also help platform providers to create more harmonious communities, e.g., by answering the question of whether certain functions (downvotes, likes, having featured comments, etc.) encourage people to engage in trolling behavior.