Unexpected expectations: Public reaction to the Facebook emotional contagion study

Abstract

How to ethically conduct online platform-based research remains an unsettled issue and the source of continued controversy. The Facebook emotional contagion study, in which researchers altered Facebook News Feeds to determine whether exposure to emotional content influences a user’s mood, has been one focal point of these discussions. The intense negative reaction by the media and public came as a surprise to those involved—but what prompted this reaction? We approach the Facebook study as a mediated controversy that reveals disconnects between how scholars, technologists, and the public understand platform-based research. We examine the controversy from the bottom up, analyzing public reactions expressed in comments on news articles. Our analysis reveals fundamental disagreements about what Facebook is and what a user’s relationship to it should be. We argue that these divergent responses emphasize the contextual nature of technology and research ethics, and conclude with a relational and contextual approach to ethical decision-making.

Keywords

Controversy analysis Facebook nonuse platform studies privacy research ethics

Introduction

The publication of “Experimental Evidence of Massive-Scale Emotional Contagion” (Kramer et al., 2014) in the Proceedings of the National Academy of Sciences (PNAS) on 2 June 2014 set the Internet ablaze. Reactions on Twitter expressed shock and outrage that Facebook was “LITERALLY playing with users’ emotions” (Altmetric, n.d.). News reports echoed and amplified public sentiment, with headlines such as: “How Facebook’s news feed controls what you see and how you feel” (Steadman, 2014), “Facebook totally screwed with a bunch of people in the name of science” (Frizell, 2014), and “So you are shocked Facebook did #psyops on people?” (Silberg, 2014). In the midst of the controversy, The Guardian conducted a reader poll where 61% of respondents reported that they were surprised to learn about the study, 84% had lost trust in the social network, and 66% were considering closing their account (Fishwick, 2014). One of the researchers involved in the study received hundreds of concerned emails from members of the public following the media attention (Hancock, 2019). As a result of the negative publicity and public reaction, both Facebook and the article’s lead author issued apologies (D’Onfro, 2014; Hiltzik, 2014) and PNAS issued a statement of editorial concern (Verma, 2014). The various apologies revealed that the negative backlash to the study came as a surprise to the researchers, the journal, and Facebook. Though the technological architecture of Facebook has long shaped possibilities for expression and social interaction, the discussion surrounding the Facebook emotional contagion (FEC) study highlighted the implications of the technological architecture for the general public and raised ethical questions about conducting research on online platforms.

But what did the study, described as “amazing scifi reading” (Altmetric, n.d.), actually entail? Conducted as a collaboration between Facebook and academic researchers, the FEC study sought to both replicate laboratory experiments and longitudinal studies on the transference of emotions, or “emotional contagion” (Fowler and Christakis, 2008; Hatfield et al., 1993; Rosenquist et al., 2011), and test the claim from prior research that repeated exposure to positive content on Facebook was making its users unhappy due to negative social comparisons (Turkle, 2011). To this end, the researchers designed and conducted an experiment on nearly 700,000 English-speaking Facebook users in which they modified users’ News Feeds, the algorithmically sorted feature that organizes and displays content generated from a user’s list of friends, according to the results of automated sentiment analysis (Pennebaker et al., 2007). One group saw a higher concentration of positive content, one group saw a higher concentration of negative content, and one group saw less emotional content of any variety. By comparing the sentiment and frequency of user posts before and after the experiment, researchers found that users exposed to higher concentrations of emotional content were slightly more likely to feature similar emotional content in their own Facebook posts for up to 3 days after exposure, and users exposed to less emotional content showed a slight decrease in engagement with the site, posting less frequently and with fewer words (Kramer et al., 2014). In short, the study offered evidence of some emotional contagion on Facebook and challenged the idea that exposure to positive content was making people sad, based on an assumption that the word choice of Facebook posts offers a reliable indicator of a person’s emotional state.

In the wake of controversies such as the FEC study and the Cambridge Analytica scandal of 2018, there has been a pronounced interest in the ethics surrounding social media research (Brown et al., 2016; Stark, 2018; Vitak et al., 2016). While issues of privacy and data use have received the most attention, the FEC study points to another important and unresolved issue—how to ethically conduct online platform-based research. The controversy that followed the publication of the FEC study provides a unique opportunity to examine responses to social computing research from members of the general public, including those who might have negative attitudes toward research or toward Facebook (e.g. Facebook nonusers). To study public reaction, we collected thousands of comments left on news articles about the FEC study. Our primary goal was to develop a deep understanding of perceptions of and attitudes toward the controversy, and by extension, research ethics for social computing platforms generally. As a result, our analysis was driven by a set of exploratory research questions: what were the patterns of public responses? What issues and considerations were most important to commenters? Simplistically, why were people so upset about this study, and what can we learn from that?

Public reactions have the potential to be an important resource for bottom-up approaches to ethical decision-making and the research ethics community generally (Nebeker et al., 2017)—especially given the prominence of normative, top-down approaches to ethical issues. However, this work faces an inherent challenge: those most negatively impacted by research and those with the most negative attitudes toward research are least likely to have their voices heard within research (Fiesler and Proferes, 2018). Studies that are interested in understanding how the public perceives and feels about research ethics typically involve deception (Hudson and Bruckman, 2004) or face an inherent selection bias toward those willing to participate in a research study (Fiesler and Proferes, 2018; Schechter and Bravo-Lillo, 2014; Williams et al., 2017). How can we take into account other relevant voices, including those that are uninterested or unwilling to participate in research? One solution is to borrow from controversy analysis (Marres, 2015; Marres and Moats, 2015) and studies of mediated public reactions (Fiesler and Hallinan, 2018; Vines et al., 2013), which is the strategy we employ in our examination of comments on news articles.

Theoretical foundations

Our analysis of public reaction to the FEC study brings together two related research traditions: (1) controversy analysis from science and technology studies and (2) expectancy violation theory (EVT) from communication. Together, these traditions provide a framework for understanding the significance of public reaction to technology controversies.

Controversy analysis establishes the value of using mediated controversies to study contested issues alongside the role of contemporary media and communication technologies (Marres, 2015; Marres and Moats, 2015), drawing attention to beliefs and values that might otherwise be overlooked or taken for granted. For example, an analysis of the Facebook Trending Topics controversy showed that news reports on the practices of the human curation team acted as a proxy for discussion about larger shifts in the news media environment (Carlson, 2018). While public expectations for Facebook typically go unstated, catalysts such as the publication of the FEC study can bring these underlying views into the foreground and reveal tensions and vulnerabilities at work in the social integration of technologies (Goodnight, 2005). In other words, mediated controversies can reveal larger tensions within the cultural positioning of technology (Satchell and Dourish, 2009).

EVT holds that individuals have expectations about the communicative behavior of others and the violation of those expectations causes people to assess their knowledge of and relationship to others (Burgoon and Le Poire, 1993). Variables that influence expectations include characteristics of the communicator, the relationship, and the context (Burgoon and Le Poire, 1993; Griffin et al., 2011). While the theory was developed in the context of interpersonal, face-to-face interactions, more recent work has extended the theory to computer-mediated contexts—for example, norms of interactions on Facebook (Bevan et al., 2014; McLaughlin and Vitak, 2012). Design choices and features of social media platforms also shape the possibilities and expectations for interaction. Previous research has examined expectations for particular features, including the Facebook Like button (Scissors et al., 2016), algorithmic curation (Bucher, 2017; Eslami et al., 2015; Rader and Gray, 2015), and design changes (Eslami et al., 2016). Together, this work demonstrates that user expectations shape assessments about the experience of social media and the desirability of particular features and practices.

Where EVT research points to the gap between knowing that expectations have been violated and knowing what those expectations are (Shklovski et al., 2014), controversy analysis prompts consideration of what large, underlying factors may be at work behind the scenes. The analysis that follows demonstrates how an understanding of expectations about platforms can contribute to ethical decision-making for researchers.

Methods

To examine public reaction, we collected and analyzed public comments on news articles about the FEC study. Analyzing the content of online news comments offers a time and resource efficient way to study public reactions (Henrich and Holmes, 2013). Previous research has used public comments to study public views on ethical and political issues related to the use of medical technologies (Chandler et al., 2017), climate change (De Kraker et al., 2014), online privacy (Fiesler and Hallinan, 2018), and even human-computer interaction (HCI) research (Vines et al., 2013). While the framing of news articles can impact comments, the FEC study was fundamentally a mediated controversy: people learned about the experiment through the publication of the research and subsequent news coverage. Therefore, it is neither possible nor desirable to separate public reaction from media coverage, since engagement with the media becomes the central site for people to analyze and understand the controversy.

As participant-driven responses, comments help reveal issues of public importance (Chandler et al., 2017; Henrich and Holmes, 2013), which is particularly important for ethics research. Comments also capture reactions and sense-making practices as they unfold and provide access to the perspectives of people who may not have social media accounts or do not use social media frequently, potentially surfacing more critical or antagonistic perspectives than user-centric social media research (Satchell and Dourish, 2009). Finally, studying public comments helps address a known limitation of ethics research: participant response bias (Fiesler and Proferes, 2018). Where surveys, interviews, and lab studies on research ethics are limited to the perspectives of those who are willing to participate in research, news comments are not subject to the same limitations. News comments provide a broader sample of online groups. Although news comments do introduce new biases—namely, people with Internet access willing to comment on news articles—they provide access to the reasoning behind different opinions. In addition, news comments are particularly impactful opinions, with previous research showing that public comments shape the views of other readers (De Kraker et al., 2014). This influence, combined with the potential to access nonusers and people uninterested in participating in research studies, makes the analysis of news comments a valuable complement to other ways of studying public reaction.

However, there are ethical considerations with respect to the collection and analysis of public data. While this is a common practice in social computing research, there are disagreements within the research community about the ethics of, for example, whether to include quotes verbatim and how—if at all—to attribute authorship of quotes (Vitak et al., 2016). Although comments are publicly available information, a best practice for making ethical decisions about the use of public data is to consider the specific context and the expectations of the people involved (Fiesler and Proferes, 2018; Nissenbaum, 2004). Arguably, comments on news sites are more “public” than some other forms of social data—that is, data from social networking sites—because comments are addressed to an audience of strangers rather than a known community of friends or followers. Commenting on a news article also indicates an interest in making one’s viewpoint known, and in the FEC study, commenters were weighing in on research ethics and the practices of social media platforms, which aligns with the context and motivations of this article. After weighing potential risks to those whose content was part of our analysis, we have decided to include quotes verbatim, without identification, which is consistent with other thematic analyses of news comments (Chandler et al., 2017; Fiesler and Hallinan, 2018; Giles et al., 2015; Glenn et al., 2012; Vines et al., 2013), and also to choose illustrative quotes that are not easily discoverable through a simple web search and that do not reveal any personal or sensitive information.

Data collection

In order to construct a dataset of public comments, we identified a set of articles starting with law professor and privacy advocate James Grimmelmann’s (2014) collection of Internet coverage about the FEC study. Because Grimmelmann’s article set included personal blog posts as well as journalist reporting, we narrowed the set into articles from news outlets that contained at least one comment, which resulted in 12 articles from that collection. Given that the earliest article on the list was published on 30 June 2014, nearly a month after the initial publication of the FEC study, we supplemented the collection with eight additional articles published prior to that date, identified using a keyword search (“Facebook + Research”) on LexisNexis and Google News for pieces published between 1 June and 30 June 2014. Our criteria for inclusion were that the article was (1) primarily about the FEC study; (2) written in English; and (3) included at least one comment; this supplemental, systematic method of adding additional articles also ensured that we included a broader set of news sources than may have been included by Grimmelmann. Our final dataset included comments from 20 articles from the following news sources: The Atlantic (3), Slate (1), Forbes (3), The New York Times (3), The Guardian (2), Wired (1), Wall Street Journal (3), The Washington Post (1), Financial Times (1), The Telegraph (1), and The Chronicle of Higher Education (1). Although this was not a criterion for inclusion, all the articles were published by organizations based in the United States and the United Kingdom. Our search uncovered a few articles published in English in other countries, but none included comments. Therefore, in addition to the limitations with news comments as a data source generally, this data may be biased toward Western voices or toward news outlets with subject matter or ideological leanings that could have influenced the decision to cover this story and our results should be interpreted with this in mind. We manually collected all available comments on the articles, including top-level comments and replies. Our final dataset consisted of 2790 total comments from 20 unique articles. The number of comments on an article ranged from 2 to 850 (M = 140; SD = 215.13; median = 42).

Data analysis

Driven by our exploratory research questions, we performed a thematic analysis (Clarke and Braun, 2006) of the data. As one of the most common approaches for studying online comments (Chandler et al., 2017; Giles et al., 2015; Holton et al., 2014; Silva, 2015; Vines et al., 2013), thematic analysis excels at revealing patterned responses in the data, especially when the analysis is concerned with meaning or explanation (Clarke and Braun, 2006). We began our analysis with the question: “What bothered people about the study?” We open coded individual comments and then developed themes inductively following the recursive steps outlined by Clarke and Braun (2006). Two of the authors met periodically to share and reconcile differences in coding, create memos, and to derive the themes discussed in the following section.

Findings

Although comment sections are notoriously antagonistic spaces, distinct patterns emerged from the thematic analysis of our data. Here, we focus on four major themes that represent public reactions, which we have labeled “Living in a lab,” “Manipulation anxieties,” “Wake up, sheeple,” and “No big deal.” Across these themes, we find divergent and contradictory understandings of Facebook as a platform, along with repeated surprise that these understandings are not universally shared. As it turns out, the researchers behind the FEC study, PNAS, and Facebook were not the only ones surprised by the reaction to the study. Some members of the public were also surprised by the expectations of their peers; in other words, there appears to be no “common” sense when it comes to social media research.

Living in a lab

The publication of the FEC study came as a surprise to some commenters who did not know that Facebook conducted experiments or collaborated with academic researchers. Their reactions were less about the specifics of testing emotional contagion and more about the revelation of experimentation as a general practice. In other words, the announcement of any experiment would violate the implicit understanding of Facebook as a place for people to connect with friends and family:

Dear Mr. Zuckerburg, Last I checked, we did not decide to jump in a petri dish to be utilized at your disposal . . . We connect with our loved ones.¹

As the reference to a petri dish suggests, the concern is with the idea of research—or “secret experiments”—taking place on online platforms. Furthermore, the concern with experimentation often conflates very different models of research, including academic research, collaborative research between academics and corporations, and applied commercial research. The tendency to conflate all forms of platform-based research into a single category is facilitated by a lack of awareness about research practices—indeed, previous research has found, for example, that nearly two-thirds of Twitter users did not know that academic researchers use public social media data (Fiesler and Proferes, 2018).

The temporal dynamics of online experiments further complicate the understanding of research on Facebook. Lab-based experiments conventionally have an obvious start and endpoint, making it clear when someone is (and is not) participating in research. With platform-based experiments, participants often have no knowledge of their own participation. In the case of the FEC study, Facebook users did not know about the experiment until it appeared in the media. Even then, people had no way of determining whether their own News Feed had been affected, despite their expressed interest—indeed, the question comes up repeatedly in our data, and one of the authors of the study received many emails with this question (Goel, 2014; Hancock, 2019). The uncertainty over participation and the lag in awareness created a sense of secrecy around research and prompted commenters to question what other kinds of experiments might be happening:

This was two years ago? Who knows what they’re doing now.

Commenters overwhelmingly characterized scientific research as negative and exploitative. Some compared the contagion study with other controversial experiments such as the Stanford prison experiment (Recuber, 2016) and Nazi medical experimentation. Others invoked the language of biomedical experiments, comparing the treatment of Facebook users with that of animal test subjects—“lab rats” or “guinea pigs”—framing scientific research as inherently dehumanizing and without benefit to the experimental subject:

At least lab rats get paid in chow. How does Facebook compensate its users to be sitting ducks for algorithms?

Even among comments defending the legitimacy of scientific research, there was little attention to any benefits, actual or potential, of the FEC study, which indicates a disconnect between the researchers’ concern with the potential negative emotional consequences of social media (Kramer et al., 2014) and the concerns expressed in public comments. The scientific value of the research and its contributions to improving user experience are not so much debated as dismissed outright; instead, comments typically frame the value of the study as serving the interests of researchers disconnected from “real world” concerns or as a proof-of-concept for the emotional exploitation of Facebook users. Where institutional decisions concerning research ethics are typically made by weighing harm and benefit, judgments from the public rarely expressed consideration for the benefit side of the equation.

These comments about “living in a lab” support the idea that some members of the public cared about the lack of transparency and consent, as well as the power dynamics between researchers and those being researched. However, concerns about experimentation were not isolated to the FEC study and instead point to discomfort with the idea of any experimentation on Facebook. Such concerns were compounded by a lack of understanding for how the research could be in service to the interests of Facebook users. As one commenter explained, the experiment demonstrated that Facebook “will pervert its stated objective of facilitating communication.” Without trust in the value of the research for Facebook users, the negative and exploitative associations of scientific research proliferated.

Manipulation anxieties

For other commenters, the FEC study was upsetting because of what the research suggested about Facebook’s powers of manipulation. While the term “manipulation” appears only once in the original publication and nowhere in the press release, it is repeated constantly in news headlines, articles, and public comments. The surprise and outrage over the “manipulation” of the News Feed suggest that many people did not realize that the News Feed selects and displays content in a particular order, or these people had assumed that content was selected according to a fair and objective standard. For example, one comment argued that Facebook “is supposed to be a neutral arbiter for its services.”

The lack of familiarity with how the News Feed works aligns with prior research (Eslami et al., 2015) and helps explain why the manipulation aspect produced such intensely negative reactions: the experiment revealed not only a single time-and-population-limited instance of manipulation, but also that manipulation is endemic to the operation of the News Feed. In other words, commenters were upset both about the experimental manipulation and about the existence of any News Feed algorithm.

Fairness is a commonly stated reason for anxiety around manipulation, tracking to findings of focus-group research on social media data mining concerns (Kennedy et al., 2015). While some commenters considered any form of manipulation to be a self-evident violation of ethics, others were worried about the specific context of manipulation on Facebook. These folks were worried that changes to the News Feed could cause them to miss out on important posts, such as an announcement of good news or a call for help:

If you were one of the friends of the almost 700,000 users, but a piece of [your] news . . . didn’t get posted . . . and this messed with your relationship to the other user? More people than just the research subject were manipulated.

From this perspective, manipulating the order of the News Feed simultaneously manipulates relationships between people that extend beyond those directly involved in the experiment. The concern over missing important content aligns with lab-based research on user reactions to revelations about the existence and operation of the News Feed algorithm (Eslami et al., 2015). However, many commenters took the concern with manipulation to more extreme ends. Our data include considerable speculation about the future implications of this research. The extrapolations were guided by examples from dystopic fiction, such as 1984 and Brave New World, and also by fears concerning politics, control, and conspiracies:

Lets see if The Algorithm can retrospectively identify the users who got the downer feeds, and when. Also those who got the happy feeds. Then there is even more useful data to be had, by medical professionals: compare the data injections against the use of health services, hospitalizations, etc. for the downers cohort and against manic spending sprees for the uppers recipients. After that’s completed, the guinea pigs can be informed of what was done to them, unless, of course, yet another health-related use can be found for the data.

Some commenters justified their far-reaching, grim extrapolations by pointing to the general lack of transparency surrounding Facebook’s practices. The public’s surprise acts as evidence of a lack of transparency, even as Facebook does disclose some information about their use of data in official policies, public-facing research, and statements to the press. The adequacy of such disclosures is outside the focus of this article, though just because information is technically available does not mean it is effectively so, as evidenced by the extensive research showing that people do not read platform policies (Martin, 2016a). As these patterns of response make clear, public perceptions of transparency do not necessarily align with company practices (Fiesler and Hallinan, 2018; Martin, 2015).

Other commenters justified their dark speculations by pointing to the subject of manipulation: emotions. For these commenters, emotional manipulation is a touchstone that enables the manipulation of what someone thinks, believes, and does—even who they are. The personal significance of emotions ups the stakes significantly, such that the experiment is understood as manipulating not only user experience, but also the very identity of the user. This kind of power is seen as having drastic political consequences that can sway elections or create, in the words of one commenter, a “herd of docile consumers”:

Don’t be fooled, manipulating a mood is the ability to manipulate a mind. Political outcomes, commerce, and civil unrest are just a short list of things that can be controlled.

There are also concerns about the relationship between emotional manipulation and mental health. Participants in the experiment received different treatments: the News Feeds of one group prioritized negative content, which some commenters interpreted as Facebook intentionally making people sad. This group received significantly more attention from comments than the group exposed to a higher concentration of positive content or the group exposed to less emotional content overall. An ethical response survey conducted shortly after the controversy broke (Schechter and Bravo-Lillo, 2014) also found greater support for a version of the study that only added more positive content to News Feeds. The addition of negative content is seen as a particularly harmful form of manipulation, a view compounded by concerns that the sample population could have included vulnerable populations:

Faecesbook [sic] is evil. What if one (or more) of their users (or victims) had been depressed and on the edge of suicide? Murdered for Zuckerbergs greater profits?

The extreme stakes of manipulation—from total political control to mass suicide—may seem out of place given the relatively minor treatment (tweaking the order in which content appears in the News Feed according to sentiment analysis of word choice) and the small effects size of the study’s findings (Kramer et al., 2014). Indeed, the findings of the study could only be significant at the scale of a massive platform like Facebook with billions of users. However, the concerns expressed in public reaction posit a much more dramatic scale of effects and make it apparent that many people do not have an accurate frame of reference to interpret these kinds of harms—or benefits.

Wake up, sheeple

Not all commenters expressed surprise about the FEC study. The theme “Wake up, sheeple” brings together comments that interpret the FEC study as a confirmation of pre-existing negative views of Facebook. These comments take a position of being profoundly unsurprised, seeing the experiment as a confirmation of the way they already understand and relate to Facebook. Similar to the “Manipulation anxieties” theme, these comments paint a negative, even dystopic picture of Facebook—but these comments also lack any sense of surprise. Experimentation and manipulation appear to be ordinary and expected behavior when considered alongside accounts of Facebook’s past bad behavior, negative perceptions of social media and Silicon Valley generally, or sharp critiques of the larger economic order. The comments tend to argue that other people need to “wise up” and either accept that this is the way the world works or opt out of using social media entirely, an attitude that has surfaced in prior work examining public reactions to privacy controversies (Fiesler and Hallinan, 2018):

The minute anyone signs up for membership to ANY group, you know that you are going to be manipulated. Ever hear the word SHEEPLE?

This antagonistic stance allows commenters to affirm their own positions, knowledge, and decisions. It also discredits the reactions of others, treating all aspects of the controversy as things that Facebook users should already expect. In doing this, the commenters shift accountability away from the company or the researchers and toward individual Facebook users:

Anyone who doesn’t realise that anything you put “out there” on Facebook (or any other social media site) is like shouting it through a bullhorn should have their internet competency licence revoked. We can’t blame all stupidity on some or other conspiracy . . .

It is notable that many of the people whose comments fell into this theme also identified as nonusers of Facebook. Some commenters framed their nonuse status as a value judgment against those who use social media. Other commenters argued that people should follow their example and decide to leave social media. These comments reframed the Facebook user base, arguing that users are actually the product that is sold to advertisers, the “real users” of Facebook. In these explanations, commenters frequently shame others for not having the same expectations they do:

Facebook is akin to an open corral baited with fake food; the herd gathers instinctively, but receives no nourishment . . . Get wise, people.

What exactly should people wise up about? Our data point to the behavior and practices of Facebook, of Silicon Valley, and of any service that is “free.” Rather than focusing on the study itself, the thrust of the indictment is that other people failed to recognize an obvious situation. However, even with this framing, there are substantial differences in opinion about what is considered obvious and how people should respond. Some call for the wholesale rejection of social media and testify to their own ability to get by without it. Others call for the adoption of a nihilistic attitude: this is the way the world works and all you can do is resign yourself to the facts. Despite disagreement over the solution, these commenters agree that the attitudes and actions of anyone who is outraged are the crux of the problem, not the experiment itself or the practices of the platform.

No big deal

Finally, even among the outrage, some commenters indicated that they had no issues with the FEC study—not necessarily because they judged it ethical, but rather because it was not just unsurprising but also unremarkable. It aligned with their expectations, whether for Facebook, advertising-supported media, or corporations generally:

The only thing that surprises me about this study is that anyone is surprised. Purveyors of information routinely attempt to manipulate their audiences and always have . . .

Similar to the “Wake up, sheeple” theme, these comments take the experiment as confirmation of their expectations and understanding of the platform. However, in contrast, these comments assess the situation as unproblematic and, if any action is required, it is the need for education about what Facebook is and how it works.

The views of some comments in this theme most strongly align with those of the researchers and with Facebook itself. Many commenters shared the view that there had been miscommunication or misunderstanding; as a result, comments explain different aspects of the situation, including the prevalence of A/B testing, industry research, and the general operation of the News Feed. Unlike those who were alarmed because of their ignorance of News Feed algorithms, these commenters formed expectations based, in part, on their understanding of those algorithms:

A/B testing (i.e. basically what happened here) when software companies change content or algorithms for a subset of users happens *all the time*. It’s standard industry practice.

Other commenters argue that emotional manipulation is not a concern because individuals have the ability to resist manipulation, whether through a skeptical disposition, education, or willpower. As one commenter put it, users are “masters of their own minds” and cannot be so easily swayed by a website.

For others, Facebook’s actions are typical of any corporation; a company is entitled to pursue its own policies and interests and if people do not like the practices of a company, they can simply choose not to use its services. This echoes the control model of privacy and supports a market approach to regulation (Martin, 2016b):

They can do whatever they want with their platform. Period. Build your own if you want to set the rules.

Other commenters point out that this is nothing new, referencing other forms of manipulation or persuasion, from advertising and marketing, to political speech, to everyday interactions. Where commenters expressing “manipulation anxieties” also considered the broader contexts for manipulation, the difference here is the normalization of manipulation as mundane rather than a dystopic version of a possible future:

So what’s new? The raison d’être for all media, even before the printing press, was to influence our emotions, secretly or otherwise.

Both this theme and “Wake up, sheeple” argue that the controversy surrounding the FEC study stems from a lack of public understanding of how social media works and propose communication solutions—albeit with radically different understandings and solutions. From avoiding all social media to knowing oneself, from embracing nihilism to education about technology, the recommendations are divergent and contradictory. The problem of communication, then, is about more than strategies and tactics and is instead based on a more fundamental disagreement about what the platform is and what people should expect from it.

Discussion

We began this research, in part, with the hope that analyzing public responses would tell us what people found to be objectionable about the FEC study, and thus what the public perceived as “unethical” in platform-based research. Our findings provide some answers to this question, including issues of transparency, manipulation, and the potential for future harm. However, just as the people involved in the research and publication of the FEC study were surprised by the public reaction to the study, our analysis reveals that members of the public were also surprised by the values and expectations of their peers. While the use of Facebook and other online platforms is widespread and frequent, a common part of people’s daily experience, the differences expressed in the comments of news articles about the FEC study highlight the lack of consensus around what these platforms are, how they should operate, and the role of platform-based research. In other words, the norms surrounding online platforms are neither unified nor settled. As a result, there is no single answer to what bothered people about this research, which means there is no single answer to what needs to be “fixed.”

While our findings do not support a one-size-fits-all solution to ensuring ethical research and avoiding controversy, our findings do support the importance of thinking about platform-based research holistically—that is, considering the relationship between academic research, collaborative research between academic and corporate researchers, and industry research both basic and applied. Although the FEC study was the product of a collaboration between academic and Facebook researchers, commenters rarely engaged with the specificity of the setup regardless of their position on the research. For example, comments in the “Living in a lab” theme tended to group all research together into a nefarious and dehumanizing category exemplified by animal testing and Nazi medical experimentation, while comments in the “No big deal” theme tended to argue for the normalcy and importance of research for improving commercial products. Certainly, neither account accurately describes the context or conduct of the FEC study. At the same time, the conflation of very different kinds of research cautions researchers against assuming that the public understands what platform-based research involves or why it might matter—questions that should, in turn, guide the design and communication of research.

Just as there is no shared understanding of platform-based research, so too is there no shared understanding of the platforms themselves. Ethical considerations for platform-based research often begin from the terms set by the platforms themselves—for example, the desirability of algorithmically sorting content in the News Feed. Despite formal agreement to a platform’s terms of service, these terms are not necessarily known or accepted by all users, to say nothing of the broader public that includes users and nonusers alike. The assumption of a shared understanding of Facebook’s News Feed algorithm and practices of research and experimentation made the negative reactions to the study genuinely unexpected to those involved. Certainly, the FEC study is not an isolated instance of researchers “badly reading the room” when it comes to expectations about social media. The public’s relationship—or rather, relationships—to platforms shape their assessment of research conducted on and about platforms. Facebook has repeatedly struggled in comparison to other platforms and tech companies in terms of public trust (Newton, 2017). The pre-existing lack of trust in the platform helps explain some of the more extreme accounts of harm in the reaction to the FEC study, which in turn further exacerbated issues of trust as The Guardian poll conducted in the wake of the controversy found (Fishwick, 2014). Complementing other calls for contextually sensitive ethical decision-making (Fiesler and Proferes, 2018; Jackman and Kanerva, 2016), we suggest a relational approach to the ethics of platform research that highlights what our data suggests is a particularly important context that researchers should be considering: the public’s relationship to online platforms.

This approach takes inspiration from work on relational ethics (Ellis, 2007), developed to guide interpersonal interactions for qualitative research. However, interactions on social media are not only interpersonal, but also involve human-machine communication, or interactions with technologies that reproduce aspects of human intelligence (Guzman and Lewis, 2019). The News Feed and other forms of recommendation offer prominent examples of this technology on Facebook, selecting and organizing content in order to show people the “stories that matter most” (News Feed, n.d.). As a result, the platform functions as a kind of third party to social media research, and a particularly important party because the relationship between the platform and the research subjects precedes and endures beyond the boundaries of any given study. Just as an ethnographer works to maintain good relationships with members of a community so that future researchers can obtain access to that community, so too should social media researchers consider ways of maintaining or improving relationships with their research populations. Such considerations may seem out of place for research practices that do not involve direct interpersonal interactions between researchers and research subjects—with the FEC study, for example, the research subjects had no way of knowing that they were part of an experiment. However, our findings illustrate that even experimental setups without interpersonal interactions can be perceived in very personal ways. These negative reactions can have a corrosive effect on trust for both platforms and research. How can we work to preserve a positive relationship instead?

For researchers, the first step in a relational approach to ethics involves understanding the public’s expectations for platforms. Although EVT was initially developed in the context of interpersonal communication, it also offers a theoretical framework for ethical considerations of online platform-based research. Relying on formal practices such as institutional review or compliance with terms of service is unlikely to address user norms and expectations because social media users are often unaware of research taking place on online platforms (Fiesler and Proferes, 2018), rarely read terms of service (Galbraith, 2017), and interpret the meaning of formal policy documents according to pre-existing expectations (Martin, 2015). Given the limitations of these formal practices, researchers can develop better understandings of user expectations from empirical ethics research (Fiesler and Proferes, 2018; Kennedy et al., 2015; Schechter and Bravo-Lillo, 2014) and from the emerging literature on folk theories of platforms (Devito et al., 2018; Eslami et al., 2016). The analysis of news comments presented here contributes to this project and demonstrates the complementary value of this methodology as a way to bring in different voices and study the relationship between expectations and arguments.

The importance of diverse relationships to platforms suggests another strategy for ethical decision-making: researchers should broaden our understanding of ethical stakeholders to include nonusers. As our data illustrate, even people who do not use Facebook have expectations for the platform and are invested enough in these expectations to react publicly when they are violated. Nonusers are also stakeholders, both because they consider themselves to be and because as social media platforms grow in terms of features and users, the influence of platforms includes broad societal effects (Baumer et al., 2015; Satchell and Dourish, 2009). Controversy analysis provides a way to surface beliefs and values that might otherwise be overlooked or taken for granted, even as these beliefs and values are central to the ways that people evaluate the actions of platforms—including research that takes place on platforms. Furthermore, the willingness of nonusers to make their interests and concerns public means that these perspectives fold back upon the platform’s user-base, shaping their expectations and concerns in turn. As a result, incorporating the expectations of nonusers into ethical decision-making can help anticipate controversies, push researchers to consider the public benefit of their research, and cultivate more beneficial ways of relating to platforms.

While we argue for the importance of considering a broader range of ethical stakeholders, we recognize that this is a challenging task. Just as previous research has argued for the importance of understanding user expectations in ethical decision-making (Fiesler and Proferes, 2018; Martin, 2015, 2016b; Schechter and Bravo-Lillo, 2014), our findings suggest that it is not feasible (or desirable) to identify a set of basic expectations common to all users. The views expressed in the “Wake up, sheeple” theme overwhelmingly begin from the premise that social media research is inherently unethical and that it either should be avoided entirely or that its use requires resignation to an unethical system. It is difficult to imagine a meaningful baseline set of expectations that include this perspective alongside the expectations of those who endorsed the experiment and see a clear social value in Facebook. However, a better understanding of the different relationships people have to platforms offers an opportunity to develop approaches that account for the needs and expectations of different relationships. Instead of simply telling people what their expectations should be, or inferring expectations from official policies (Gelinas et al., 2017), there is value in empirically studying expectations. In addition to formalized responses such as official policies that govern platform conduct, we should consider initiatives designed to cultivate informal norms and expectations. Compared to other forms of regulation such as legislation or formal policies, norms offer greater flexibility to adapt to particular contexts and technological developments.

Expectation violation can have substantial ramifications on the public perception of research and—potentially—support for future research. Controversies can also drive change, such as the development and implementation of industry review of research at Facebook (Jackman and Kanerva, 2016). The case of the FEC study offers some insight into what was poorly received and why. We can clearly say, for instance, that an approach to platform-based research based on implicit consent for research via terms of service is unpopular among the commenting public. A study that places specific opt-in requirements on its participants, even if the study design is kept hidden, may be received more positively and resolve some of the more prominent concerns, including the cloud of secrecy around research, not knowing if and when one has been the subject of an experiment, and the inclusion of vulnerable populations. Even an opt-out option could address some of these concerns, as it would allow people with specific objections to research to be removed from it without requiring them to stop using the platform entirely. Fundamentally, a relational approach to ethical decision-making for platform-based research begins with an understanding of public expectations for platforms and uses that understanding to inform the design and communication of research.

Footnotes

Acknowledgements

The authors would like to thank Michael Zimmer, Katie Shilton, Caitlin Reynolds, the anonymous reviewers, and members of the Identity Lab and the Internet Rules lab for all of their help cultivating this paper.

Authors’ note

Blake Hallinan is now affiliated to Hebrew University.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This work was funded by NSF award IIS-1704369 as part of the PERVADE (Pervasive Data Ethics for Computational Research) project.

ORCID iD

Blake Hallinan

Notes

Author biographies

Blake Hallinan is a postdoctoral fellow in the Department of Communication and Journalism at Hebrew University of Jerusalem. From the Lazarsfeld-Stanton Program Analyzer to the Like button, Blake researches the cultural politics of the transformation of feelings and emotions into information.

Jed R Brubaker is an assistant professor in the Department of Information Science at the University of Colorado Boulder. His research focuses on how identity is designed, represented, and experienced through technology.

Casey Fiesler is an assistant professor in the Department of Information Science at the University of Colorado Boulder. She studies social computing and governance, including the ethical and legal implications of researching and designing technology.

References

Altmetric (n.d.) Overview of attention for article published in Proceedings of the National Academy of Sciences of the United States. Altmetric. Available at: https://www.altmetric.com/details/2397894 (accessed 15 April 2018).

Baumer

EPS

Ames

Burrell

, et al. (2015) Why study technology non-use? First Monday 20(11). Available at: https://dx-doi-org.web.bisu.edu.cn/10.5210/fm.v20i11.6310

Bevan

Ang

Fearns

(2014) Being unfriended on Facebook: an application of expectancy violation theory. Computers in Human Behavior 33: 171–178.

Brown

Weilenman

McMillan

, et al. (2016) Five provocations for ethical HCI research. In: Proceedings of the SIGCHI conference on human factors in computing systems, Montréal, QC, Canada, 22–27 April. San Jose, CA: ACM Press.

Bucher

(2017) The algorithmic imaginary: exploring the ordinary affects of Facebook algorithms. Information Communication and Society 20(1): 30–44.

Burgoon

Le Poire

(1993) Effects of communication expectancies, actual communication, and expectancy disconfirmation on evaluations of communicators and their communication behavior. Human Communication Research 20(1): 67–96.

Carlson

(2018) Facebook in the news: social media, journalism, and public responsibility following the 2016 Trending Topics controversy. Digital Journalism 6(1): 4–20.

Chandler

Sun

Racine

(2017) Online public reactions to fMRI communication with patients with disorders of consciousness: quality of life, end-of-life decision making, and concerns with misdiagnosis. AJOB Empirical Bioethics 8(1): 40–51.

Clarke

Braun

(2006) Using thematic analysis in psychology. Qualitative Research in Psychology 3(2): 77–101.

10.

De Kraker

Kuys

Cörvers

RJM

, et al. (2014) Internet public opinion on climate change: a world views analysis of online reader comments. International Journal of Climate Change Strategies and Management 6(1): 19–33.

11.

Devito

Birnholtz

Hancock

, et al. (2018) How people form folk theories of social media feeds and what it means for how we study self-presentation. In: CHI ’18: Proceedings of the 36th annual ACM conference on human factors in computing systems, Montreal, QC, Canada, 21–26 April.

12.

D’Onfro

(2014) Facebook researcher responds to backlash against “creepy” mood manipulation. Business Insider, 29 June. Available at: https://www.businessinsider.com.au/adam-kramer-facebook-mood-manipulation-2014-6

13.

Ellis

(2007) Telling secrets, revealing lives: relational ethics in research with intimate others. Qualitative Inquiry 13(1): 3–29.

14.

Eslami

Rickman

Vaccaro

, et al. (2015) “I always assumed that I wasn’t really that close to [her]”: reasoning about invisible algorithms in news feeds. In: CHI ’15: Proceedings of the 33rd annual ACM conference on human factors in computing systems, Seoul, Republic of Korea, 18–23 April, pp. 153–162. New York: ACM Press.

15.

Eslami

Karahalios

Sandvig

, et al. (2016) First I “like” it, then I hide it: folk theories of social feeds. In: CHI ’16: Proceedings of the 2016 CHI conference on human factors in computing systems, San Jose, CA, 7–12 May, pp. 2371–2382. New York: ACM Press.

16.

Fiesler

Hallinan

(2018) We are the product”: public reactions to online data sharing and privacy controversies in the media. In: Proceedings of the 2018 CHI conference on human factors in computing systems (CHI ’18), Montreal, QC, Canada, 21–26 April.

17.

Fiesler

Proferes

(2018) “Participant” perceptions of Twitter research ethics. Social Media + Society 4(1): 1–14. DOI: 10.1177/2056305118763366.

18.

Fishwick

(2014) Facebook’s secret mood experiment: have you lost trust in the social network? The Guardian, 30 June. Available at: https://www.theguardian.com/technology/poll/2014/jun/30/facebook-secret-mood-experiment-social-network

19.

Fowler

Christakis

(2008) Dynamic spread of happiness in a large social network: longitudinal analysis over 20 years in the Framingham heart study. BMJ 337: a2338.

20.

Frizell

(2014) Facebook totally screwed with a bunch of people in the name of science. TIME, 28 June. Available at: http://time.com/2936729/facebook-emotions-study/

21.

Galbraith

(2017) Terms and conditions may apply (but have little to do with ethics). American Journal of Bioethics 17(3): 21–22.

22.

Gelinas

Pierce

Winkler

, et al. (2017) Using social media as a research recruitment tool: ethical issues and recommendations. American Journal of Bioethics 17(3): 3–14.

23.

Giles

Holmes

McColl

, et al. (2015) Acceptability of financial incentives for breastfeeding: thematic analysis of readers’ comments to UK online news reports. BMC Pregnancy and Childbirth 15(1): 116.

24.

Glenn

Champion

Spence

(2012) Qualitative content analysis of online news media coverage of weight loss surgery and related reader comments. Clinical Obesity 2(5–6): 125–131.

25.

Goel

(2014) As data overflows online, researchers grapple with ethics. The New York Times, 12 August. Available at: https://www.nytimes.com/2014/08/13/technology/the-boon-of-online-data-puts-social-science-in-a-quandary.html

26.

Goodnight

(2005) Science and technology controversy: a rationale for inquiry. Argumentation and Advocacy 42(1): 26–29.

27.

Griffin

Ledbetter

Sparks

(2011) Expectancy violations theory. In: Griffin

(ed.) A First Look at Communication Theory. 9th ed. New York: McGraw-Hill, pp. 84–92.

28.

Grimmelmann

(2014) The Facebook emotional manipulation study: sources. The Laboratorium, 30 June. Available at: http://laboratorium.net/archive/2014/06/30/the_facebook_emotional_manipulation_study_source

29.

Guzman

Lewis

(2019) Artificial intelligence and communication: a human–machine communication research agenda. New Media & Society 22(1): 70–86.

30.

Hancock

(2019) The ethics of digital research. In: Welles

Gonzalez-Bailon

(eds) The Oxford Handbook of Networked Communication. Oxford: Oxford University Press. Available at: https://www.oxfordhandbooks.com/view/10.1093/oxfordhb/9780190460518.001.0001/oxfordhb-9780190460518-e-25.

31.

Hatfield

Cacioppo

JTJ

Rapson

(1993) Emotional contagion. Current Directions in Psychological Science 2(3): 96–99.

32.

Henrich

Holmes

(2013) Web news readers’ comments: towards developing a methodology for using on-line comments in social inquiry. Journal of Media and Communication Studies 5(1): 1–4.

33.

Hiltzik

(2014) Facebook on its mood manipulation study: another non-apology apology. Los Angeles Times, 2 July. Available at: http://www.latimes.com/business/hiltzik/la-fi-mh-facebook-apology-20140702-column.html

34.

Holton

Lee

Coleman

(2014) Commenting on health: a framing analysis of user comments in response to health articles online. Journal of Health Communication 19(7): 825–837.

35.

Hudson

Bruckman

(2004) ‘Go Away’: Participant Objections and the ethics of Chatroom Research. The Information Society: An International Journal 20(2): 127–139.

36.

Jackman

Kanerva

(2016) Evolving the IRB: building robust review for industry research. Washington and Lee Law Review 72(3): 442–457.

37.

Kennedy

Elgesem

Miguel

(2015) On fairness: user perspectives on social media data mining. Convergence: The International Journal of Research into New Media Technologies 23: 270–288.

38.

Kramer

ADI

Guillory

Hancock

(2014) Experimental evidence of massive-scale emotional contagion through social networks. Proceedings of the National Academy of Sciences 111(24): 8788–8790.

39.

McLaughlin

Vitak

(2012) Norm evolution and violation on Facebook. New Media & Society 14(2): 299–315.

40.

Marres

(2015) Why map issues? On controversy analysis as a digital method. Science, Technology, & Human Values 40(5): 655–686.

41.

Marres

Moats

(2015) Mapping controversies with social media: the case for symmetry. Social Media and Society 1(2): 1–17.

42.

Martin

(2015) Privacy notices as Tabula Rasa. Journal of Public Policy & Marketing 34(2): 210–227.

43.

Martin

(2016a) Formal versus informal privacy contracts: comparing the impact of privacy notices and norms on consumer trust online. The Journal of Legal Studies 45(Suppl. 2): S191–S215.

44.

Martin

(2016b) Understanding privacy online: development of a social contract approach to privacy. Journal of Business Ethics 137(3): 551–569.

45.

Nebeker

Harlow

Espinoza Giacinto

, et al. (2017) Ethical and regulatory challenges of research using pervasive sensing and other emerging technologies: IRB perspectives. AJOB Empirical Bioethics 8(4): 266–276.

46.

News Feed (n.d.) Facebook for media. Available at: https://www.facebook.com/facebookmedia/solutions/news-feed (accessed 9 December 2018).

47.

Newton

(2017) America doesn’t trust Facebook. The Verge, 27 October. Available at: https://web.archive.org/web/20190402070156/https://www.theverge.com/2017/10/27/16552620/facebook-trust-survey-usage-popularity-fake-news

48.

Nissenbaum

(2004) Privacy as contextual integrity. Washington Law Review 79(119): 1119–1158.

49.

Pennebaker

Chung

Ireland

, et al. (2007) The Development and Psychometric Properties of LIWC2007 (LIWC2007 Manual). Austin, TX: LIWC.

50.

Rader

Gray

(2015) Understanding user beliefs about algorithmic curation in the Facebook news feed. In: Proceedings of the 33rd annual ACM conference on human factors in computing systems (CHI’15), pp. 173–182. Available at: https://dl.acm.org/citation.cfm?id=2702174

51.

Recuber

(2016) From obedience to contagion: discourses of power in Milgram, Zimbardo, and the Facebook experiment. Research Ethics 12(1): 44–54.

52.

Rosenquist

Fowler

Christakis

(2011) Social network determinants of depression. Molecular Psychiatry 16(3): 273–281.

53.

Satchell

Dourish

(2009) Beyond the user: use and non-use in HCI. In: Proceedings of the 21st annual conference of the Australian computer-human interaction special interest (OZCHI ’09), Melbourne, VIC, Australia, 23–27 November, pp. 9–16. New York: ACM Press.

54.

Schechter

Bravo-Lillo

(2014) Using ethical-response surveys to identify sources of disapproval and concern with Facebook’s emotional contagion experiment and other controversial studies. Available at: https://www.microsoft.com/en-us/research/wp-content/uploads/2016/02/Ethical-Response20Survey202014-10-30.pdf

55.

Scissors

Burke

Wengrovitz

(2016) What’s in a like? Attitudes and behaviors around receiving likes on Facebook. In: Proceedings of the 19th ACM conference on computer-supported cooperative work & social computing (CSCW’16), San Francisco, CA, 27 February–2 March, pp. 1499–1508. New York: ACM Press.

56.

Shklovski

Mainwaring

Skúladóttir

, et al. (2014) Leakiness and creepiness in app space. In: Proceedings of the 32nd annual ACM conference on human factors in computing systems (CHI’14), pp. 2347–2356. Available at: https://dl.acm.org/citation.cfm?id=2557421

57.

Silberg

(2014) So you are shocked Facebook did #psyops on people? Huffpost, 29 August. Available at: https://www.huffpost.com/entry/so-you-are-shocked-facebo_b_5542094.

58.

Silva

(2015) What do users have to say about online news comments? Readers’ accounts and expectations of public debate and online moderation: a case study. Participations: Journal of Audience & Reception Studies 12(2): 32–44.

59.

Stark

(2018) Algorithmic psychometrics and the scalable subject. Social Studies of Science 48(2): 204–231.

60.

Steadman

(2014) How Facebook’s news feed controls what you see and how you feel. New Statesman, 30 June. Available at: https://www.newstatesman.com/future-proof/2014/06/how-facebooks-news-feed-controls-what-you-see-and-how-you-feel

61.

Turkle

(2011) Alone Together: Why We Expect More from Technology and Less from Each Other. New York: Basic Books.

62.

Verma

(2014) Editorial expression of concern: experimental evidence of massive scale emotional contagion through social networks. Proceedings of the National Academy of Sciences 111(29): 10779–10779.

63.

Vines

Thieme

Comber

, et al. (2013) HCI in the press: online public reactions to mass media portrayals of HCI research. In: Proceedings of the SIGCHI conference on human factors in computing systems (CHI ’13), pp. 1873–1882. Available at: https://dl.acm.org/citation.cfm?id=2466247

64.

Vitak

Shilton

Ashktorab

(2016) Beyond the Belmont principles: ethical challenges, practices, and beliefs in the online data research community. In: Proceedings of the 19th ACM conference on computer-supported cooperative work & social computing (CSCW’16), pp. 939–951. Available at: https://dl.acm.org/citation.cfm?id=2820078

65.

Williams

Burnap

Sloan

Jessop

, et al. (2017) Users’ views of ethics in social media research: Informed consent, anonymity, and harm. In: Woodfield

(ed) The Ethics of Online Research (Advances in Research Ethics and Integrity, Vol. 2). Emerald Publishing Limited, pp. 27–52.