media regulation

https://doi.org/10.1177/1556264619874646

Journal of Empirical Research on Human Research Ethics 2020, Vol. 15(1-2) 77 –86 © The Author(s) 2019 Article reuse guidelines: sagepub.com/journals-permissions DOI: 10.1177/1556264619874646 journals.sagepub.com/home/jre

Everyday Experience and Ethics of Social Media Research Practice

Introduction

The ethical treatment of data collected from social media platforms has been subjected to renewed debates in recent years. In 2018 alone, the Facebook Cambridge Analytica scandal around data misuse, the suggestion of net neutrality regulations, and the realization of the General Data Protection Regulation (GDPR) have all featured on the pub- lic agenda as well as in academic circles. Following years of controversies around the misuse of data by platform provid- ers and the circulation of data assumed by users to be pri- vate (Leetaru, 2018), social media platforms have gained a reputation for being ethically questionable. Aside from by now common concerns around privacy, consent, confidenti- ality, and data storage, researchers are therefore facing addi- tional difficulties in gaining consent when users may be reluctant to provide information that is not mandatory for using the platforms. While online ethics remain important, particularly in human research, there are still questions around what should guide ethical decision-making on social media: laws, platform terms, cultural conventions, tradi- tional and Internet-specific ethical guidelines (Sormanen & Lauk, 2016), top-down or bottom-up decisions, and regula- tion-driven or context-specific approaches (Markham & Buchanan, 2012). Overall, the ethical treatment of social media data remains contentious, particularly since social media present a double challenge: ethics themselves are

known to evolve constantly and in the case of social media research they do so in ever-changing environments and therefore contexts.

In response to these concerns, this article addresses the issue of context. Context has been a prominent topic in newer literature on social media ethics (e.g., Nissenbaum, 2004; boyd & Crawford, 2012; Association of Internet Researchers [AOIR] guidelines/Markham & Buchanan, 2012). In particular, Nissenbaum’s (2004) recommenda- tion for researchers to have “contextual integrity” for ethi- cal research, meaning a consideration of what is assumed by audiences in terms of privacy, has shaped research eth- ics in social media research (see application in Zimmer, 2018; for further exploration, see Schultze & Mason, 2012; for recommendations, see AOIR guidelines/Markham & Buchanan, 2012). Although context as a concept has been widely addressed in the computational sciences (see Seaver, 2015, for overview), ethics literature commonly addresses it as an issue of reframing data or taking data out of context, particularly in Twitter research (e.g., Ahmed,

874646 JREXXX10.1177/1556264619874646Journal of Empirical Research on Human Research EthicsÖzkula research-article2019

1The University of Sheffield, UK

Corresponding Author: Suay Melisa Özkula, Department of Sociological Studies, Elmfield, The University of Sheffield, Northumberland Road, Sheffield S10 2TN, UK. Email: [email protected]

The Issue of “Context”: Data, Culture, and Commercial Context in Social Media Ethics

Suay Melisa Özkula1

Abstract One of the central concerns in research ethics in recent years has been the vast amount of data available from social media platforms and the related concerns around what establishes an ethical use of data. Toward addressing these challenges, researchers have therefore called for the consideration of “context” in Internet research. However, context remains a fuzzy concept and little guidance exists on its different dimensions. In response to this issue, this article uses worked examples from three data sets to discuss three different dimensions of “context”: data context, cultural context, and commercial context. The article problematizes these dimensions and offers suggestions toward creating ethical sensibility to these by drawing on two data sets from 2017: (a) climate change imagery scraped from five social platforms and (b) digital-ethnographic work at the climate summit COP23.

Keywords context, Internet research ethics, social media ethics, contextual integrity, data context, commercial context, cultural context

78 Journal of Empirical Research on Human Research Ethics 15(1-2)

Bath, & Demartini, 2017; boyd & Crawford, 2012; Fiesler & Proferes, 2018), or, in relatively vague terms, as an issue that helps to determine the assumed privacy of a group (e.g., Fiesler & Proferes, 2018; Nissenbaum 2004; Zimmer, 2018). While a lot of ethics literature mentions context as relevant, little work (beyond Nissenbaum’s seminal text in 2004 and Schultze and Mason’s work on entanglement in 2012) has been dedicated to offering explicit guidance on what kinds of contexts researchers need to consider (beyond network context in Twitter networks) and how, an issue this article seeks to address.

In what follows, this article will explore some ethical concerns that follow from new social media research and methodology. It draws on two studies on social media com- munications of climate change, which were conducted in 2017, to discuss different dimensions of context. It will be argued that the ethical challenges of social media research depend on a particular set of contextual factors beyond pri- vacy settings and content that researchers need to consider for online study. Based on more recent trends in digital studies scholarship, it will be recommended that research- ers consider the context of data, individuals’ cultural con- texts, and platform factors in their assessment of ethical research. These issues will be presented in three parts: (a) data context, (b) cultural context, and (c) commercial con- text, after an overview of context in ethical social media research and the methodological background of the datasets.

“Context” in Internet Research Ethics

Context has been a central concern in Internet research eth- ics (see AOIR guidelines/Markham & Buchanan, 2012; Marwick & boyd, 2014; Nissenbaum, 2004). Above all, Nissenbaum’s (2004) concept of contextual integrity has been central in promoting the idea that ethical Internet research is not grounded in universally applicable guide- lines, but is, instead, tied to a contextual understanding of the specific space under study. Even so, there is little clar- ity on what context captures and entails. Context in itself is a fickle and unbounded—even fuzzy—concept (see Seaver, 2015; for a description of context as “entanglement,” see Schultze & Mason, 2012). This issue is illustrated in the title of Seaver’s (2015) article: “The nice thing about con- text is that everyone has it”—essentially: context very much depends on context.

According to Dey (2001), this issue of definition stems from scholars having used different and often vague defini- tions of context. For example, context has been conceptual- ized both as location and nearby people/objects and simply as environment or situation (Dey, 2001). Based on the dif- ficulty of applying such definitions in practice (within com- puting studies), Dey offers a different definition:

Context is any information that can be used to characterise the situation of an entity [ . . . ], a person, place, or object that is considered relevant to the interaction between a user and an application, including the user and applications themselves. (Dey, 2001, p. 5)

In line with some prominent guidance in Internet research ethics (e.g., Markham & Buchanan, 2012), Dey’s definition suggests that the significance of a particular connection, object, or data depends on whether it is relevant in a given context. Thus, the individual elements that are relevant in the ethical treatment of data (i.e., context) depend very much on the nuances of an individual case. In the AOIR guidelines on Internet research ethics, Markham and Buchanan illustrate this issue as follows:

Individual and cultural definitions and expectations of privacy are ambiguous, contested, and changing. People may operate in public spaces but maintain strong perceptions or expectations of privacy. Or, they may acknowledge that the substance of their communication is public, but that the specific context in which it appears implies restrictions on how that information is—or ought to be—used by other parties. (Markham & Buchanan, 2012, p. 6)

Markham and Buchanan’s text suggests that there are three issues that make a clearer understanding of context rel- evant for scholarly study: (a) context is significant, (b) what precisely is relevant depends on the individual case, and (c) context is therefore contested, but it is also constructed (for context as constructed and contested, see Seaver, 2015), issues also mirrored by Nissenbaum (2004) and Marwick and boyd (2014). Thus, although the relevance of context has been acknowledged across Internet-related fields, the concept remains fuzzy in terms of guidance.

The fuzziness of the concept makes its practical appli- cation difficult. Unsurprisingly, and perhaps as a result of this fuzziness, scholarship offers little specific guidance toward understanding context. An exception are the AOIR guidelines (Markham & Buchanan, 2012) that offer a set of questions that help researchers to determine context under two overarching questions: “How is the context defined and conceptualized?” and “[h]ow is the context (venue/participants/data) being accessed?” (Markham & Buchanan, 2012). The guidelines additionally provide a chart listing “types of venues/contexts,” which includes direct communication, special interest forums, social net- working, personal spaces/blogs, avatar-based social spaces, virtual worlds, online gaming spaces, commercial web services, and databanks/repositories (Markham & Buchanan, 2012, p. 18). While the list is comprehensive, context here is limited to the type of space—essentially

Özkula 79

platforms or platform types. Thus, context here relates to data or platform context.

This view of platforms as context has become common- place among ethics scholars. For instance, Nissenbaum’s concept of contextual integrity is based on the notion that individual platforms constitute cultural spaces or commu- nities that assume different practices around privacy and anonymity. Such approaches are mirrored in common dis- tinctions by platform. For example, while Twitter is gener- ally considered more public, Facebook—due to its privacy settings—is usually seen to be more private (e.g., Ahmed et al., 2017; Townsend & Wallace, 2016). Although these are all valuable guidelines for understanding and address- ing the specific context, they are limited to an individual digital space (what I will call data context), but do not con- sider larger patterns of potentially relevant contexts.

Furthermore, while such distinctions have provided researchers with some guidance toward understanding their research spaces, understanding context in a space that is immediate and has archiving capabilities and viral- ity potential has proven difficult. For example, in the infa- mous Facebook study titled the “T3 project” in 2008, researchers collected data from 1,700 Facebook profiles of students of an anonymized university (Zimmer, 2010). However, due to other information provided, the univer- sity was soon identified, as were the individual students as their individual networks and cultural tastes were unique— like “fingerprints” (Stutzman in Zimmer, 2010, p. 316). Even when identifiable markers were deleted in response to the public controversy surrounding the study, the dam- age was done and the data still retrievable (Zimmer, 2010). In this case, context was considered (i.e., eliminated) toward anonymization, but it was also the very reason the case was identified. Thus, although the relevance of data context was acknowledged here (by omitting the organiza- tion’s and students’ names), not all types of context (such as the digital fingerprint of the network) were considered in the anonymization.

In part, researchers have therefore relied on platforms’ terms and conditions as well as individuals’ privacy settings as a form of context guidance. While consent and terms and conditions are significant standalone issues in Internet research ethics, they also provide contextual parameters as they help users and researchers to gauge and assess plat- form cultures and norms, and the related user expectations, an issue most recently highlighted by the Cambridge Analytica scandal. Even so (and despite certain types of consent having been written into the terms), it remains questionable whether participants are aware of what they have consented to. In fact, according to a Pew Research Center survey, half of U.S. Americans did not correctly understand what a privacy policy provides (Smith, 2014). Swirsky, Hoop, and Labott (2014) articulate similar con- cerns, explaining that information shared online may

contain additional information that users are not aware of. They may therefore not truly comprehend what their posts implicate for their privacy (Swirsky et al., 2014). Even when consent is obtained, it is uncertain whether users are conscious of what data they have provided.

An added challenge are individual platforms’ terms and services that (even when fully read) may change over time and without notice to users. Swirsky and colleagues cite Google’s terms as an example, explaining that the Google terms provide them with a license to any content transmit- ted through these channels; thus, privacy of online com- munications cannot be guaranteed (Swirsky et al., 2014) and users may be unaware of what they have consented to at any given time. Even when deleted, certain links and information cannot be unmade due to the archiving prop- erties of digital networks. After all, what happens online, stays online (Couts, 2011; Waddell, 2016). Thus, user per- spectives on privacy, anonymity, and consent depend on various factors beyond contextual integrity and beyond the immediate spatial context often highlighted in context lit- erature. They include users’ reading and understanding of data uses (data literacies), cultural understandings and expectations of these norms (techno-cultural context), as well as their understanding of what data may be found, where, and how it may be used (commercial context).

In response to these concerns, scholars have sought a range of guidance and advice. These include ethical guide- lines provided by the employing university, national research councils, field-specific national or international associations (on an international level most famously the AOIR), individual platforms’ terms and conditions, ethical conduct in national and cultural contexts, as well as guid- ance from equivalent traditional (non-digital) methods. Even so, there is little explicit guidance as (a) there are dis- crepancies between the advice given, based on national, cultural, and field-specific contexts; (b) many recommen- dations merely suggest the “consideration” of context and platform; and (c) are limited to an understanding of context as spatial—meaning platform- and community-specific, but do not offer any specific advice on how contextual factors need to be reflected in an ethical assessment (exceptions: Nissenbaum, 2004; Schultze & Mason, 2012). This article therefore offers a set of more specific contextual factors (data, cultural, and commercial context) for researchers’ consideration in ethical social media research toward awak- ening their ethical sensibilities to multiple dimensions of context. These suggestions are based on studies that are out- lined in the following section.

Data and Methods

The ethical reflections presented here are based on case studies from the Economic and Social Research Council (ESRC)-funded project “Making Climate Social”. These

80 Journal of Empirical Research on Human Research Ethics 15(1-2)

include empirical image–based research, empirical text– based research, and methodological research from two dis- tinct datasets and projects (named by the event where data were collected): (a) DMI—the Digital Methods Initiative’s summer school at the University of Amsterdam in July 2017 (in short: DMI) and (b) COP23—the United Nations’ (UN) global climate summit hosted in Bonn (Germany) in November 2017 (in short: COP).

The DMI project included the scraping of climate change–related imagery from platforms Facebook, Twitter, Instagram, Reddit, Tumblr, and Google (see Pearce et al., 2018; for methodology, also see DMI Wiki: https://wiki .digitalmethods.net/Dmi/MakingClimateVisible). The vary- ing visualizations offered both quantitative and qualitative data. The collected images were visualized in a range of different formats, including image plots, heat maps, and image stacks (for examples see the above Wiki page and Pearce et al. 2018). The image plots and co-hashtag analy- ses are based on a large data set (the entire collection of images spanned several thousand images) and displays large-scale patterns. In the case of heat maps and images stacks, on the other hand, the datasets were capped to 10 images per platform (highest ranking images) to provide more qualitative detail and analysis. While no personal data were used in the image scraping, the visual orientation, obscure data paths, and combination of quantitative and qualitative elements created an ethically interesting case.

The COP23 project, however, drew on more traditional methods including ethnographic observation and inter- views. Ethnographic immersion took place across plat- forms, devices, and in person at the event. Online, observation included coverage of the event on the official COP23 websites and social media channels and Twitter communication under the event hashtag. Offline, ethno- graphic observation took place at the event for its entire duration (November 06-17, 2017) in Bonn. As part of the ethnographic participation, parts of the event were photo- and video-recorded. In addition, participants were inter- viewed in formal and informal interviews and asked to allow the observation or video-recording of their social media activity during the event. Thus, while both cases were used to understand social media communications of climate change, they were different in their empirical focus (text versus image), data scale and detail (particularly quan- titative and qualitative focus), and in their combination of traditional and newer methods.

Note on ethical approval: Ethical approval was sought and given through the employing institution for all research conducted as part of the COP23 project. However, for the DMI image scraping, consent could not be obtained, due to the relatively unplanned, flexible, and freely collaborative nature of data sprints as part of summer school at an exter- nal institution. To mitigate for any issues surrounding image uses, in the public versions, the images were displayed

through either quantitatively oriented visualizations or (for qualitative depictions) through the creation of image stacks where images are overlaid and their opacity reduced (see Wiki for examples).

The next section offers a reflection on ethical concerns that arose from these studies with a focus on data scale and data context, platform culture, user’s geo-culture, and data paths in converging platforms and describes how these were addressed. It will be recommended that aside the more com- mon guidelines (institutional guidelines, platform terms and conditions, national and cultural contexts, and online– offline comparisons), researchers consider and embed sev- eral dimensions of contexts in their ethical treatment of social media data: (a) data context, (b) cultural context, and (c) commercial context.

Added Data Context: Data Scale, Form, and Context

The scale of the data collected differed across the studies, which was incorporated into the approach to the ethical treatment of the data. Where larger datasets were collected, for example, the COP tweet collection or image scraping, there was an underlying assumption that the contents were less sensitive. This logic followed from collecting little to no contextual data (such as profile data). While all the qualitative COP data were heavily contextualized within a specific event, location, participant base, and official social media coverage, all of which were interconnected, the big datasets were more decontextualized. The individual pieces of data were “shorter” in form, for example, individual tweets or images, and little to not at all connected to a par- ticular profile (depending on dataset). For instance, where images were scraped, we initially obtained metadata for the scraping, but it was the scraped images only (or in addition to keywords and tags) that were analyzed. These data were therefore considered less personal and less sensitive, a trend that is not uncommon with quantitative research (boyd & Crawford, 2012), but is also problematic as all researchers interpret data (boyd & Crawford, 2012).

One issue arose on the basis of network visualizations. Although qualitative data include less context, network graphs essentially recreate it. They are, as in the T3 Facebook study, “fingerprints.” Although these data were drawn from a platform (Twitter) that is generally consid- ered and, according to its terms and conditions public, this context was not part of the platform’s data. It was underlying context that, without researcher manipulation, would not have existed. The issue is, as boyd and Crawford (2012) argue, not that the dataset is big, but that it can be aggregated, searched, and analyzed for patterns. Thus, even though Twitter explicitly states that its content may be used for academic purposes, such uses do not spe- cifically outline their recreation and re-contextualization,

Özkula 81

applications users may not anticipate. This reflects some known concerns that platform use does not necessarily mean that a given user has in fact read or understood the terms (see Smith, 2014; Swirsky et al., 2014). Even where users fully understand terms, they may not wish to be framed in a particular way, an issue also observed by Ahmed et al. (2017) in their Twitter work. While in the case of the climate change data, there was no political or similarly contentious framing (it was little more than a metric), the supplication of network information to tweets produced context that participants may have been unhappy to share—added context.

While the scale of the data certainly played a role in the assessment of ethical concerns, so did the added con- text. The image scraping was very much decontextual- ized in our data presentation (and in part already in the data collection), but the tweets were not. For Twitter studies overall, this creates an interesting problematic. Twitter studies are often used to depict relationships, affiliations, and wider networks around issues/causes and keywords—it is what the medium lends itself to (Ahmed et al., 2017). Thus, researchers here run the risk of revealing additional information that embeds data and displays participants’ activities by visualizing digital fin- gerprints—essentially a contextualization of data. For predominantly organization-produced data (and simi- larly verified accounts), as was the case with the COP data here, this was, as such, little problematic. Nevertheless, in this case, it was decided not to visualize the tweets. As the overall methodological approach here was ethnographic, a larger network visualization was omitted. Although two colleagues generated a separate tweet collection of the summit for a COP side event, these data were not used as part of the ethnography for two reasons: (a) a network visualization was not abso- lutely necessary for the argument of the study and (b) ethnographic research is by default very contextualized and typically consists of a smaller sample size. It is therefore riskier, an issue also highlighted by Schultze and Mason (2012). Thus, additional considerations were given to the data’s degree of contextualization, the sig- nificance of any supplemented data for publication, the publicness of users, and the specific combination of methods and their capacity to create added context.

Cultural Contexts in Digital Space: Platform Culture and Users’ Geo- Culture

Beyond data context in form and scale, a common theme that denoted the individual studies was the perception of platform culture. In both the interviews and the collabora- tive research on climate change imagery, participants and

collaborators related ethical concerns around individual platform cultures, rhetorics, “grammars,” or “vernaculars” (see M. Gibbs, Meese, Arnold, Nansen, & Carter, 2015). Platform vernaculars constitute a “unique combination of styles, grammars, and logic” (M. Gibbs et al., 2015, p. 257) that can be derived from platform affordances and practices. Such vernaculars became evident in COP par- ticipants’ descriptions of individual platforms. Regardless of privacy settings, they generally considered Facebook a more private platform than Twitter. Although this differen- tiation is already partially acknowledged in ethical guide- lines for Internet research (e.g., Townsend & Wallace, 2016), participants’ reflections suggest that this recogni- tion is not merely an infrastructural fact, meaning that the view of Facebook as a private platform is not solely based on platform capabilities, but it is also a cultural assump- tion. This is mirrored in participants’ descriptions of the platforms. Twitter was generally recognized as a more professionally oriented platform, while Facebook was (in part derogatorily) described as a platform for more per- sonal sharing such as personal news and entertainment. These descriptions do not only denote platform capabili- ties, but assumed purposes. This suggests that the ethical treatment of social media data is subject not only to plat- form settings and users’ platform knowledge but also to users’ assumed treatment of their data based on pre-exist- ing platform vernaculars.

Cultural assumptions also call into question what users may justifiably expect around the use of their data. Unless information sheets are provided and consent is specifically sought, it remains for researchers to establish whether par- ticipants are aware of potential public uses of their data. Although certain knowledge bases may justifiably be assumed for certain platforms in specific cultural and social contexts, those contexts may vary depending on individual users’ national and cultural backgrounds. Thus, as such, national or cultural contexts may provide some guidance for ethical research (Sormanen & Lauk, 2016), but that context is not necessarily identifiable online, an issue that became problematic in the observation of Facebook groups as part of the COP study. Although the COP23 summit was held in the Western hemisphere, a cultural setting that suggests a good working knowledge of popular social media platforms such as Facebook and Twitter, the user base of these groups was largely undetermined. Due to the summit’s development and UN orientation, it also hosted participants from very diverse geographical, ethnic, and cultural backgrounds. It may therefore be presumed that the working knowledge of indi- vidual participants may have been tied to digital connectiv- ity and the individual platform’s popularity in their geo-cultural context. For instance, although Facebook and Twitter are largely seen as mainstream social media plat- forms in the global West, other platforms such as Weibo and Odnoklassniki have shown much wider distribution in other

82 Journal of Empirical Research on Human Research Ethics 15(1-2)

countries such as China and Russia, respectively (Hutt, 2017; Statista, 2017). Thus, the popularity of certain plat- forms over others as well as geo-cultural views of particular platforms also play a role in how expectations of privacy and data uses are judged.

This geo-cultural sensitivity became apparent in COP participants’ descriptions of individual technologies. Although Facebook was generally considered private, geo- cultural context still mattered. For instance, one participant described how in Ukraine Facebook was used for profes- sional as well as private purposes. Thus, Facebook groups were considered viable spaces for professional communica- tions and decision-making, a cultural norm that affects indi- viduals’ behaviors on and attitudes toward the platform. In the participant’s words, “There it’s just normal.” A rather contrasting view was exhibited by a German participant, who referred to Germany’s strict views on privacy as a way of explaining the limited use of Facebook. Here, Facebook was seen as unsuitable for extensive sharing based on cul- tural values that were deemed to be in conflict with Facebook vernaculars, and ethics are, on a basic level, “a social construct that reflects a group’s culture and value sys- tem” (Schultze & Mason, 2012, p. 302). Thus, unless researchers are aware of the individual users’ personal cul- tural contexts and how a given platform is perceived in that specific context, it remains difficult or potentially even impossible to judge user expectations. While user expecta- tions embody Nissenbaum’s notion of contextual integrity, such integrity remains difficult to assess in fora where there is no identifiable community, but rather networks connected by themes, causes, or events. In that sense, cultural context is a combination of platform cultures and geographically based cultures, not only the sum of cultural norms of a given digital space. Thus, in this case, geo-cultural platform norms and practices (as narrated by the participants) were used as a way of judging user expectations.

Commercial Context: Data Paths in Converging Platforms

The difficulty of defining platform culture and boundaries is further exacerbated by the relative lack of transparency in data paths, an issue that became apparent in the image scraping (DMI study). As such, privacy issues constitute by now more familiar—even typical—concerns. However, social media image scraping is a relatively new territory, for which ethical guidelines have not been developed as yet and questions remain with regard to ownership and privacy. In the case of climate change visuals, images that were col- lected were (unintentionally) largely impersonal in that they mostly did not feature people, suffering, or conflict. Even so, concerns may be raised here around consent as related to intertextuality and ownership in modified imagery. While the original production of textual and visual content may be

attributed to an individual, textual inclusion in and refram- ing of imagery is more contentious. This is particularly wor- risome as the origin and pathway of images remain somewhat obscure due to multimedia and converging com- ponents of modern social media platforms. In the case of Google, the images were publicly available, and although a public orientation may be claimed on that basis, there remains some obscurity around how Google image data are compiled. Thus, neither platform-cultural nor geo-cultural elements could be defined with certainty.

In theory, some options for tracing origin and authorship exist for images, similar to searching authors by the use of text (e.g., tweet searches on Google or Twitter). For exam- ple, TinEye and Google Reverse Image Search allow for searching the origin or publication of images, a relatively new concern in rapidly evolving digital methods. This cre- ates ethical problems in itself though. Newer digital meth- ods mean that contents, including images, become easier to trace—a benefit in the research process, but a disadvantage for the ethical treatment of data, as readers can search for the data, and anonymization therefore becomes ineffective. It also suggests that future tools may be able to extract con- texts that researchers cannot anticipate as yet. To illustrate, prior to social media data mining, researchers would realis- tically not have anticipated that any digital context or fin- gerprint their data create would eventually become easily and freely retrievable by a multitude of audiences.

These issues in data tracing follow in part from and are exacerbated by the convergence and multimedia capabili- ties of modern social media. The commercial orientation of platforms means not only that terms are set by private com- panies, but also that platforms, contents, and attached data are purchasable products. Thus, when companies merge, individual platforms may converge. This creates two prob- lems. The first problem is that the terms and conditions of data use may be subject to change if not partial merger. Thus, privacy and consent are not situated in a static and secure environment. The second problem is that users may not necessarily be aware of platform links, especially due to tendencies of platformization—an infrastructural develop- ment in which applications are built onto existing platforms (Helmond, 2015). For instance, when Facebook bought WhatsApp, the company announced that data between the two platforms would not be linked. However, later data flows suggested that the combined ownership of the plat- forms had been utilized for a more physical link and was in breach of that separation (Gibbs, 2018; Merriman, 2018).

In terms of privacy and consent, this creates a consider- able obstacle for the Facebook–WhatsApp case as the two platforms have traditionally provided different functions. For instance, in the COP interviews, participants described WhatsApp as a practical communications tool across con- texts. For the purpose of the conference, it was also described as a more immediate and user-friendly equivalent

Özkula 83

to email and therefore a pragmatic substitute for it. Facebook, however, was seen as a more entertainment-ori- ented platform, more private, and therefore more suitable to personal use. The convergence of Facebook and WhatsApp then creates an interesting dilemma. In linking two cultur- ally different spheres, it creates context collapse—the col- lapse of individual networks and the associated personal contexts (Marwick & boyd, 2014)—beyond a platform- specific network. As with the previously mentioned dimen- sions of context, this commercial context of a given platform may not be known to individuals.

This issue also became apparent in the COP study. During the summit, many interviews were arranged on WhatsApp. The tool was considered an email and telephone substitute, of particular benefit at the global summit as a way of avoiding excessive phone charges (via Wi-Fi). The sole purpose of the tool in this scenario was then pragmatic, not social. However, following the connection, these con- tacts were also suggested as potential contacts via Facebook, an unintended consequence based on the WhatsApp– Facebook link, and information disclosure that was not intended by the researcher or expected by the contact. As such, this link created an unforeseen “intervention” (word borrowed from Schultze & Mason, 2012). Thus, the com- mercial context of the platform created additional connec- tions, revealing information that the user had not consented to. This encounter highlights how context may move beyond a specific platform, due to platform connections and convergence, a commercial context that should ideally be considered in Internet research ethics.

Discussion: Barriers in Identifying Context in Social Media Data

While these examples have been provided in the hope to support the scholarly community, there are several obsta- cles in pursuing ethical research of this kind. For example, on one hand, commercial elements and aspects of platform context could realistically be identified by researchers. On the other hand, it could be questioned whether this much knowledge can realistically be expected from researchers when research itself can be time-consuming and digital space constitutes a rapidly changing environment. After all, digital space is like no other: public–private and public–commercial. This creates an interesting conundrum. The worked exam- ples suggest that researchers familiarize themselves with the elements that are unknown to them. At the same time, in social media space, researchers lack some of the more advanced technical and contextual knowledge with which to judge ethical decision-making. Thus, more user studies are needed in order for scholars to gauge users’ knowledge bases and assumptions as a way to navigate this difficult territory.

A different issue arises with assumed platform cultures and users’ geo-cultural context. Unless researchers explore those as part of informal conversations or formal interviews with the participants, these areas remain difficult to judge. After all, the indication of a home city on Facebook may not be provided or even be accurate, and the language used for a particular social media network may not represent the eth- nic, cultural, or even language background of the partici- pant. In that regard, scholars would need to judge the feasibility of exploring geo-cultural context based on their research design. Thus, even where researchers endeavor to conduct social media research ethically, they remain com- promised by the relative lack of control over their research sites in a commercially oriented environment as well as restrictions in transparency.

The commercial context of social media platforms, in that sense, creates an unprecedented ethical problematic for researchers. Although commercial regulations have long played a role for scholars researching commercial spaces and institutions, boundaries and contexts have tradi- tionally been easier to identify, and scholars have often taken the route to discuss the research arrangement within the chosen community or organization. These issues differ in social media spaces due to a range of reasons tied to the difficulty in identifying the digital space under study. Although platforms visibly exist, digital space—due to its relative detachment or disembeddedness from specific national or cultural boundaries—does not conform to tradi- tional spatial organization. Data paths and platform links are less transparent (an issue shown both in the Facebook– WhatsApp sharing and the Facebook Cambridge Analytica scandal). Digital space is also less regulated, its regulations are less widely known, and its regulations are less stable. While platforms provide terms and conditions, and par- tially outline their cultural norms and moderation practices, these rules are not as detailed and consistent as state-spe- cific cultural norms, regulations, and legal documents, or regional culturally tighter contexts. This issue is exacer- bated by the comparative lack of knowledge about those regulations in digital space.

This issue is heightened by platformization, as boundar- ies are therefore difficult to identify. Thus, the commercial context determines where and how data are published, but the lack of transparency also means that this remains diffi- cult for researchers to assess. As a result, researchers have to work with users’ assumed contexts, intended as well as imagined audiences (Litt & Hargittai, 2016). Even so, the contextual ramifications remain to some extent obscure. This is different in many offline settings due to the pro- vided context and researchers’ access to information. Physical space tends to offer more context for several rea- sons: A research site is geo-culturally situated. Even in eth- nically and culturally diverse settings, dominant cultural norms can be identified based on the specific national and

84 Journal of Empirical Research on Human Research Ethics 15(1-2)

regional context, as well as the prevailing ethnic and cul- tural representation. Unlike digital space, an offline site presents more contextual information due to the access it provides. Compared with many online spaces, there is a higher degree of “multisensoriality.” Scholars gain addi- tional cues due to richer regional knowledge including regional norms, cultural expectations, and the demographic makeup including the distribution of wealth and cultural diversity. These cues can be obtained ahead or experienced through a multisensorial experience in situ. Likewise, offline spaces potentially offer better opportunities for par- ticipant contact as living in an offline space allows for a more “natural” and again multisensorial experience of its inhabitants. For example, even without any prior contact, researchers can gain visual, audio, and behavioral informa- tion. Such information does not per se disappear in digital space—the body does not disappear but manifest itself in chosen spaces and visual culture (Rudnicki, 2017), but it is often incoherent, fragmented, and therefore difficult to synthesize for ethical purposes. Thus, although context is important, it remains difficult to identify in complex social media platforms. The recommendations that follow are therefore predominantly intended toward awakening researchers’ sensibilities toward the multiple dimensions of context in social media research.

Best Practices: Gauging Ethical Sensitivities in Social Media Research

The multiple dimensions of context presented here demon- strate that context is an unbounded fuzzy concept. At the same time, these different dimensions highlight that context is highly complex and cannot be understood as a monolithic static set of data points, but, instead, a range of different con- texts that interweave in the research setting. In that regard, the worked examples raised a number of questions that may sup- port researchers in judging the ethical treatment of their data. In particular, there are three contexts that social researchers should consider in their determination of the ethical treatment of user data. These include (a) data context (data scale, form, and added context), (b) cultural context (platform-cultural and users’ geo-cultural context), and (c) commercial context (data paths and converging platforms). While this article illustrated how these contexts are both problematic and often difficult to identify, there is merit in including and discussing them as part of ethical considerations in social media research, that is, the development of ethical sensibility toward these multiple fac- ets of context in research ethics. In particular, the DMI and COP studies suggest that there may be additional ways in which researchers can consider context in the ethical assess- ment of their research, which have been exemplified through the worked examples in the relevant sections and can be described as follows.

Data Context

As part of data context, researchers may consider not just the context of the group they study but also the context that is added to the data—its contextual embedding. While this has been a known concern for qualitative case studies, the visualizations show that such concerns also apply to differ- ent data scales. Recognizable networks or “digital finger- prints” establish valid ethical concerns. Nevertheless, qualitative research may also be risky when smaller and detailed datasets (such as ethnographic case studies) pro- vide additional context. I therefore suggest that researchers consider following questions for their added data context:

•• How contextualized is the data (data form, size, and profile information based on platform)?

•• Is the use of identifiable information such as Twitter handles absolutely essential for the research?

•• What types of users (e.g., official versus individual, anonymous versus pseudonymous, or real name) does the dataset primarily contain, and do their pro- files suggest a wide-scale public use?

•• Do additional methods used in a given study add or recreate context and therefore make participants more easily identifiable?

Cultural Context

Reflections from the COP study further suggest that users’ geo-cultural norms carry weight in ethical social media research. This has, elsewhere, already been suggested for individual online communities, for example, by Marwick and boyd (2014). The COP findings suggest, however, that such considerations should include platform-cultural and users’ geo-cultural contexts including platform (vernacu- lar) expectations. Beyond privacy settings, researchers may consider users’ geo-cultural contexts by trying to identify the cultural makeup of the group under study and what cultural expectations around privacy may be made in individual users’ geo-cultural contexts. On that basis, I suggest that ethical judgments on context should include following considerations:

•• What do platform grammars and geo-cultural plat- form practices suggest about users’ expectations of privacy and do additional methods (such as inter- views) allow for a judgment of such cultural norms?

•• What kinds of national, ethnic, or otherwise geo- graphically based cultures are prominent in the stud- ied space/community and what kinds of platform practices and expectations may reasonably be assumed on that basis?

Özkula 85

Commercial Context

Finally, as part of the commercial context, contextual ele- ments that should be considered include a given platform’s ownership, infrastructure, as well as recent and predicted mergers (media convergence) toward gauging what users may justifiably assume or not be sensitive to with regard to their data use. In particular, an awareness of platform own- ership and mergers may be beneficial in understanding how data across platforms may be considered (especially when the collected data were generated by users for the sole pur- pose of participating in the study) and support users in understanding how their data may be used. As such, researchers are advised to consider following questions in their assessment of commercial context:

•• What is the commercial context of the researched platform (e.g., platform ownership, mergers, and links)?

•• Does the research design allow for researchers to inform participants about these links (especially where researcher contact creates data that would oth- erwise not have existed)?

Research Agenda

This article problematized three contexts in ethical assess- ments of social media research: data context (data scale, form, and added context), cultural context (platform-cul- tural and users’ geo-cultural context), and commercial context (data paths and converging platforms). Beyond raising these contexts as both problematic and significant for ethical research, this article offered these worked examples in the hope that other researchers (and particu- larly early-career scholars) would find these reflections and the related best practices helpful in their assessment of ethical research.

Nevertheless, this article adds to the corpus of work that cautions about issues in social media research and rec- ommends the consideration of additional factors, without offering any universally applicable solutions. Thus, this article agrees with literature that has stressed that ethics are case-dependent. However, although the reflections and recommendations made here are neither universal nor absolute, there is merit in considering ethical concerns by the different contextual dimensions as they offer an approximate framework and suggest certain ethical sensi- bilities not (fully) addressed elsewhere. Such an approach would be particularly beneficial for early-career research- ers for whom the broad and sometimes conflicting guid- ance may offer little clarity.

It is therefore hoped that future research will build on this work by refining these categories toward the creation of

a more detailed framework for social media ethics. In par- ticular, it would be beneficial to see more studies that explore avenues for circumventing the display of digital fingerprints, empirical work on cultural expectations of pri- vacy on mainstream social media platforms, and research that addresses transparency and power in the commercial context.

Acknowledgments

The author would like to thank the Economic and Social Research Council for funding Making Climate Social (ES/N002016/1) through the Future Research Leaders program as well as the Principal Investigator of the project Dr Warren Pearce. Additional thanks go to the Digital Methods Initiative (DMI) of the University of Amsterdam where one of the projects referred to in the article was conducted. Without the DMI’s summer school and of course the colleagues who joined the project, the data collection would not have taken place and these reflections would likely not have been produced.

Declaration of Conflicting Interests

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

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: The author thanks the Economic and Social Research Council for funding Making Climate Social (ES/N002016/1) through the Future Research Leaders program.

ORCID iD

Suay Melisa Özkula https://orcid.org/0000-0002-1674-5491

References

Ahmed, W., Bath, P., & Demartini, G. (2017). Using Twitter as a data source: An overview of ethical, legal, and methodologi- cal challenges. In K. Woodfield (Ed.), Advances in Research Ethics and Integrity: Vol. 2. The ethics of online research (pp. 79-107). Emerald. doi:10.1108/S2398-601820180000002004

boyd, D., & Crawford, K. (2012). Critical questions for big data. Information, Communication & Society, 15, 662-679. doi:10. 1080/1369118X.2012.678878

Couts, A. (2011). What happens on the Internet stays on the Internet. Digital Trends. Retrieved from https://www.digi- taltrends.com/computing/what-happens-on-the-internet- stays-on-the-internet/

Dey, A. K. (2001). Understanding and using context. Personal and Ubiquitous Computing, 5, 4-7.

Fiesler, C., & Proferes, N. (2018). “Participant” perceptions of Twitter research ethics. Social Media + Society, 4(1). doi:10.1177/2056305118763366

Gibbs, M., Meese, J., Arnold, M., Nansen, B., & Carter, M. (2015). #Funeral and Instagram: Death, social media, and platform

86 Journal of Empirical Research on Human Research Ethics 15(1-2)

vernacular. Information, Communication & Society, 18, 255- 268. doi:10.1080/1369118X.2014.987152

Gibbs, S. (2018, March 14). WhatsApp sharing user data with Facebook would be illegal, rules ICO. The Guardian. Retrieved from https://www.theguardian.com/technol- ogy/2018/mar/14/whatsapp-sharing-user-data-facebook- illegal-ico-gdpr

Helmond, A. (2015). The Platformization of the web: Making web data platform ready. Social Media + Society, 1(2). doi:10.1177/2056305115603080

Hutt, R. (2017). The world’s most popular social networks, mapped. World Economic Forum. Retrieved from https:// www.weforum.org/agenda/2017/03/most-popular-social-net- works-mapped/

Leetaru, K. (2018, December 5). Social media platforms are still powerless to stop data misuse. Forbes. Retrieved from https://www.forbes.com/sites/kalevleetaru/2018/12/05/ social-media-platforms-are-still-powerless-to-stop-data- misuse/#71d7a3c5742b

Litt, E., & Hargittai, E. (2016). The imagined audience on social network sites. Social Media + Society, 2(1). doi:10.1177/2056305116633482

Markham, A., & Buchanan, E. (2012). Ethical decision-making and Internet research: Recommendations from the AOIR eth- ics working committee (Version 2.0). Retrieved from http:// www.aoir.org/reports/ethics2.pdf

Marwick, A. E., & boyd, D. (2014). Networked privacy: How teenagers negotiate context in social media. New Media & Society, 16, 1051-1067. doi:10.1177/1461444814543995

Merriman, C. (2018, May 24). WhatsApp and Facebook are shar- ing user data after all and it’s legal. The Inquirer. Retrieved from https://www.theinquirer.net/inquirer/news/3032946/ whatsapp-and-facebook-are-sharing-user-data-after-all- and-its-legal

Nissenbaum, H. (2004). Privacy as contextual integrity. Washington Law Review, 79, 119-158.

Pearce, W. Özkula, S. M., Greene, A. K., Teeling, L., Bansard, J. S., Omena, J. J., & Rabello, E. T. (2018). Visual coss- platform analysis: digital methods to research social media images. Information, Communication & Society, 0(0), 1-20. doi: 10.1080/1369118X.2018.14868

Rudnicki, S. (2017). The body, technology and translation: Mapping the complexity of online embodiment. Sociological Research Online, 22(2), 1-13. doi:10.5153/sro.4322

Schultze, U., & Mason, R. O. (2012). Studying cyborgs: Re-examining internet studies as human subjects research. Journal of Information Technology, 27, 301-312.

Seaver, N. (2015). The nice thing about context is that every- one has it. Media, Culture & Society, 37, 1101-1109. doi:10.1177/0163443715594102

Smith, A. (2014). Half of online Americans don’t know what a privacy policy is. Pew Research Center. Retrieved from http:// www.pewresearch.org/fact-tank/2014/12/04/half-of-ameri- cans-dont-know-what-a-privacy-policy-is/

Sormanen, N., & Lauk, E. (2016). Issues of ethics and methods in studying social media. Media and Communication, 4(4), 63- 65. doi:10.17645/mac.v4i4.793

Statista. (2017). Most famous social network sites worldwide as of September 2017, ranked by number of active users (in mil- lions). Statista—The Statistics Portal. Retrieved from https:// www.statista.com/statistics/272014/global-social-networks- ranked-by-number-of-users/

Swirsky, E. S., Hoop, J. G., & Labott, S. (2014). Using social media in research: New ethics for a new meme? The American Journal of Bioethics, 14(10), 60-61. doi:10.1080/15265161.2 014.948302

Townsend, L., & Wallace, C. (2016). Social media research: A guide to ethics. Retrieved from http://www.gla.ac.uk/media/ media_487729_en.pdf

Waddell, K. (2016, June). Your data is forever: The best parts of the internet are in danger of disappearing over time—But personal information has an exceptionally long half-life. The Atlantic. Retrieved from https://www.theatlantic.com/tech- nology/archive/2016/06/your-data-is-forever/485219/

Zimmer, M. (2010). “But the data is already public”: On the ethics of research in Facebook. Ethics and Information Technology, 12, 313-325. doi:10.1007/s10676-010-9227-5

Zimmer, M. (2018). Addressing conceptual gaps in big data research ethics: An application of contextual integrity. Social Media + Society, 4(2). doi:10.1177/2056305118768300

Author Biography

Suay Melisa Özkula is university teacher and post-doctoral research associate in digital sociology at the University of Sheffield. Her research focuses on social media cultures and socio- political empowerment around large-scale societal phenomena like climate change and human rights activism. Prior to Sheffield, she completed a PhD in Sociology at the University of Kent.