Inflammatory Comments Elicit Less Outrage When Made in Anonymous Online Contexts

Abstract

Digital communities often face difficulties in limiting inflammatory social exchanges. The present studies test one potential obstacle to combating malicious comments online: that characteristics of specific online environments dull emotional reactions to inflammatory speech. Across four studies, results suggest that online contexts, particularly those lacking social information such as names and profile pictures, attenuate negative reactions to malicious behavior relative to face-to-face contexts. Shifting expectations and perceptions of harm may partly account for varying outrage across face-to-face and digital environments.

Keywords

Internet/cyberpsychology digital outrage morality online flaming

You have just got to accept that mean words on the Internet do not hurt anyone…What actually affects people are things that happen in the real world.

—Milo Yiannopoulos on Real Time with Bill Maher (2017)

Social interactions increasingly take place online. People argue about politics, play games, and even meet their soul mates without sharing physical space, an unprecedented form of human communication. Unfortunately, the darker sides of our nature frequently infiltrate digital communities. One might not expect music videos and movie trailers to bring out the worst in humans, but comments below them on YouTube regularly house inflammatory speech rarely encountered in daily life (Moor, Heuvelman, & Verleur, 2010). While others have written on the disinhibitions afforded by digital anonymity (Suler, 2004), here we entertain a perspective echoed by polemicist Milo Yiannopoulos above that digital words carry less weight than words spoken in the “real world.” While Yiannopoulos’s perspective may seem extreme, the digital divide may impede mind perception—a crucial component of moral cognition (Gray, Young, & Waytz, 2012). Relative to face-to-face contexts, inflammatory speech in certain digital environments may feel less harmful and more common, dulling reactions to malicious language and fostering sentiments that mean words on the Internet are less harmful.

We examine three characteristics common to online communities that may dull reactions to inflammatory language. First, because digital environments are typically more deprived of social cues than face-to-face environments, victims may be less salient, making malicious speech seem less harmful. Second, norms in digital contexts may be more permissive of inflammatory comments than those in face-to-face environments. Lastly, we consider whether people make weaker character attributions for malicious speech shared anonymously, a feature afforded by many digital communities. Character attributions are closely tied to moral judgments (Uhlmann, Pizarro, & Diermeier, 2015), and the low costs of anonymous, malicious speech may produce weaker internal attributions to perpetrators.

Person Perception and Expectations in Digital Space

The dyadic morality framework postulates that all moral interactions consist of an actor and a recipient of action (Gray et al., 2012). Perceiving the experiences of victims or benefactors activates cognitive templates for moral interactions, and online behavior may appear less morally relevant because key components of moral cognition are muted (McCall, 2013; Sparrow & Chapman, 2013). Research on computer-mediated communication notes its lack of social cues (Kiesler, Siegal, & McGuire, 1984) and deficiencies in nonverbal and paraverbal communication (Daft & Lengel, 1983; Short, Williams, & Christie, 1976). Self-centered thinking dominates youths’ approaches to online life while considerations of others’ perspectives are less prevalent (Flores & James, 2012). Moreover, cyber bullies expect posting embarrassing jokes or rumors online to cause much less distress to targets than they actually do (Gibb & Devereux, 2016).

Harmful comments may also evoke less negativity when committed online because of differing norms in digital environments. Flaming behavior (or hostility characterized by offensive language) is common on websites such as YouTube, possibly because of reduced awareness of others’ feelings (Moor et al., 2010), and online political discussions socialize individuals to see flaming as more acceptable (Hmielowski, Hutchens, & Cicchirillo, 2014). Observations of Usenet newsgroups find that participants consider flaming an unfortunate but acceptable way of interacting virtually (Lee, 2005), and the use of “flames” becomes quite common during the development of online communication norms (Postmes, Spears, & Lea, 2000). Moreover, common behaviors are judged as more moral than rare behaviors (Lindström, Jangard, Selbing, & Olsson, 2017).

Lastly, digital communications frequently afford users anonymity. In public, identifiable contexts, people’s malicious words, and actions can affect their reputations, which should lead observers to make strong character attributions (e.g., “that person must be a real jerk for saying that”; Kelley, 1973). Online communities often afford complete anonymity, severing behaviors from their impact upon reputations. Audiences may make weaker character attributions and instead attribute actors’ behaviors to external forces (e.g., Internet norms). Viewing perpetrators as harboring wicked desires increases perceived blameworthiness (Inbar, Pizarro, & Cushman, 2012; Uhlmann et al., 2015). Thus, more forgiving perceptions of perpetrators may also mute reactions to inflammatory comments made in anonymous, online contexts.

Immoral Behavior in Face-to-Face and Digital Space

Direct comparisons of reactions to immoral behavior across face-to-face and digital environments are rare, possibly because many daily encounters with morality (e.g., an able-bodied person parking in a handicapped spot) do not have obvious counterparts online. However, Poole (2007) asked junior high through college students about the acceptability of 13 technology-based unethical behaviors (e.g., You delete a file that a classmate saved on the school’s network) and 13 parallel nontechnology behaviors (e.g., As students pass papers forward to the teacher, you remove one of the papers and throw it out). Students rated the unethical behaviors as more acceptable when embedded in technology-based scenarios compared to nontechnology scenarios. Similarly, university students consider it more acceptable to illegally download music or movies from the Internet than to shoplift media from physical stores (Lysonski & Durvasula, 2008). A text analyses of comics found that offensive, stereotypical jokes may be more acceptable online (Shifman & Lemish, 2010), and online video game environments can be especially permissive of sexism (Fox & Tang, 2014).

The Present Studies

Across four studies, we examine reactions to malicious comments made in face-to-face and various online environments, testing four key hypotheses. First, people will react with less outrage to inflammatory comments made in online relative to face-to-face environments. Second, because we theorize that victim salience contributes to outrage, offensive language will evoke more outrage as the richness of social information (e.g., profile pictures and real names) increases in digital environments. Third, both harm perception and expectedness will mediate the effects of an inflammatory comment’s context (online, low social information vs. online, high social information vs. face-to-face) upon felt outrage. Fourth, anonymous environments will reduce outrage by making perpetrators’ character appear less malicious. Study 1 tests our first hypothesis by measuring outrage in response to a controversial comment from an online versus face-to-face class discussion. Study 2 uses an insulting comment made either by a passerby on a city street or across two online environments that vary in social information. Study 3 uses a similar design as Study 2 while also measuring perceived harm and expectations of encountering insulting comments. Finally, Study 4 specifically tests the effects of anonymity upon perceptions of the perpetrator by manipulating both anonymity and an online versus face-to-face environment.

Study 1

We aimed to recruit roughly 128 participants from the University of South Florida psychology participant pool to achieve power of .80 for α = .05 and d = .50 when comparing two independent groups using two-tailed tests. Final recruitment fell just short of this at 119 (82% women, M_age = 20.42, SD_age = 2.28). No participants were excluded from analyses.

Participants read an abbreviated version of an e-mail sent by Yale University’s intercultural affairs council discouraging students from wearing culturally insensitive Halloween costumes. In the online condition, participants saw a screenshot of an online class discussion in which a student made a potentially offensive comment about political correctness (excerpts include, “I’m so sick of this political correctness bs,” “students at the school are pampered little children,” and “grow a pair and get on with your lives”). Participants in the off-line condition saw this same comment in a transcript from a face-to-face class discussion.

Participants then rated their reaction to the student’s comment using 7-point Likert-type scales. One item measured their agreement with the comment (anchors: not at all to very much). Three items, adapted from Tetlock, Kristel, Elson, Green, and Lerner (2000), evaluated felt outrage (anchors: not at all upsetting to highly upsetting, no anger to a great deal of anger, not at all offensive to highly offensive, α = .87). Finally, we asked participants “if part of the student’s grade is based upon class discussion, how should this comment affect the grade?” (very negatively to very positively).

Results

Participants felt more outraged over the controversial comment when it occurred face-to-face than online, t(117) = 2.52, p = .01. Participants were no more likely to agree with the comment online than face-to-face, t(116) = −.61, p = .55, or indicate the comment should impact the student’s grade, t(117) = −.08, p = .94 (see Tables 1 and 2 for means and effect sizes). Following recommendations to examine indirect effects in the absence of direct effects (Hayes, 2013), we estimated the indirect effect of the face-to-face versus online environment upon punishing the student’s grade through outrage. The mediation model (analyzed using the PROCESS macro for SPSS using bootstrapping, 10,000 iterations) indicated a significant effect of the experimental condition upon outrage (b = −.72, p = .01), a significant relationship between outrage and punishing the student’s grade while controlling for experimental condition (b = −.23, p = .001) and a significant indirect effect (indirect effect = .17; 95% confidence interval [CI] = [.03, .44]).

Table 1.

Summary of Raw Means and Standard Deviations Across All Studies.

		Online (LS)	Online (HS)	Face to Face
	Outcome	M (SD)	M (SD)	M (SD)	M (SD)
Study 1	Outrage	3.10 (1.43)^a	—	3.81 (1.67)^b	—
	Agree	3.79 (2.02)^a	—	3.58 (1.73)^a	—
	Punish	4.08 (1.24)^a	—	4.07 (1.22)^a	—
Study 2	Outrage	3.13 (1.57)^a	3.67 (1.45)^b	3.64 (1.42)^a,b	—
	Wrong	4.95 (1.58)^a	5.46 (1.33)^a	5.37 (1.52)^a	—
Study 3	Outrage	3.30 (1.70)^a	3.80 (1.78)^b	4.42 (1.71)^c	—
	Harm	3.54 (1.40)^a	3.69 (1.40)^a	4.45 (1.03)^b	—
	Surprise	2.83 (1.81)^a	3.19 (1.77)^a	4.32 (1.74)^b	—
	Ban	3.39 (2.13)^a	3.64 (2.23)^a	4.39 (2.24)^b	—
		Online Anonymous	Online Identified	Face-to-Face Identified	Face-to-Face Anonymous
Study 4	Outrage	3.89 (1.58)^a	3.67 (1.47)^a	4.74 (1.30)^b	4.48 (1.42)^b
	Character	5.09 (1.20)^a	5.13 (1.30)^a	5.04 (1.28)^a	5.32 (1.12)^a
	Surprise	3.67 (1.73)^a	3.63 (1.76)^a	4.67 (1.54)^b	4.65 (1.75)^b
	Harm	2.37 (1.23)^a	2.62 (1.21)^ab	3.41 (.86)^c	2.89 (1.10)^b

Note. Columns are relabeled for Study 4 to reflect the manipulation of anonymity specifically. Means within the same row that do not share at least one superscript letter in common are significantly different from one another at p < .05. Online (LS) = online, low social information conditions (i.e., those lacking real names and profile pictures), Online (HS) = online, high social information conditions (i.e., containing profile pictures and real names).

Table 2.

Summary of Effects Sizes for Outrage.

	Effect Sizes for Online (LS) Versus FtF					Effect Sizes for Online (LS) Versus Online (HS)					Effect Sizes for Online (HS) Versus FtF
	N	d		95% CI		N	d		95% CI		N	d		95% CI
Study 1	119	.46*		[.15, 1.29]		—	—		—		—	—		—
Study 2	174	.34^†		[−1.04, .02]		179	.36*		[−1.05, −.01]		184	.02		[−.54, .50]
Study 3	409	.66***		[.71, 1.52]		403	.29*		[−.90, −.09]		408	.36***		[−1.02, −.22]
	Effect Size for Online Versus FtF							Effect Size for Identified Versus Anonymous
	N		d		95% CI			N		d			95% CI
Study 4	283		.57***		[.49, 1.17]			283		.001			[−.34, .36]

Note. Effect sizes for each independent variable in Study 4 are estimated while collapsing across the other independent variable. Online (LS) = online, low social information conditions (i.e., those lacking profile pictures), Online (HS) = online, high social information conditions (i.e., containing profile pictures). Dependent variable = outrage.

^†p < .10. *p < .05. **p < .01. ***p < .001.

Study 2

Study 1 provided initial support for our first hypothesis—that inflammatory comments made online evoke less outrage than when made in person. Interestingly, we found that the environment of the comment did not affect participants’ agreement with it, and we observed only an indirect effect upon desire to see the commenter punished. However, the online condition somewhat ambiguously manipulated our proposed mechanisms. We did not include real names or profile pictures, but participants may still have inferred an identified context because screenshots were modeled after the Canvas learning management system. Since social information varies considerably across online contexts, generalizations beyond this test of one specific digital environment are limited. Study 2 attempted to replicate this effect in a different context while also testing the hypothesis that outrage triggered by malicious behavior online will vary depending upon the amount of social information available. Study 1 also used a politically loaded comment (i.e., cultural insensitivity), whereas Study 2 employed a generic, malicious comment. Finally, in addition to measuring participants’ outrage toward the malicious behavior, we measure judgments of the comment’s moral wrongness. For instance, participants may feel less outrage in response to malicious comments made online but cognitively judge them as no less wrong.

Participants and Procedure

We targeted a larger sample size than Study 1 based upon the average effect size observed in social psychology of d = .43 (Richard, Bond, & Stokes-Zoota, 2003). Analyses for power = .80, α = .05, and d = .43 suggest a sample size of 86 per cell. Our final sample overshot this slightly yielding 270 total participants, recruited from the University of South Florida participant pool (73% women, M_age = 21.07, SD_age = 4.88). No participants were excluded from analyses.

Participants each saw an image depicting someone insulting another because of a “nerd culture” reference in one of three environments: face-to-face; online, high social information (HS); and online, low social information (LS). In both online conditions, participants saw a screenshot of one user sharing an image with the phrase “Choose you weapon” underneath several different gaming dice used in board games. Beneath a user has replied to this post saying “Go back to your mommy’s basement nerd.” In the face-to-face condition, the image of gaming dice was depicted on a T-shirt, and participants were told that a passerby on a city street shouted the comment. The online (HS) condition was modeled after Twitter, featuring the victim’s name and profile picture. The online (LS) condition was modeled after 4chan and contained no profile pictures or usernames. Study 2 used the same measure of outrage from Study 1 with one additional item assessing participants’ outrage from 1 (not at all outraged) to 7 (very outraged), yielding a 4-item measure of outrage (α = .92). Additionally, participants rated how morally wrong the insulting comment was from 1 (not at all wrong) to 7 (extremely wrong).

Results

A one-way analysis of variance revealed a significant effect of the environment of the insulting comment upon outrage, F(2, 264) = 3.64, p = .03, η² = .03. Tukey tests indicated that the online (LS) condition differed from the online (HS) condition and from the face-to-face at levels below or approaching statistical significance, p = .04 and p = .06, respectively. A one-way ANOVA also revealed a marginal effect of the comment’s environment upon judgments of moral wrongness, F(2, 264) = 2.96, p = .05, η² = .02. Tukey tests indicated marginal differences between the online (LS) condition and both the online (HS) and the face-to-face conditions, p = .06 and p = .15, respectively. However, the online (HS) condition did not evoke less outrage or judgments of wrongness relative to the face-to-face condition (see Tables 1 and 2 for means and effect sizes).

Study 3

Study 2 mostly supported our first two hypotheses. The online condition lacking social information elicited less outrage than both the face-to-face condition and the online condition containing social information. However, the online condition containing real names and profile pictures did not evoke less outrage than the face-to-face condition, suggesting that moral outrage may be dampened specifically when humanizing social cues are lacking. Study 3 attempted to replicate Study 2 in a noncollege student sample using online environments modeled after other well-known websites. Study 3 also measured the expectedness of the malicious comment and perceived pain of the victim, allowing us to test whether each uniquely accounted for the effects of online environments upon outrage (our third hypothesis).

Participants and Procedure

Given the smaller effect sizes observed in Study 2, we collected a high-powered sample of roughly 200 participants per condition. Our final sample of 610 total participants (43% women, M_age = 36.14, SD_age = 12.46) yielded power of .92 to detect the effect size observed between the online (low social information) and face-to-face conditions in Study 2. Participants were recruited from Amazon’s Mechanical Turk website and compensated US$.40.

Participants read an insulting comment made in one of three contexts: face-to-face; online, high social information (HS); and online, low social information (LS). Participants were told that the comments were taken from a discussion about changes to a statewide education curriculum. The online (HS) condition resembled Facebook, featuring both names and profile pictures. The online (LS) condition resembled the social media site Reddit and contained no images or full names. Users in the low social information condition instead had usernames resembling the names in high social information conditions (e.g., the username Apple4Laura was used in the place of Laura Fisher). The face-to-face condition featured an image of people talking above a transcript from the conversation. Each condition showed several comments; however, two were blurred out leaving only the target comments for participants’ evaluation. The first comment read, “I personally like the new curriculum. If we aren’t willing to adapt to better teach our students then how can we expect them to adapt while learning?” followed by a response reading, “Well, that’s the dumbest thing I’ve heard all day. Thank you for lowering everyone here’s intelligence.”

Participants then completed a single-item measure of perceived pain adapted from the Wong-Baker FACES Pain Scale (Wong & Baker, 1988; used with permission), previously used in the studies of mind perception and moral judgment (Gray & Wegner, 2009). Participants indicated how much pain they thought the target of the insulting comment felt by clicking on one of the six faces, with facial depictions of pain ranging from no hurt to hurts worst. This was followed by the same 4-item measure of outrage from Study 2 (α = .96) and 2 items assessing the unexpectedness of the comments, not at all surprised to very surprised and not at all caught off guard to very much caught off guard (α = .89). Finally, a single item asked how likely participants would be to ban the perpetrator if they could (anchors: not at all likely to ban to definitely would ban).

Results

One-way analyses of variances revealed a significant effect of the environment of the insulting comment upon outrage, F(2, 607) = 21.34, p < .001, η² = .07, perceived pain, F(2, 490) = 23.86, p < .001, η² = .09), felt surprise, F(2, 603) = 39.10, p < .001, η² = .11, and likelihood to ban, F(2, 607) = 11.64, p < .001, η² = .04. Tukey tests indicated that participants in the online (LS) condition experienced significantly less outrage than participants in the face-to-face condition (p < .001) and the online (HS) condition (p = .01). Participants in the online (LS) and online (HS) conditions were significantly less surprised, perceived less pain, and were less likely to ban, compared with the face-to-face condition but did not significantly differ from one another (see Tables 1 and 2 for means and effect sizes).

Next, we examined the predicted mediation model using the PROCESS macro (bootstrapping 10,000 iterations). We entered the environment of the comments as the independent variable, dummy coding the face-to-face condition as the reference group. Perceived pain and felt surprise were entered as mediators and outrage as the dependent variable. This model produced three regression analyses with surprise, perceived pain, and outrage as outcomes in separate models. Environment predicted surprise and perceived pain in their respective models (see Figure 1 for the effects of each dummy-coded variable). Importantly, in the model predicting outrage, both perceived pain (b = .47, p < .001) and surprise (b = .39, p < .001) explained unique portions of the variance in outrage. Neither dummy-coded variable had a direct effect upon outrage while controlling for felt surprise and perceived pain (see Figure 1 for model statistics). Mediation analysis revealed significant indirect effects through both surprise (effect = −.60, CI = [−.81, −.4]2; and effect = −.47, CI =[ −.67, −.31] for the dummy variables with the online (LS) condition coded as 1 and the online (HS) condition coded as 1, respectively), and perceived pain (effect = −.42, CI [−.60, −.28]; and effect = −.35, CI [−.52, −.21] for both respective dummy variables), suggesting that both online conditions (relative to the face-to-face condition) had significant effects upon outrage through the proposed mediators.

Figure 1.

Mediation model testing indirect effects of the online conditions upon outrage through perceived harm and surprise. The independent variable was dummy coded with the face-to-face condition as the reference group. Coefficients labeled b₁ correspond to comparisons between the online, low social information condition and the face-to-face condition (i.e., dummy variable coding the online, low social information condition as 1). Coefficients labeled b₂ correspond to comparisons between the online, high social information condition and the face-to-face condition (i.e., dummy variable coding the online, high social information condition as 1). Direct effects are calculated while controlling for mediators.

Study 4

Study 3 provided initial evidence that both perceptions of harm and surprise contribute to attenuated outrage in certain online contexts. Study 4 aimed to provide a preregistered replication (materials located at https://osf.io/hzrxt/) of the attenuating effects of certain digital contexts upon outrage while also testing the effects of anonymity specifically. We predicted that making a malicious comment in a public, identified context (relative to anonymous) will be perceived as more indicative of the actor’s character across both face-to-face and online contexts. However, we expected malicious comments to be perceived as less harmful and surprising even in identified digital environments. Finally, in Study 4, we controlled for the possibility that participants may perceive different size audiences witnessing the malicious comment.

Participants and Procedure

We targeted statistical power of .80 to examine medium effect sizes (d = .50 between individual cells in our design). A power analysis for comparing two independent groups suggested a sample size of 64 participants per cell to meet this target. We overshot this slightly with a final total of 283 participants. Participants were recruited from Amazon’s Mechanical Turk website and compensated US$.20.

Participants read a remark in response to a woman commenting on infrastructure issues. The insulting comment read, “Well no duh Sherlock. Like that’s anything we don’t already know. What an idiot.” We manipulated whether the comment was made online versus face-to-face and was anonymous versus identified. In the online conditions, participants saw a screenshot of a message board discussion with two comments isolated for evaluation. The face-to-face conditions contained a transcript of a conversation at a public event. The identified conditions contained real names, Emily Riker and Ed Foster. In the online-identified condition, the message board profiles displayed users’ real names. In the face-to-face, identified conditions, discussants were shown wearing name tags and the transcript contained real names. The anonymous conditions simply read “Woman Two” and “Man Two” or “♀user5237” and “♂user3721” for the face-to-face and online conditions, respectively. Additionally, the face-to-face, anonymous condition explicitly stated that the perpetrator walked away before anyone could recognize him.

Participants completed the same measures for perceived pain, surprise, and outrage from Study 3. Additionally, participants rated the moral character of the perpetrator on two items (e.g., Do you think that [Man Two] is mainly a good person or a bad person? anchors: Mainly a bad person to mainly a good person; Inbar et al., 2012). Finally, participants provided a numerical estimate of the number of people who witnessed the malicious comment.

Results

In all analyses, we controlled for participants’ estimates of the number of people who witnessed the malicious comment (log transformed because of substantial positive skewness; skewness = 16.46). A factorial analysis of covariance revealed a main effect of online (vs. face to face), F(1,264) = 24.00, p < .001, η² = .08, but not of anonymity, F(1,264) = .02, p = .88, η² < .001, upon outrage. Online context and anonymity did not interact, F(1,264) = 1.37, p = .24, η² = .005. To estimate indirect effects through each of our proposed mediators and to test the interaction between anonymity and online context, we ran a moderated-mediation model using the PROCESS macro. The analysis generated four regression models testing main effects and interactions for online context and anonymity upon each mediator and the relationship between each mediator and outrage (see Table 3 for detailed results). The online context reduced perceptions of pain (b = −.57, p = .005) and surprise (b = −1.12, p < .001) but not character evaluations (b = −.26, p = .19). Anonymity decreased perceptions of pain (b = −.42, p = .007) but not surprise (b = −.05, p = .84) or character evaluations (b = .03, p = .84). No interactions were significant (p > .53). Perceived pain (b = .49, p < .001), surprise (b = 31, p < .001), and character (b = .17, p = .01) all uniquely predicted outrage. Since no interactions between anonymity and online context emerged, we estimated indirect effects (via bootstrapping, 10,000 iterations) of online (vs. face-to-face) upon outrage while collapsing across the anonymous/identified conditions. Indirect effects of online context upon outrage emerged through both perceived pain (effect = −.30, CI −.57, −.06) and surprise (effect = −.33; CI [−.56, −.13]) but not through character evaluations (effect = −.04, CI [−.12, .03]). A similar analysis found an indirect effect of anonymity upon outrage through perceived harm (effect = −.23, CI [−.41, −06]) but not through surprise (effect = −.02, CI [−.15, .13]) or character evaluations (effect = .004, CI [ −.04, .06]).

Table 3.

Study 4 Regression Output for Models Predicting Outrage and Proposed Mediators.

	Perceived Pain		Surprise		Character		Outrage
Predictors	b (SE)	t	b(SE)	t	b(SE)	t	b(SE)	t
Online	−.57 (.20)	−2.83**	−1.12 (.31)	−3.64***	−.26 (.20)	−1.33	−.49 (.18)	−2.65**
Anonymous	−.42 (.15)	−2.72**	−.05 (.23)	−.21	.03 (.15)	−.21
Online × Anonymous	.19 (.30)	.62	−.21 (.47)	−.45	−.11 (.30)	−.36
Perceived pain							.52 (.07)	7.58***
Surprise							.30 (.05)	6.57***
Character							.14 (.06)	2.15*

*p < .05. **p < .01. ***p < .001.

Discussion

Across four studies—with diverse samples and types of online environments—we found evidence that malicious behavior in certain digital environments evokes less outrage than in face-to-face contexts. Participants were also more likely to endorse punishing commenters in face-to-face contexts in Studies 1 and 3 (albeit only indirectly, through outrage, in Study 1). The impoverished social information of online communication may make targets of provocations less salient. In support of this, participants in Studies 2 and 3 indicated feeling less outrage in the online conditions without social data than the online conditions with social data. However, Study 3 found no differences between the two online conditions in either perceived pain or felt surprise, leaving some uncertainty regarding what drove the differences between the online conditions high and low in social information. In Studies 3 and 4, both the perceived harmfulness and expectedness of comments independently mediated the effects of the comment’s environment upon outrage, providing some evidence for our proposed mechanisms. In Study 4, the digital context itself and anonymity decreased the perceptions of harm but only the former reduced feeling surprised by the malicious comment. Character judgments of the perpetrator predicted outrage but were unaffected by the digital context or anonymity. In sum, we find evidence that reactions to inflammatory comments in certain digital environments are muted relative to face-to-face contexts, and this effect is likely partly accounted for by victim perception and shifting norms. The social information present in digital contexts also appears to affect reactions to malicious behavior, but we find mixed evidence for why this is the case.

The present findings may at first appear to conflict with recent work suggesting that digital media amplify outrage (Crockett, 2017). Crockett argues that online contexts increase exposure to extreme moral transgressions, decrease the effort required for outrage expression, and reinforce habitual outrage. We do not think these claims conflict with the present findings, rather we believe our results point to nuance in the Internet’s effects upon felt outrage. Much of Crockett’s discussion targets the context in which immoral events are learned of and shared, whereas our studies manipulate the context in which the immoral behavior itself takes place. For example, people may be more likely to express outrage over inflammatory behavior when discussing with others on social media but may feel less outrage if the inflammatory behavior in question took place in an anonymous, digital context than in a face-to-face context.

Limitations

The present studies examined one type of malicious behavior—inflammatory comments, but we found effects for politically charged and generally insulting comments. Moreover, our tests of mechanisms were limited to one approach to measuring perceived harm and the expectedness of inflammatory comments. While the results of Studies 3 and 4 support our predictions, we cannot conclude whether the causal chain implied by our model is correct. For instance, it is possible that participants’ outrage caused them to perceive more harm. The relationship between outrage and harm perception may be bidirectional. Furthermore, while the proposed mediators uniquely predict outraged responses, they may also reinforce one another (e.g., decreased surprise reduces harm perception).

We also caution generalizing the present findings to all online versus face-to-face contexts. We propose specific features of online environments drive their dampening effects upon outrage. Digital environments rich in social information and supportive of norms against inflammatory behavior may produce more outrage than face-to-face interactions lacking these conditions. However, literature on computer-mediated environments highlights factors such as anonymity and social deprivation as defining characteristics of many digital communities (Bargh & McKenna, 2004). So while we recommend caution in drawing broad conclusions about online versus face-to-face contexts, the characteristics we propose attenuate reactions to inflammatory speech are likely more common in digital spaces.

It is important to note that participants only imagined face-to-face interactions. They did not witness firsthand facial expressions and other social cues linked to harm perception. If anything, we might expect that witnessing face-to-face, malicious behaviors firsthand would increase outrage, making the present experiments conservative tests of the hypotheses. Still, this may cast some doubt onto whether victim salience explains the present results. First, in Studies 3 and 4, participants view the face-to-face comment as more harmful even while controlling for comment expectedness. So it is unlikely that the attenuated perceptions of harm in the online conditions are explained by the behaviors seeming more normal. In fact, the present results are counterintuitive in one sense. Should not broadcasting malicious behavior to thousands of viewers increase perceived harm? We are unable to definitively conclude that imagining a face-to-face interaction evokes more perceptions of mind than viewing a screenshot of an online interaction, but even if the attenuated perception of harm observed in certain online contexts are explained by general beliefs that digital, immoral behavior is relatively victimless, we argue that these beliefs still likely stem from barriers to mind perception online.

Finally, one variable left unexplored in the present article is the closeness of the relationships between the interactants. Witnessing someone attack a close friend may evoke a different reaction than witnessing someone attack a stranger. This may contribute to some of our findings (e.g., less outrage in the Reddit than in the Facebook condition in Study 3), but we argue other results cannot be easily explained by this variable. For instance, in some face-to-face conditions (i.e., Study 2 and Study 4), perpetrators are described more as passersby, suggesting weak relationships. Nevertheless, weaker ties between perpetrators and victims in online contexts may influence the strength of the effects observed here.

Conclusions

How widespread is the belief that “mean words on the Internet do not hurt anyone?” The answer likely depends on the part of the Internet to which one refers. Reactions vary across environments that differ in anonymity and the presence of social data such as profile pictures. However, even when people are identifiable, we find initial evidence that inflammatory speech is less shocking in digital contexts. This may be concerning as social interactions move increasingly into the digital realm, potentially causing the same malicious behavior to evoke less outrage and to be viewed as more acceptable over time. The United States is currently led by a president with a history of attacking others via social media in perhaps the highest profile example of societies’ willingness to look past malicious behavior online. The present research does, however, point to strategies (i.e., supportive norms and humanizing information) that may prevent users from trivializing malicious behavior in digital contexts.

Supplemental Material

Supplemental Material, SPPS806350_suppl_mat - Inflammatory Comments Elicit Less Outrage When Made in Anonymous Online Contexts

Supplemental Material, SPPS806350_suppl_mat for Inflammatory Comments Elicit Less Outrage When Made in Anonymous Online Contexts by Curtis Puryear and Joseph A. Vandello in Social Psychological and Personality Science

Footnotes

Authors’ Note

The authors thank Jesse Graham and an anonymous reviewer for their contributions to the design and rationale for Study 4.

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

Supplemental Material

The supplemental material is available in the online version of the article.

References

Bargh

J. A.

McKenna

K. Y.

(2004). The Internet and social life. Annual Review of Psychology, 55, 573–590. doi:10.1146/annurev.psych.55.090902.141922

Crockett

M. J.

(2017). Moral outrage in the digital age. Nature Human Behaviour, 1, 769–771. doi:10.1038/s41562-017-0213-3

Daft

R. L.

Lengel

R. H.

(1983). Information richness. A new approach to managerial behavior and organization design. In Staw

B. M.

Cummings

L. L.

(Eds.), Research in organizational behavior (Vol. 6, pp. 191–233). Greenwich, CT: JAI.

Flores

James

(2012). Morality and ethics behind the screen: Young people’s perspectives on digital life. New Media & Society, 15, 834–852. doi:10.1177/1461444812462842

Fox

Tang

W. Y.

(2014). Sexism in online video games: The role of conformity to masculine norms and social dominance orientation. Computers in Human Behavior, 33, 314–320. doi:10.1016/j.chb.2013.07.014

Gibb

Z. G.

Devereux

P. G.

(2016). Missing link: Exploring repetition and intentionality of distress in cyberbullying behaviors within a college population. Translational Issues in Psychological Science, 2, 313–322. doi:10.1037/tps0000083

Gray

Waytz

Young

(2012). The moral dyad: A fundamental template unifying moral judgment. Psychological Inquiry, 23, 206–215. doi:10.1080/1047840x.2012.686247

Gray

Wegner

D. M.

(2009). Moral typecasting: Divergent perceptions of moral agents and moral patients. Journal of Personality and Social Psychology, 96, 505–520. doi:10.1037/a0013748

Hayes

A. F.

(2013). Introduction to mediation, moderation, and conditional process analysis: A regression-based approach. New York, NY: The Guilford Press.

10.

Hmielowski

J. D.

Hutchens

M. J.

Cicchirillo

V. J.

(2014). Living in an age of online incivility: Examining the conditional indirect effects of online discussion on political flaming. Information, Communication & Society, 17, 1196–1211. doi:10.1080/1369118x.2014.899609

11.

Inbar

Pizarro

D. A.

Cushman

(2012). Benefiting from misfortune: When harmless actions are judged to be morally blameworthy. Personality and Social Psychology Bulletin, 38, 52–62.

12.

Kelley

H. H.

(1973). The processes of causal attribution. American psychologist, 28, 107–128.

13.

Kiesler

Siegel

McGuire

T. W.

(1984). Social psychological aspects of computer-mediated communication. American Psychologist, 39, 1123–1134. doi:10.1037/0003-066x.39.10.1123

14.

Lee

(2005). Behavioral strategies for dealing with flaming in an online forum. The Sociological Quarterly, 46, 385–403. doi:10.1111/j.1533-8525.2005.00017.x

15.

Lindström

Jangard

Selbing

Olsson

(2017). The role of a “common is moral” heuristic in the stability and change of moral norms. Journal of Experimental Psychology: General. Advance online publication. doi:10.1037/xge0000365

16.

Lysonski

Durvasula

(2008). Digital piracy of MP3s: Consumer and ethical predispositions. Journal of Consumer Marketing, 25, 167–178. doi:10.1108/07363760810870662

17.

McCall

(2013). Social cognition in the cyborg age: Embodiment and the internet. Psychological Inquiry, 24, 314–320. doi:10.1080/1047840x.2013.839921

18.

Moor

P. J.

Heuvelman

Verleur

(2010). Flaming on YouTube. Computers in Human Behavior, 26, 1536–1546. doi:10.1016/j.chb.2010.05.023

19.

Poole

(2007). A study of beliefs and behaviors regarding digital technology. New Media & Society, 9, 771–793. doi:10.1177/1461444807080341

20.

Postmes

Spears

Lea

(2000). The formation of group norms in computer-mediated communication. Human Communication Research, 26, 341–371. doi:10.1111/j.1468-2958.2000.tb00761.x

21.

Richard

F. D.

Bond

C. F.

Stokes-Zoota

J. J.

(2003). One hundred years of social psychology quantitatively described. Review of General Psychology, 7, 331–363. doi:10.1037/1089-2680.7.4.331

22.

Shifman

Lemish

(2010). “Mars and Venus” in virtual space: Post-feminist humor and the Internet. Critical Studies in Media Communication, 28, 253–273. doi:10.1080/15295036.2010.522589

23.

Short

Williams

Christie

(1976). The social psychology of telecommunications. London, England: Wiley.

24.

Sparrow

Chatman

(2013). Social cognition in the Internet age: Same as it ever was? Psychological Inquiry, 24, 273–292. doi:10.1080/1047840x.2013.827079

25.

Suler

(2004). The online disinhibition effect. Cyberpsychology & Behavior, 7, 321–326.

26.

Tetlock

P. E.

Kristel

O. V.

Elson

S. B.

Green

M. C.

Lerner

J. S.

(2000). The psychology of the unthinkable: Taboo trade-offs, forbidden base rates, and heretical counterfactuals. Journal of Personality and Social Psychology, 78, 853–870. doi:10.1037/0022-3514.78.5.853

27.

Uhlmann

E. L.

Pizarro

D. A.

Diermeier

(2015). A person-centered approach to moral judgment. Perspectives on Psychological Science, 10, 72–81.

28.

Wong

D. L.

Baker

C. M.

(1988). Pain in children: Comparison of assessment scales. Pediatric Nursing, 14, 9–17.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

5.32 MB