Ambiguity and Ambivalence: Revisiting Online Disinhibition Effect and Social Information Processing Theory in 4chan’s /g/ and /pol/ Boards
Numerous communication theories have been established to articulate key differences between computer-mediated communication (CMC) and traditional face-to-face (FtF) interactions in the physical world. Although many aspects of communication remain the same in online settings, people do behave differently once no longer speaking FtF. Social Information Processing (SIP) theory holds that CMC is limited due to the lack of communication cues available to a conversation’s participants. Online Disinhibition Effect (ODE) further predicts that due to the presence of anonymity in CMC, individuals will be more willing to reveal their emotions and beliefs in online settings. This study uses a series of posts from 4chan’s /g/ and /pol/ imageboards to revisit SIP and ODE and examine the extent to which they can apply universally to all CMC. Overall, the study found that even though anonymity is held constant between the two online spaces, there are significant differences in the overall sentiment of communication. These findings suggest that while aspects of SIP and ODE ring true, they cannot be applied equally to all CMC settings. Instead, this study considers differences between online communities and the inherent ambivalence that is present when attempting to study the motives behind online communication.
Keywords: 4chan, computer-mediated communication, online disinhibition, social information processing theory
Whenever you enter into a public space, some uncertainty is always inevitable. There could be any number of other people there as well, doing any number of possible actions, and saying any number of possible things. In these public settings you can never be quite sure of what you may inadvertently overhear in passing. But despite this uncertainty, there are some things that would certainly cause a great deal of shock to hear.
“why would you dress like that? literally asking to be raped or badly beaten.”
“For Pete's sake...ENOUGH with these Jewish fellas already. Take care of America. It's been like three years. like...come on...this is gay.”
I think, and sincerely hope, that it is not too radical of a position to assume that if you were to overhear any one of these utterances in some public space, you would react with some combination of shock, disgust, and disbelief. These kinds of statements are completely reprehensible and would be completely out of place given the expectations of civility and basic human decency. But such standards may only apply in face-to-face (FtF) settings. Our expectations for civility and decency are often different in Computer Mediated Communication (CMC) contexts. The above quotations were not spoken aloud in an FtF setting, but rather are sourced from anonymous 4chan comments. In their original settings, the /g/ and /pol/ boards respectively, this type of language would be considered much more normative. Misogyny, racism, and anti-Semitism certainly exist in both CMC and FtF settings, but these two examples underscore how they can be much more widespread and rampant in certain online spaces.
To suggest that CMC is different from FtF communication is hardly a new argument, and there have been numerous studies that have mapped some of the primary distinctions (For instance: Bagozzi & Dholakia, 2002; Jarvenpaa & Leidner, 1999; Joinson, 2001; Parks & Floyd, 1996). Additionally, specific theories have been developed to explain why people may be more willing to communicate differently in online settings. Social Information Processing (SIP) theory posits that due to the lack of non-verbal cues in CMC, individuals are required to adapt and work within significant limitations in order to communicate effectively (Walther, 1992, 1996). SIP suggests that CMC is similar to FtF, but relationships may progress at a slower rate. Applied to the above examples, it may be the case that each anonymous user was purposely trying to be provocative in order to stand out more, and to catch the attention of passers-by. Of course when studying online communication, determining intent is really an exercise in futility (Phillips, 2015; Phillips & Milner, 2017).
So it is also equally possible that these individuals genuinely hold these hateful and bigoted viewpoints, and feel empowered to express these views due to the affordances of CMC. Online Disinhibition Effect (ODE) is a theory that describes this behavior (Suler, 2004). According to ODE, because the characteristics of CMC are different from that of FtF communication, individuals inherently behave differently. Specifically, the fact that CMC allows for increased anonymity and the reduction of individual consequences, people are more willing to express themselves and their viewpoints online than they are offline.
Taken together, SIP and ODE are useful communication theories for conceptualizing CMC and the ways that human interaction changes once it moves beyond FtF settings. However, both SIP and ODE tend to lump all of CMC together, and do not wholly account for possible differences between separate online spaces and communities. The two theories are too quick to ascribe the general distinctions between CMC and FtF to the communication technology itself, and overlook the role of the individual persons in the communication process. While it certainly is the case that CMC has unique technological characteristics from FtF communication settings, the role of individual people and their communities cannot be overlooked. It is in this spirit that I turn to the /g/ and /pol/ 4chan imageboards and their communities as a means to revisit these two communication theories and reconsider the role of the individual and distinct community standards.
It is critical for communication scholars to challenge the monolithic view of CMC and account for differences in different online communication platforms and spaces. Since the development of these two theories, the Internet has become increasingly intertwined within nearly everyone’s daily lives. Whereas our email inboxes used to be something we might check once per day, many of us now carry small devices in our pockets that can notify us immediately when there is a new message demanding our attention. For many people, it may be difficult—or even impossible—to imagine participating in modern society while also being wholly disconnected from the Internet. CMC is unavoidable.
But while CMC may be largely unavoidable, it would be folly to suggest that the Internet means the same thing to every one of its users. Not every person uses the Internet in the same ways, and not all online communities and spaces operate in identical fashions. This may seem like an obvious statement to make, but it is nevertheless an important reminder as we continue to apply communication theories to CMC. Yes, some people become absolutely nasty in the ways that they interact and behave online. But to ascribe this nastiness to the anonymity of the technology completely misses part of the story, such as the reality that some people are just unpleasant–no matter the communication setting. To suggest that the anonymity of CMC makes people more willing to express hateful views than in FtF settings also overlooks the countless individuals who actually benefit from the possibilities of CMC. Simply put, SIP and ODE reproduce many of the shortcomings of technological determinism, the belief that media technology is the primary driving factor in human behavior. However, I believe that a social shaping lens may be more appropriate and lead to richer communication theories that acknowledge the co-construction and co-influence of both media technology and individual human behavior upon one another.
In this paper, I use a sampling of threads from the /g/ and /pol/ 4chan imageboards to revisit SIP and ODE through a social shaping lens. I argue that although CMC does differ from FtF communication in many ways, ascribing these differences only to anonymity and the technology itself overlooks the human component of CMC. SIP and ODE are strong communication theories, but can be further enhanced through more consideration of the differences between distinct online communities. After reviewing the literature, I examine the dataset of /g/ and /pol/ threads scraped between April 1 and April 8, 2019. The front page posts were analyzed using natural language processing sentiment analysis, which was complemented by a close reading the full contents of 20 randomly selected threads from each board. Through this analysis, I demonstrate that despite anonymity being present in both online spaces, there are still many differences in the type of communication that takes place in each. Because of this, theories such as ODE and SIP that attempt to predict all online communicative behavior may be limited in their applicability. There are different community standards and practices in all online spaces, and this ambiguity and ambivalence may limit the universality of communication theories that work in online spaces (Phillips & Milner, 2017).
There is a widespread tendency to consider the Internet as some kind of mythical space that exists wholly separated from the physical world, and in response some scholars have pointed out the shortcomings of this mythical conception of cyberspace, and remind us that the virtual world is still situated within the “real” world (Robins, 1995). But while online and offline spaces may indeed be deeply interconnected, the fact remains that CMC is significantly different than interpersonal communication in embodied space. Internet-based communication is often marked with greater levels of vitriol, hostility, and other abrasive characteristics (Phillips, 2015). There is certainly no shortage of scholarship that examines CMC in the 21st century and identifies ways that it different from, and in many cases is more negative than, FtF communication (Crosset, Tanner, & Campana, 2018; N. Douglas, 2014; Hine et al., 2017; Waterloo, Baumgartner, Peter, & Valkenburg, 2018). However, it would be inaccurate to write this off as a relatively new phenomenon. The notion that online communication could be more toxic and negative than face to face communication is not limited to just the last decade (Dibbell, 1996; Machado, 1996).
Although there has been significant emphasis on CMC that is more hostile and negative than FtF communication, it is important to note that this is not necessarily universal for all cases. There are many examples of opportunities afforded by CMC that would not be possible or as accessible in embodied settings (Al-Qasimi, 2011; Ellison, Steinfield, & Lampe, 2007; Gal, Shifman, & Kampf, 2016; Ging & Garvey, 2018; Zhao, Grasmuck, & Martin, 2008). Stating that all CMC is negative is an overgeneralization, and overlooks the rich body of literature that suggests the important role that online communication can have for many individuals and groups. However, the fact remains that CMC has different characteristics than FtF communication.
But while there is general sense that online communication has different characteristics, there is still opportunity to pinpoint some of the specific reasons that people behave differently when they are no longer communicating face to face. The question of why online communication is different has been explored in both academic and non-academic settings. One webcomic, Penny Arcade (2004) has described this with the “Greater Internet Fuckwad Theory” (GIFT), which describes the nature of online communication as: “Normal Person + Anonymity + Audience = Total Fuckwad.” In more academic terms, Suler’s (2004) Online Disinhibition Effect (ODE) describes several factors, including anonymity and the lack of consequences, that cause individuals to communicate differently in online settings than they would in person. Similarly, Social Information Processing (SIP) theory posits that because fewer communication cues are available in CMC, it is necessarily less rich than embodied communication. However, both ODE and SIP discuss CMC in very broad strokes, and generalize all online communication as behaving in a similar fashion. Leaning on just ODE and SIP, one is lead to believe that all CMC is essentially the same, and differs from FtF communication because of anonymity and the availability of communication cues.
SIP and ODE
SIP describes that despite computer-mediated communication (CMC) being more limited than FtF communication, users nonetheless can adapt to the medium’s restrictions and develop close relationships (Walther, 1992). Griffin (2012) describes the original definition of SIP “as goes FtF communication, so goes CMC” (139). However, the theory nevertheless holds that CMC lacks the nonverbal cues that are typically present in FtF settings, thus necessitating the use of linguistic cues to accommodate this lack of cues. However, the theory as first initially proposed fails to acknowledge the possibility that while many nonverbal cues may be missing, CMC may still have nonverbal cues that influence the communicative act. For instance, the specific online location certainly influences the norms and expectations.
To its credit, SIP has evolved significantly, often alongside the development of new CMC platforms and technologies, including the proliferation of visual communication online as well as social networking services (Tidwell & Walther, 2002; Walther, 1996; Walther & Bunz, 2005; Walther, Van Der Heide, Hamel, & Shulman, 2009). While the theory has developed to acknowledge cues beyond the nonverbal cues of FtF communication, SIP remains primarily focused on the presence (or absence) of specific cues in communication contexts. In other words, it still largely emphasizes that CMC affords its parties with less information at once. There is still the possibility that CMC simply offers different cues that influence our ability to effectively communicate.
Suler’s ODE theory offers another possible explanation for how CMC is different than FtF communication. However, unlike SIP this theory specifically engages with why individuals are more willing to readily reveal their opinions and emotions in CMC as opposed to FtF. Suler offers anonymity and invisibility as two primary factors that lead people to “say and do things in cyberspace that they wouldn’t ordinarily say and do in the face-to-face world” (2004, p. 321). ODE fits well alongside SIP because it offers additional communicative cues that are different between CMC and FtF. Indeed, ODE includes six distinct characteristics that differentiate CMC, though anonymity generally emerges as the primary reason that it differs from FtF. However, ODE overemphasizes the connection between anonymity and negative online behavior. It leaves out the possibility for anonymity to actually incentivize increased cordiality, as well as the possibility that even when anonymity is present, there may be other factors that influence the nature of CMC.
Furthermore, both SIP and ODE overgeneralize their theories as essential characteristics of CMC, and suggest that they will hold true in all online communication settings. While Suler (2004) at least offers some acknowledgement of the myriad CMC contexts, there is still significant research opportunities to consider differences between specific online platforms. Furthermore, there remains the possibility that even within the same platform, CMC is not wholly homogenized. Although this paper only considers a single online platform, 4chan, it may serve as a pilot for future work on other platforms to consider factors beyond anonymity and lack of cues that affect CMC.
By putting these two theories in conversation with one another, there is opportunity to complicate the potential factors that contribute to differences between CMC and FtF communication. I acknowledge that anonymity and the availability of cues do contribute to CMC being different, but suggest that only considering this small handful of factors actually does a disservice to communication research. These theories can be strengthened by considering other factors, such as the ambiguity and ambivalence of many online communication settings. Furthermore, I argue that these theories are necessarily limited in their generalizability and ability to describe all CMC; even within the same online platform, there may be significant differences in how negative the communication is in different contexts.
4chan - Doing Dirt Work Online
4chan is a website built around the bulletin board model typical of many earlier websites and CMC systems. It is organized into several distinct imageboards, each named for a specific topic such as /a/ Anime & Manga, /tv/ Television & Film, or /biz/ Business & Finance. Though each of these boards has its own distinct community, the general model remains the same throughout the website. A thread is started by an “original poster” which includes a single image. The first post in a thread must include an image, but this is optional in subsequent replies. It is also possible to link to previous posts, quote text, and set a custom user name–although it is most common for users to post as “Anonymous.” Despite the distinct boards and unique communities, the entire 4chan website is often most well known for its “rhetoric of hate and racism” (Hine et al., 2017, p. 1). While this is certainly true in many instances, it is a bit of a stretch to apply to label to all of 4chan’s boards, and to extend the characterization to all CMC as hostile and vitriolic. Instead, I argue that it is more appropriate to consider each instance of CMC separately and consider the frames of ambiguity and ambivalence to complicate the universal aspects of ODE and SIP.
I draw upon Phillips and Milner’s (2017) definition of ambiguity and ambivalence in the context of CMC to revisit ODE and SIP. Ambiguity refers to the general unknowability of context and intention in online communication settings, while ambivalence refers to the broader underlying tension between users online. In online communication, the meaning of any given message is continually negotiated by both its sender and receiver, and this negotiation often can never fully be resolved. For example, “trolling” has become an imprecise catch-all term that may represent anything from playful joking to actual malicious intent (Phillips, 2015; Phillips & Milner, 2017). Although ODE and SIP attempt to identify how individuals communicate differently in online spaces, they overlook the role of ambiguity and ambivalence. As Phillips and Milner (2017) note, individuals rarely fit cleanly into one category or another, and in fact may often move back and forth between different frame. The same people that are “total fuckwads” online may be perfectly friendly when offline, and vice versa.
I reframe ODE and SIP within the context of ambiguity and ambivalence, and use the theories in combination to examine a selection of posts and comments from 4chan’s /pol/ “Politically Incorrect” and /g/ “Technology” boards. 4chan, an imageboard website, is widely considered to be part of “the Internet’s dark underbelly” and /pol/ specifically is a site where Online Disinhibition is readily apparent (Hine et al., 2017). However, 4chan has many boards beyond just /pol/, each of which has its own distinct community. /g/, therefore, is a useful comparison because it exists on the same platform, and thus the same affordances, but represents an entirely different community and communication context.
4chan threads might be described by Brunton (2017) as digital middens, accumulations of junk and trash that were not intended to be archived. But as Brunton (2017) notes, it is often the scraps that are meant to be thrown away that can be the most enlightening about a group’s behaviors and values. By this logic, 4chan can still be a useful object of study for understanding CMC. 4chan is often considered to be indicative of the “weird” part of the Internet, and may be overlooked for this reason (Phillips & Milner, 2017). However taking the time to examine what is weird or taboo is still significant, because these characteristics are typically established vis-à-vis what is normative (M. Douglas, 1966). While I am cautious to avoid overgeneralizing the characteristics of /pol/ and /g/ to the Internet at large, they may nevertheless be useful case studies to complicate the assumptions of SIP and ODE by suggesting other possible differences between CMC and FtF.
As mentioned briefly above, both /g/ and /pol/ are separate 4chan imageboards, with different audiences and intended uses. While /g/ is meant for general discussions about technology, /pol/ engages directly with politics and world issues. A cursory glance at the front page of both boards is enough to surmise that the communication on /pol/ is more hostile and aggressive than that of /g/. To better understand the differences, I pose the following research question:
RQ: What is the general sentiment of communication on the /g/ and /pol/ 4chan boards?
However, because both boards exist on the 4chan platform, they contain the same affordances. There are some minor differences, such as the inclusion of country flags and unique poster tripcodes on /pol/ threads. While these do provide a small amount of identification, they are more useful for estimating the approximate number of unique users in a given thread but are not typically used for the identification of specific users. Indeed on both /g/ and /pol/, the majority of users choose to post anonymously. Thus, the two 4chan boards provide useful opportunity to compare differences in communication, even when anonymity is held constant. The /g/ and /pol/ communities are different from each other which may account for differences in communication sentiment more so than the =characteristic of anonymity. With this in mind, I offer the following hypothesis:
H1: There is no correlation between anonymity and differences in communication sentiment on the /g/ and /pol/ 4chan boards.
Data were collected via an automated Python script that scraped the front page content of both /g/ and /pol/. For a period of one week (April 1 – April 8, 2019), the front page content was scraped every hour and saved into an individual comma-separated variable (CSV) file on a Google Cloud Compute Virtual Machine. The April 1-8 timeframe was chosen arbitrarily, and does not represent a specific event on either board. The purpose was to capture a general snapshot of each board’s typical posting behavior, and the week timeframe was chosen given considerations of research time and computational limitations. Additionally, the content of each individual front page thread was saved as a separate CSV file. Reply text was saved, though due to technical and storage limitations any image attachments were not retained. The script also collected general metadata, including the date and timestamp of each thread and reply.
After data collection, general information about each board and community was compiled. To assess the overall sentiment of communication on each board, I consider the average number of swear words per top-level post to provide a general evaluation of the thread’s language. To complement this, I ran the raw text of each board’s top level threads through the Google Cloud Natural Language sentiment analysis API to generate a score ranging from -1 (very negative) to 1 (very positive) as well as its magnitude, the strength of each score. Taken together, the score and magnitude can be used to compare texts from multiple sources (Google Cloud, 2018).
Sentiment analysis is a method to rapidly analyze large sets of text to determine its overall sentiment. Although this is task that human readers can perform accurately, it can be significantly more difficult to use computer-based natural language processing (Nasukawa & Yi, 2003). One of the primary challenges of sentiment analysis is ensuring that the machine learning algorithm is provided with a sufficient set of human-selected training data to fine tune its accuracy. This influenced the decision to utilize the Google Cloud Natural Language API, as its connection to the larger suite of Google products as well as widespread usage may make it more reliable than a small-scale solution. Despite the challenges of accurate sentiment analysis, the method has been used to assess consumer opinions (Pang & Lee, 2008) as well has been successful utilized to classify text from the online microblogging platform Twitter (Patodkar & I.R, 2016). However, given that I was using Google Cloud’s own API and training data rather than sample data specifically from 4chan, there was the potential that the automated sentiment analysis may have limited reliability. Thus, to compare the front page threads from /g/ and /pol/ I did not rely on the sentiment analysis alone.
After compiling basic statistics and computing the sentiment and magnitude of each front page thread, I randomly selected 20 threads from both /g/ and /pol/ to analyze. For these threads, I performed close readings and textual analyses of each reply within the thread. I took notes on specific words and phrases that stood out, as well as the broad communication characteristics within the thread. As I discuss later, even though both /g/ and /pol/ have a distinct overall community culture that is evident across the board there are still significant differences between individual threads within the same board. By performing this qualitative analysis, I was able to more thoroughly describe the communication sentiment of each board; though two threads may have similar computed language sentiment scores, it is only through close readings that characteristics such as the difference between complaining about a consumer product on /g/ and outright anti-Semitism on /pol/ become evident. This type of mixed methods research approach builds on the strengths of both qualitative and quantitative research, while also minimizing the potential limitations of each (Creswell & Creswell, 2018, p. 216).
Although 4chan is a public website and does not require registration to use, I still recognize the importance of protecting individual users and their privacy. The general guidelines of the Association of Internet Researchers state “The greater the vulnerability of the community / author / participant, the greater the obligation of the researcher to protect the community / author / participant” (2012, p. 4). With this in mind, I assessed that the population of 4chan users was not vulnerable, and have chosen to include direct quotations from selected 4chan posts and replies whem appropriate. Most users do post anonymously, but may include some personally identifying information elsewhere in their text. Any such instances were not directly quoted within this paper. Although 4chan does have an implied ephemerality, there are several 4chan archive websites, and it is not uncommon for 4chan users to post screenshots of previous threads to the imageboards. This, combined with the public nature of 4chan, informed my decision to work directly with 4chan content, and include direct quotations as appropriate. However, I am also mindful that the decision to reproduce hateful language, even for the purposes of critique in an academic setting, can still contribute to its spread and amplification (Phillips, 2018). Thus, when choosing to directly quote from the 4chan dataset, I only do so when the words directly relate to my argument and err on the side of paraphrasing rather than reproducing certain instances of hateful and harmful language.
Between April 1 and April 8, a total of 3,941 threads were scraped from the /g/ and /pol/ boards. On /g/, 97.60% of threads were posted anonymously, and on /pol/, 98.02% of threads were posted anonymously. The language sentiment and magnitude scores for each thread was retrieved via the Google Cloud Natural Language API, and basic statistics were calculated for each imageboard.
On both /g/ and /pol/, the general characteristics and sentiment of communication differed from what would typically be considered appropriate in FtF settings. Whereas an overall sentiment score of 0 would represent truly neutral communication, the general sentiment of the threads tended to be mostly negative, and the use of expletives was fairly common. However, there are still notable differences between the two boards. First, the overall sentiment score of /pol/ was much lower than that of /g/. Second, there was a difference in volume of threads–with nearly twice as many threads on /pol/ than /g/ within the same timeframe. Even taking the volume discrepancy into consideration, there is still a notable difference in the use of expletives, as shown in Table 2. While 3.39% of scraped /pol/ posts contained the N-word, compared to only 0.68% of /g/ posts. There use of other expletives on /pol/ were more common than on /g/, but the difference was less extreme: 10.21% compared to 8.65% for the word “fuck” and 3.25% compared to 2.23% for the word “fag.” The word “shit” was used more frequently on /g/ than /pol/, 8.48% compared to 7.28%.
An independent samples t-test was performed to compare the overall language sentiment between /g/ and /pol/ front page threads. There was a significant difference in the sentiment analysis scores for /g/ (M = -0.0803, SD = 0.3713) and /pol/ (M = -.01513, SD = 0.4258) front page threads; t(2493) = -5.2343, p < 0.001.
Based on the sentiment analysis scores alone, the data show that the overall language sentiment on /g/ and /pol/ is different, even though anonymity is present in both online settings. This demonstrates that not all online communication share similar characteristics, and that any differences in the overall communication sentiment may be correlated with factors other than the presence of anonymity.
Close Reading of 4chan Threads
To better understand the other factors that may be at play, the quantitative sentiment analysis was followed up with close reading of threads from both boards. On both /g/ and /pol/, 20 threads were randomly selected from the dataset. For each thread, basic information was compiled including the total number of replies, the number of replies that quoted a previous reply, and any users not posting anonymously. Then, I performed a close reading of the entire thread and its individual replies, making note of the thread’s general topic, overall sentiment and characteristics of communication, and any noteworthy quotes.
Before examining the 20 randomly selected threads from each board, I examined the threads that were “pinned” to the top of /g/ and /pol/. Pinned threads are set by one of the board’s administrators to always appear at the top of the front page, regardless of their age. Additionally, pinned threads are locked for commenting and are therefore often used to explain the basic rules and expectations for each board. Pinned posts, then, are a key location for the articulation of each community’s ethos and contribute to the overall tone of the communication taking place in each online space.
On /g/, there is one pinned post that explicitly states what the community is meant for. “/g/ is for the discussion of technology and related topics. /g/ is NOT your personal tech support team or personal consumer review site.” /g/ also includes a link to the “Install Gentoo Wiki,” a separate website that contains specific resources and information from previous /g/ threads, as well as explanations of the board community’s in-jokes (“InstallGentoo Wiki,” 2019). Though there is not an overt statement of the type of communication that is expected or allowed on /g/ threads, the pinned post nonetheless influences the board’s overall communication sentiment by framing the topics that the community is meant for. But despite these clearly framed topics, on /g/ there are still numerous examples of threads that are only tangentially related to technology as well as the use of hateful language. Although the pinned post defines the board and its community to a certain extent, it should not be considered singly.
Similar to /g/, the /pol/ pinned posts outline the expected topics for the board and defines the type of community that the administrators and moderates hope to foster. /pol/ has two separate pinned posts, both of which emphasize that the /pol/ community is intended for a specific topics–discussion of politics–and a specific type of discourse. “If your thread is not specifically about politics, then it does not belong on /pol/.” Furthermore, /pol/’s pinned posts attempt to differentiate its community from other 4chan boards and online spaces. “Off-topic and /b/-tier threads will be deleted (and possibly earn you a ban, if you persist).” But despite these pinned posts which define /pol/ as a place for a high level of discourse, much of the threads I examined contained hostility, hate, and other harmful language—hardly high-level discourse.
In this sense, the articulation of community standards and norms through the 4chan pinned post is inherently ambivalent. These posts are places where the community defines itself and its ethos, except for the times where they are not and the board content deviates from these norms. They’re a serious articulation of the rules and expectations, but do not necessarily matter in every situation. It is because of this ambivalence, and ambiguity over the “seriousness” of rules, that makes it difficult for any communication theory to universally predict CMC behaviors. A user could point to rules to call out another user’s content as being off-topic or inappropriate, but in the very next post critique the rules as not actually mattering.
One the one hand, ODE might explain this behavior through the fact that the rules have little enforceability; even if a user receives an IP ban for breaking the rules, it is a trivial matter to use a proxy or VPN service to circumvent the ban. In short, the lack of repercussions is what leads to the toxic behavior online. SIP, on the other hand, might attribute the general ignorance of the rules to the lack of communication cues. For instance, even when administrators or moderators enforce rules, there may be little visible trace of their actions. When a user is banned, there is no large announcement of their removal; Similarly a user could easily reappear under a new name and new IP address, and there would be no way to tell. However, it is not simply the case that 4chan has less communication cues for individuals to read, but rather that there are just different cues than in FtF communication. For instance, any post that contains questionably relevant content, or negative language sentiment, serves as a cue to other users that that type of behavior actually falls within the scope of the rules and the board community. However, if such a post is met with negative responses it may be a cue that such behavior is actually not acceptable. Each specific online community, therefore, is responsible for defining the type of acceptable communication and influences the overall sentiment of a given online space.
Language sentiment of selected /g/ threads.
From the 20 randomly selected threads from each board, the overall quality of communication on /g/ appears to be nicer and more conversational. In general, responses to a thread seem to come from a place of genuinely wanting to answer a question, or share expertise. However, this is not to say that the /g/ board is entirely devoid of nastiness or language that would be entirely out of place in FtF settings. For instance, it is not uncommon for a thread reply to include sexist, racist, or other tasteless language.
One /g/ thread is dedicated to discussing and sharing sources of cheap electronics and other products from overseas. Based on the widespread belief that China is a leading supplier of cheap consumer guides, this thread is labeled using a racial slur as the “Chink Shit General” thread. Throughout the thread, anonymous users freely use the words “chink” and “chinked” to describe their experiences ordering products from Chinese suppliers. One /g/ user asked, “Are these space blankets supposed to be slightly transparent or have I been chinked?” Other users avoided using the racial slur, but still generalized just a few experiences to all Chinese goods. Another /g/ users posted, “Chinese electronics work great and are cheap but they always always fucking break after a ridiculously short amount of time. Have had this happen with so many cables, adapters and a humidifier.”
This particular thread was difficult for me to work through. In many ways it highlights the inherent ambivalence of online communication. Would any of these users actually say such things in FtF communication? There is simply no way to tell. The thread is a valuable online discussion about ordering goods from international sources, and indeed does include useful information, such as details about how to work with customs and delivery logistics. But simultaneously, truly harmful language is tossed around casually seemingly without a second thought. These instances of negative language seem to be accepted by the /g/ community, and not called into question, which can contribute to the normalization of such racial slurs. While it is entirely possible that the use of such language really is truly “just a joke,” there is no way to determine a 4chan user’s actual intent, nor is it possible to deny that even if such words are used in a joking fashion that they can have real negative impacts. This is the inherent crux of online ambivalence.
That said, however, the quality and overall sentiment of the language on the /g/ threads was notably different than what appeared in the /pol/ threads. Despite the somewhat common presence of slurs, sexism, and racism, /g/ threads were noticeably less hostile and confrontational than /pol/ threads. Furthermore, /pol/ posters were typically much more open and direct with their racism and sexism, often using their hurtful language to directly target actual people.
Language sentiment of selected /pol/ threads.
On /pol/, there is much more common and much more overt instances of misogyny, racism, and anti-Semitism. One thing to note is that there tended to be a difference between the first few replies to a post and the ones that came much later. Often, the original post and the initial replies set the tone of the discourse. For example, in one thread the original poster posted a link to an article, and posed a genuine question about whether or not President Trump would really shut down the U.S.-Mexico border. The first few replies are real responses, which the users’ genuine thoughts, opinions, and reasoning. One /pol/ user wrote, “He threatened to send the national guard and declare national emergency about a week before he did both of those things. So idk [sic], it’s possible.” However, as the thread received more replies, the quality of conversation quickly deteriorated, with many users now posting anti-Semitic comments. “this meme is getting old. Trump is a puppet of the zionist Jew faction.”
In another thread, /pol/ users discussed whether the Holocaust actually happened. The discussion was hostile and confrontational from the beginning, the original poster asking, “Can you faggots put some real evidence this event is fabricated?” Unsurprisingly, replies were posted with anti-Semitic comments and other hateful language. Outside a very small minority of posts, the thread is otherwise filled with Holocaust-deniers. This is one major difference between the /g/ and /pol/ threads I observed. On /pol/ it was much more common for outright anti-Semitism to be posted, and go unchallenged. But just as in the case of /g/, the challenge of online ambivalence is that it is nearly impossible to accurately correlate the presence of hateful language on 4chan’s /pol/ board with the genuinely-held beliefs of actual people. Though it would be really awful and tasteless humor, the excuse of “it’s just a joke” is always a possibility. To be clear, I am not suggesting that we should completely ignore /pol/’s harmful language but rather that we must exercise caution when drawing direct connections between CMC and FtF communication.
That said, the close reading of several /g/ and /pol/ complicates the comparison between communication in the two online spaces. While the quantitative sentiment scores do underscore the differences in the type of communication, it is still important to “do the dirt work” of sifting through the actual 4chan comments. There is a lot that can be hidden or swept away by a mere sentiment score, and it is our responsibility as researchers to fully consider all aspects of a communicative exchange.
RQ1 asked about the general sentiment of communication was on 4chan’s /g/ and /pol/ boards. Through the close reading of specific threads I show that not only is the sentiment of communication different between /g/ and /pol, but the ways in which it is different. /g/ is much less hostile and directly confrontational, and on /pol/ there are much more instances of overt misogyny, racism, and anti-Semitism. In both boards, the use of expletives and certain racial slurs was somewhat common–a notable difference between online communication on 4chan and FtF communication settings.
SIP theory posits that CMC does differ from FtF communication primarily due to the reduced number of communication cues that are available. Because of this, it may take longer to establish connections via CMC, and users communicating in online settings may be less willing to immediately speak their minds. However, the fact that /g/ and /pol/ have such notably different communication sentiments suggests that the lack of communication cues may not be the case in all online settings. Indeed, a more accurate statement may be simply that there are different communication cues available in CMC than in FtF communication. It is true that on 4chan you may only be able to get a “sip” of information, but there are many different places that users are able to sip from simultaneously. For instance, 4chan users may observe what types of threads often receive large numbers of replies, what types of comments are quoted most frequently, as well as what common phrases other users tend to use. These observations can be made before an individual ever posts something to 4chan. Thus, there are many possible communication cues present in online communication. CMC is indeed different from FtF, as SIP theory suggests, but to ascribe this difference wholesale to the reduced number of communication cues overlooks the potential differences in online spaces, as well as the multiple online communities that exist even within the same platform.
H1 predicted that there would be differences in the sentiment of communication between /g/ and /pol/, even when anonymity was held constant. In the dataset for both boards, the vast majority of threads and replies were posted anonymously. However, both the quantitative language sentiment analysis and the qualitative close reading demonstrated that there were significant differences in sentiment.
In addition to the different communication cues as discussed above, both /g/ and /pol/ have very different communities of users. It is possible that these are contributing factors to the difference in communication sentiment. However, the study nevertheless identifies a shortcoming in ODE, as well as in general perceptions of Internet communication. ODE suggests that anonymity is a major reason that CMC is different that FtF communication. However, this study demonstrates that there can be factors beyond just anonymity that contribute to differences in the type and quality of communication.
Furthermore, the study underscores the inherent limitation of SIP and ODE and their attempt to broadly predict the differences between CMC and FtF communication. As the comparison study of /g/ and /pol/ demonstrates, there can be significantly different forms of communication between different online communities, even when those communities exist on the same platform. While there are indeed many differences between CMC and FtF communication, and while anonymity may indeed be one of those key differences, it is still a stretch to apply that universally to all CMC. There are many different types of online spaces and online communities. By lumping them all together, we are doing ourselves a disservice and limiting both the number and types of questions that we are able to ask.
Rather than attempting to develop a theory that universally applies to all CMC, I suggest that we instead embrace the inherent ambivalence of online communication. The things that make CMC different than FtF communication are neither inherently good nor inherently bad. We should not use the harmful and negative language present on 4chan to write off all CMC as doomed. In many cases they are both, or neither. This can be incredibly frustrating to researchers, but is important to understand and embrace in order to better perform Internet-based research and examine online communities. When trying to describe ambivalence and the quality of online communication Phillips and Milner write:
“Our position is simple. We are staunch advocates of the democratic process and think that problematic speech should be countered through more speech. Except actually maybe not, because not all speech, and not all voices, are given equal weight, and that position privileges those whose voices already carry further and louder than others … So okay, we are staunch advocates of the democratic process…as our voice trails off and we stare blankly into the distance.” (Phillips & Milner, 2017).
Studying the beast that is online communication is inherently tricky, and requires confronting questions that may indeed be unanswerable. However, embracing the inherent ambivalence is an important first step beyond earlier theories that treat all CMC as identical.
This study has confirmed both ODE and SIP, as well as the general knowledge that communication that takes place on the Internet is often radically different than that which takes place in physical embodied settings. However, by revisiting these theories, this study has introduced more nuance and limitations to the universality of each. In many cases, such as in 4chan’s /g/ and /pol/ boards, there may be several instances of repulsive language including racism, sexism, and anti-Semitism. However, to extrapolate these instances as indicative of all online communication, and to ascribe these differences between CMC and FtF communication to anonymity, overlooks several other potential factors.
Online communication is inherently ambivalent, and any attempt to define a single concrete meaning is largely an exercise in futility. There are simply so many other potential factors that may contribute to this difference. By comparing the /g/ and /pol/ boards, I have demonstrated that in online settings it may be the specific community that defines the quality of communication, and not just the technology’s affordances. While the anonymity of CMC may indeed be a contributing factor, there are simply too many other potential intervening factors to lean solely on the anonymity. To apply these kinds of blanket statements to the communication technology itself rather than the actual individual people behind it shifts the responsibility and accountability away from where it ought to be. Whether in CMC or an FtF setting, it is the individual that is responsible for their words and their impact. By ascribing the differences between CMC and FtF to the mere presence of anonymity is in a sense providing a “free pass” to these individuals and the communities that they are a part of, ultimately doing more of a disservice than anything else.
Al-Qasimi, N. (2011). Ladies and gentlemen, boyahs and girls: Uploading transnational queer subjectivities in the United Arab Emirates. Circuits of Visibility: Gender and Transnational Media Cultures, 284–302.
AoIR Ethics Committee. (2012, September). Ethical decision-making and Internet research 2.0: Recommendations from the AoIR ethics working committee. Retrieved March 15, 2019, from https://aoir.org/reports/ethics2.pdf
Bagozzi, R. P., & Dholakia, U. M. (2002). Intentional social action in virtual communities. Journal of Interactive Marketing, 16(2), 2–21. doi: 10.1002/dir.10006
Brunton, F. (2017). Notes from/dev/null. Internet Histories, 1(1–2), 138–145. doi: 10.1080/24701475.2017.1307059
Creswell, J. W., & Creswell, J. D. (2018). Research Design: Qualitative, Quantitative, and Mixed Methods Approaches (Fifth edition). Los Angeles: SAGE.
Crosset, V., Tanner, S., & Campana, A. (2018). Researching far right groups on Twitter: Methodological challenges 2.0. New Media & Society, 21(4), 939–961. doi: 10.1177/1461444818817306
Dibbell, J. (1996). A Rape in Cyberspace: How an Evil Clown, a Hatian Trickster Spirt, Two Wizards, and a Cast of Dozens Turned a Database into a Society. In M. Stefik (Ed.), Internet Dreams: Archetypes, Myths, and Metaphors. Cambridge, Mass: MIT Press.
Douglas, M. (1966). Purity and Danger: An Analysis of Concepts of Pollution and Taboo. London ; New York: Routledge Classics.
Douglas, N. (2014). It’s supposed to look like shit: The Internet ugly aesthetic. Journal of Visual Culture, 13(3), 314–339.
Ellison, N. B., Steinfield, C., & Lampe, C. (2007). The Benefits of Facebook “friends:” Social Capital and College Students’ Use of Online Social Network Sites. Journal of Computer-Mediated Communication, 12(4), 1143–1168. doi: 10.1111/j.1083-6101.2007.00367.x
Gal, N., Shifman, L., & Kampf, Z. (2016). “it Gets Better”: Internet Memes and the Construction of Collective Identity. New Media & Society, 18(8), 1698–1714. doi: 10.1177/1461444814568784
Ging, D., & Garvey, S. (2018). ‘written in These Scars Are the Stories I Can’t Explain’: A Content Analysis of Pro-Ana and Thinspiration Image Sharing on Instagram. New Media & Society, 20(3), 1181–1200. doi: 10.1177/1461444816687288
Google Cloud. (2018, November 5). Natural Language API Basics. Retrieved April 27, 2019, from Google Cloud website: https://cloud.google.com/natural-language/docs/basics
Griffin, E. (2012). A First Look at Communication Theory (Eighth International Edition). McGraw-Hill.
Hine, G. E., Onaolapo, J., De Cristofaro, E., Kourtellis, N., Leontiadis, I., Samaras, R., … Blackburn, J. (2017). Kek, cucks, and god emperor trump: A measurement study of 4chan’s politically incorrect forum and its effects on the web. Eleventh International AAAI Conference on Web and Social Media.
InstallGentoo Wiki. (2019, January 20). Retrieved April 28, 2019, from https://wiki.installgentoo.com/wiki/Main_Page
Jarvenpaa, S. L., & Leidner, D. E. (1999). Communication and Trust in Global Virtual Teams. Organization Science, 10(6), 791–815. doi: 10.1287/orsc.10.6.791
Joinson, A. N. (2001). Self-disclosure in computer-mediated communication: The role of self-awareness and visual anonymity. European Journal of Social Psychology, 31(2), 177–192. doi: 10.1002/ejsp.36
Machado, J. (1996). Netiquette 101. In M. Stefik (Ed.), Internet Dreams: Archetypes, Myths, and Metaphors. Cambridge, Mass: MIT Press.
Nasukawa, T., & Yi, J. (2003). Sentiment Analysis: Capturing Favorability Using Natural Language Processing. Proceedings of the 2Nd International Conference on Knowledge Capture, 70–77. doi: 10.1145/945645.945658
Pang, B., & Lee, L. (2008). Opinion Mining and Sentiment Analysis. Foundations and Trends® in Information Retrieval, 2(1–2), 1–135. doi: 10.1561/1500000011
Parks, M. R., & Floyd, K. (1996). Making Friends in Cyberspace. Journal of Computer-Mediated Communication, 1(4). doi: 10.1111/j.1083-6101.1996.tb00176.x
Patodkar, V. N., & I.R, S. (2016). Twitter as a Corpus for Sentiment Analysis and Opinion Mining. IJARCCE, 5(12), 320–322. doi: 10.17148/IJARCCE.2016.51274
Penny Arcade. (2004, March 19). Green Blackboards (And Other Anomalies). Retrieved February 18, 2019, from Penny Arcade - Comic website: https://www.penny-arcade.com/comic/2004/03/19
Phillips, W. (2015). This Is Why We Can’t Have Nice Things: Mapping the Relationship Between Online Trolling and Mainstream Culture. Cambridge, Massachusetts: The MIT Press.
Phillips, W. (2018). The Oxygen of Amplification: Better Practices for Reporting on Extremists, Antagonists, and Manipulators (p. 128). Retrieved from Data & Society website: https://datasociety.net/output/oxygen-of-amplification/
Phillips, W., & Milner, R. M. (2017). The Ambivalent Internet: Mischief, Oddity, and Antagonism Online (1st edition). Malden, MA: Polity.
Robins, K. (1995). Cyberspace and the World We Live in. Body & Society, 1(3–4), 135–155. doi: 10.1177/1357034X95001003008
Suler, J. (2004). The Online Disinhibition Effect. CyberPsychology & Behavior, 7(3), 321–326. doi: 10.1089/1094931041291295
Tidwell, L. C., & Walther, J. B. (2002). Computer-Mediated Communication Effects on Disclosure, Impressions, and Interpersonal Evaluations: Getting to Know One Another a Bit at a Time. Human Communication Research, 28(3), 317–348. doi: 10.1111/j.1468-2958.2002.tb00811.x
Walther, J. B. (1992). Interpersonal Effects in Computer-Mediated Interaction: A Relational Perspective. Communication Research, 19(1), 52–90. doi: 10.1177/009365092019001003
Walther, J. B. (1996). Computer-Mediated Communication: Impersonal, Interpersonal, and Hyperpersonal Interaction. Communication Research, 23(1), 3–43. doi: 10.1177/009365096023001001
Walther, J. B., & Bunz, U. (2005). The Rules of Virtual Groups: Trust, Liking, and Performance in Computer-Mediated Communication. Journal of Communication, 55(4), 828–846. doi: 10.1111/j.1460-2466.2005.tb03025.x
Walther, J. B., Van Der Heide, B., Hamel, L. M., & Shulman, H. C. (2009). Self-Generated Versus Other-Generated Statements and Impressions in Computer-Mediated Communication: A Test of Warranting Theory Using Facebook. Communication Research, 36(2), 229–253. doi: 10.1177/0093650208330251
Waterloo, S. F., Baumgartner, S. E., Peter, J., & Valkenburg, P. M. (2018). Norms of Online Expressions of Emotion: Comparing Facebook, Twitter, Instagram, and Whatsapp. New Media & Society, 20(5), 1813–1831. doi: 10.1177/1461444817707349
Zhao, S., Grasmuck, S., & Martin, J. (2008). Identity Construction on Facebook: Digital Empowerment in Anchored Relationships. Computers in Human Behavior, 24(5), 1816–1836. doi: 10.1016/j.chb.2008.02.012