Ethical Dilemmas Using Social Media in Qualitative Social Research: A Case Study of Online Participant Observation

Risk of harm

There is a general consensus that social science research should do no harm (Webster et al., 2013) and while we recognise that the potential for social research methods to cause any serious physical harm to participants is low (Dingwall, 2008), our understanding of harm is as a wide-ranging concept, including serious physical harm, but also psychological and social harms (European Commission, 2010). Harms can include, but are not exclusive to users being embarrassed by the research, the research leading to reputational damage affecting a user’s social or family life or their career, online or offline harassment, or being subject to menacing behaviour or differences in how people treat them. Dawson (2014) describes a study where a researcher published the usernames of members of an online community, which led to unwelcome, additional attention and scrutiny of one of the participants, who felt they had to leave the community after being involved for many years. Williams et al. (2016) advise that the publication of content may also open up the possibility of harm if the content is sufficiently inflammatory, citing the example of racist content. These wider harms are possible in the context of social media research particularly if user anonymity is breached or if content is shared with new and unintended audiences at the risk of embarrassment, reputational damage, and/or physical harm (Townsend and Wallace, 2016).

Private vs public data?

The public nature of much social media content (Beninger, 2017) leads to a blurring of the boundaries between the public and private domains, thus creating challenges around the level of ethical regulation required for the collection, analysis, and dissemination of data gathered from different online contexts (Robards, 2013; Sloan and Quan-Haase, 2017).

To aid decision-making in this arena, much of the debate has centred around the public versus private nature of this type of data, addressing issues around defining privacy in this very public domain, particularly when access to content is unrestricted (Beninger, 2017; Woodfield and Iphofen, 2017). Accordingly, a general distinction has been made in the literature between social media data that are publicly accessible (and potentially therefore ethical for research use), and social media accounts that make use of restricted permissions with the highest possible privacy protections (thus potentially unethical to use for research purposes without consent) (Woodfield and Iphofen, 2017). Based on these distinctions, Facebook and Instagram content can be regarded as private, as users can control who views their data through privacy settings (Kosinski et al., 2015). This is despite some Facebook data (namely basic Facebook profile information) being publicly available and indexable by search engines. Conversely, the majority of Tweets are in the public domain and accessible by anyone, even without a Twitter account. Although Twitter users can choose to protect their Tweets and restrict them to approved followers, in practice this is rare (Lafferty and Manca, 2015; Woodfield et al., 2013). However, privacy and expectations of privacy are not straightforward.

Zimmer (2010) explores some of the key issues relating to the public versus private status of data, providing an account of how Facebook data from a group of students gathered for a specific project (the 3 T project), was released and used for other purposes. Zimmer argues that the students in that study did not consent to or expect their publicly available social media data to be used in this way, and thus the researchers effectively breached their contextual privacy. That is, the assumed level of privacy is based on the norms and expectations within any given social media platform (Brown et al., 2016). For example, sharing with friends on Snapchat and posting on a chatroom forum are likely to be governed by different information norms. There were also additional concerns expressed about the potential for identification of individuals through the combination of the two data sources, emphasising the blurring of the public/private line (Zimmer, 2010).

The status of data as public or private is not definitive or static, since the affordances of social media enable text and non-text content to be recirculated, commented on, shared, and changed. The effect is that in some cases, data that were originally been considered private, have come to be made public – a form of data described by Lange (2007) as publicly private/privately public. Facebook’s permitted data use policy informs users that information and content shared by users but created by another person is not under the control of the person who originally posted the content (Facebook, 2016a). This policy is effectively making private content, shared by others, public. In addition, the ‘tagging’ of friends (the identification of someone else in a post, status update or image), makes them visible to the wider social network of the user posting the content, connecting users who would not otherwise have been connected (Altshuler et al., 2013).

In this contested area of private versus public data, some scholars have further distinguished between what data are ethical to use and what are not based on the type of data rather than the restrictions available. For example, user generated social media images (e.g. pictures or videos) are generally considered to be more personal and more sensitive than textual data (Humphreys, 2016). Although these images maybe publicly available, there are additional issues to be considered which render them private, such as individuals being identifiable, additional individuals may be included, or the research may be of a sensitive topic, for example, mental health or sexual violence (Humphreys, 2016; Williams et al., 2017b).

Given the complexity of the public/private nature of social media postings, debate continues around the use of social media data for research purposes. For Beninger (2017), this debate raises the issue of data ownership. She distinguishes between legal and moral ownership of data and argues that regardless of legal ownership, moral ownership of social media content is with the user. In her study on social media users’ views on the ethics of online research, some users were uneasy about their data being used for research and concerned that their data may be misrepresented and taken out of context. For these reasons, Beninger’s participants did not regard their data as public (Beninger, 2017). Studies on Twitter users’ views of their data use, conclude that the context of the original tweet is important for understanding user expectations of privacy (Fiesler and Proferes, 2018; Williams et al., 2017a).

Consideration around the private/ public aspects of data can assist both the researcher and institutional ethics committees to establish the level of ethical regulation required. However, where studies gather data through virtual social networks this becomes problematic. A priori definitions of data as public or private may be compromised because the data content may change through links to other interactions, evolving and recirculating in the virtual space and visual images may include non-research participants. Hence, a ‘situational ethics’ approach maybe most appropriate in this setting (Kosinski et al., 2015).

Informed consent

A further key ethical principle in social science research relates to informed consent. This relates to the provision of clear explanations of all aspects of study participation, enabling potential participants to completely understand the participant role, to place trust in the researcher and subsequently secure agreement for participation in the proposed study (Wiles et al., 2005). However, as Salmons (2017) points out, the traditional consent protocols based on conversations between researcher and participant (and which are usually signed before the onset of the study) are more difficult in practice in the digital context. Consent in this space is complicated by the blurring of boundaries and associated ownership of data (Beninger, 2017), linked to the earlier discussion around how social media content evolves and re-circulates (Halford, 2017). Furthermore, building trust between the researcher and the participant – important for gaining consent – is also challenging in the virtual world (Salmons, 2017) and practically impossible in the case of ‘big data’ research (Halford, 2017). However, studies of users’ views on the use of their social media content suggest that some users want to be asked for consent (Fiesler and Proferes, 2018), and they expect anonymity when research is published (Williams et al., 2017). Salmons (2017) suggests that, similar to researchers in offline settings, researchers operating in online spaces need to build trust by conveying messages to potential participants that exhibit credibility and trustworthiness. In the context of data collection through virtual social networks, this can translate into the creation of an online identity that demonstrates and displays both credibility and trustworthiness (Salmons, 2017). For example, if a potential participant undertakes an online search of the researcher and fails to discover any evidence of their existence or expertise this is likely to negatively affect their credibility. Credibility can be established through the creation of an online identity that links to the research, a relevant research institution and perhaps, an online blog, webinar or other studies undertaken by the researcher (Salmons, 2017). In this way, potential participants who search for the researcher will find a wealth of information that conveys credibility and authenticity.

Social media researchers using online participant observation methods share some similar issues relating to informed consent as those undertaking other more traditional types of longitudinal studies. The need to regard ‘informed consent’ as an ongoing process, which is continually revisited and contextualised throughout the study (Salmons, 2017), is a key aspect of research in the online space. Similarly, the BSA (2002) recommends that where the research is conducted over a period of time consent should be regarded as a process, which includes renegotiation. This is particularly important where research is over a prolonged period and participants may forget they are being studied as can easily be the case with online research (BSA, 2002). For this reason, it is important in online observational studies that consent is continually revisited and renegotiated.

Anonymity

Ensuring anonymity is a particular challenge in relation to publication of research conducted in the online world (Dawson, 2014; Zimmer, 2010). This is challenging, since we know from studies of users that they are most comfortable with their content being used for research purposes when that content is anonymised (Fiesler and Proferes, 2018). Anonymisation procedures are frequently problematic when using social media data in any study as there is the potential for the publication of social media data to result in it being traced, revealing the identity of the user and exposing them to the risk of harm (Townsend and Wallace, 2016; Williams et al., 2016). This is particularly concerning when data are sensitive or users are from vulnerable groups.

However, one of the issues associated with maintaining anonymity online is the traceability of content, comments, statuses, and quotes (Beninger, 2017; Woodfield et al., 2013). Search engines have become so effective that it is possible to trace much content back to its origin using a simple search function (Johnson et al., 2013; Roberts, 2015). This traceability highlights the potential for identifying both the online profile of the user and their offline identity (Zimmer, 2010). To prevent the identification of research participants to third parties, Robards and Lincoln (2017) made the decision not to quote directly from the research participants’ profile pages in their study. A further option is to paraphrase parts of quotes or posts that are obviously searchable, such as names, places, and sometimes text (BPS, 2013). This type of content editing has been challenged as this has the potential to change the meaning of the original content, and affect the authenticity of the reporting (Johnson et al., 2013). The traceability debate is further complicated where both text and image-based data are collected, since images are less conducive to such alterations. In addition, social media organisations are using images and facial recognition software to identify individuals (Facebook, 2016b), another factor mitigating against changes to images. Traceability is enhanced where metadata (e.g. location and author data) are stored in some images (Johnson et al., 2013), and where some social media platforms require content to be reproduced in its original format and attibuted to the username (Sloan and Quan-Haase, 2017; Townsend and Wallace, 2016).

We now move on to the empirical study, and explore the emerging ethical concerns.

Study overview

The ethical dilemmas and concerns discussed relate to an in-depth public health PhD qualitative study drawing on both social media data and group interviews, collected from mixed gender friendship groups over 14 months in the North of England. Three friendship groups of young people (total 23 people, aged between 16 and 21 years) were recruited to the study through offline methods using a snowballing approach. These initial contacts subsequently functioned as gatekeepers to wider networks of young people, enabling friendship groups to be recruited to the study. The research was discussed with potential participants in face-to-face meetings, with study information provided, and consent was established. Consent was subsequently re-confirmed through social media after 24 hours. During the 14-month period of the study each group participated in three or four interviews, with each discussion lasting up to 75 minutes. Social media content from Twitter, Facebook, and Instagram including visual images, hashtags, microblogs, online conversations, and memes relating to drinking practices, competing practices, and interconnecting practices was collected through online participant observation through the researcher linking to the participant’s own social networks.

Interpretation of the data involved the integration of all source data using NVivo software. Thematic analysis was used to identify and analyse patterns of meaning (Braun and Clarke, 2006) and analysis of visual images drew on the work of Banks (2008), who proposes the notion of an internal narrative focusing on content and an external narrative referencing the context of the image. The empirical findings of the study will be available in forthcoming papers.

This particular ethical journey started in a similar way to those of other researchers, with an anticipatory and pre-study review of ethical considerations. Following this, the study then became the subject of ethical scrutiny by the formally constituted institutional Faculty Ethics Committee (FEC). The FEC undoubtedly had considerable expertise and knowledge from which the study benefitted. However, new ethical issues arose within the study which related to the use of social media; these were either unforeseen during the pre-study review or were identified during the course of the study through wider reading. The contemporary nature of using social media as a data collection tool in research, combined with the continually changing nature of social media affordances, meant that the ethical challenges encountered were outside the practical experience of the FEC.

The following sections of this article provide a reflection of the ethical problems related to the use of social media and from joining the participant’s social networks that arose during the study, addressing the four areas of ethical concern discussed in the literature review and exploring how these issues were addressed.

Addressing the risk of harm

Particular safeguarding issues arose relating to the observations of the virtual everyday contexts of the drinking practices of participants. For example, it was deemed likely during the anticipatory review that data would show participants who were younger than the legal drinking age consuming alcohol, and there was a possibility that other illegal or illicit practices may also be exposed. While these concerns also feature within offline studies, the use of social media as a data collection platform has an added dimension of revealing activities that may show the participants and/or others at risk of harm in real time (Moreno et al., 2013). This is particularly relevant with the advent of Facebook live, Twitter live, and Instagram Live where users can stream live, unedited, and unregulated video footage (Hern, 2017). The use of these online spaces may have resulted in the researchers observing these events as they are happening. As such the researcher, similar to other professionals working with young people, has a duty of care and responsibility to respond accordingly (BSA, 2002). The ethical challenge linked to the nature of an appropriate response in an online space has been recognised previously by other researchers (Barnes et al., 2015; Johnson et al., 2013). Barnes et al. (2015) collected Facebook status updates and made ethical decisions on a case-by-case basis, using the knowledge and experience gained by their lead researcher as a school teacher to support their decision-making. Similar to Barnes et al. (2015), we chose to use this type of ethics-in-practice approach, assessing situations as they arose on an individual basis, the likelihood of harm happening, and gauging the seriousness and severity of any illicit activity. Participants were informed that there were limits to confidentiality; should the lead researcher consider there to be a likelihood of serious harm to themselves or others, then they would be obliged to report it. In making the decision whether significant harm was likely, we took into account who else was aware of the situation (such as venue staff, parents, and the police), the support available to the young person at the time from friends and family, and whether the young person or someone else was in imminent danger. No incidents were deemed serious enough to require an intervention to secure personal safety.

A further ethical dilemma related to protecting the integrity of the data, while also protecting the online identities of gatekeepers and limiting the possibilities for harm to the lead researcher. Approval from the FEC was dependent on the creation of an online research project identity, considered to be an effective way of protecting the integrity of the data and reducing the likelihood of harm to both participants and the researcher. However, creating an online research project identity created both practical issues and additional ethical dilemmas. Practically and ethically, creating a unique research account for both Twitter and Instagram was easily accomplished, as both social media support the use of pseudonyms and the only requirement is a valid email address not currently linked to an alternative account. A credible online research project identity was therefore created on Twitter and Instagram using a university email address, and both accounts were closed on completion of the study.

Two options were investigated for setting up a project-based Facebook account. The first option was setting up a Facebook page similar to that of a business or company. However, this was not found to be feasible as there was no facility for accessing participants’ status updates or newsfeed from the page. The second option was setting up a personal profile using a fake account. This had the outcome of meeting the FEC requirements, but raised the ethical dilemma of contravening Facebook rules as using a fake account contravenes Facebook policy (Facebook, 2016c). The use of a fake identity also raised the ethical dilemma of potentially misleading the research participants as to the identity of the researcher (Carter et al., 2016), and thus not exhibiting the credibility and trustworthiness that Salmons (2017) advocates for creating a research relationship.

Hence, although we agree that covert research methods should not necessarily be dismissed as being intrinsically unethical (Calvey, 2017), in this study building a relationship and rapport with the participants was important to the success of the project, and the creation of a false identity (one potentially uncovered by something as simple as a brief look at the University website pages) may well have reduced the credibility of the researcher. Furthermore, Elovici et al. (2014) argue that fake accounts create unnecessary costs for the provider, as they police this policy, and even have the potential to lead to a prosecution. It was this policing that ended the deception as an automatic alarm was triggered when connecting with participants and Facebook froze the fake research profile account. In this instance, we explained to the participants why a false research identity had initially been set up which did not negatively impact on the study.

The remaining option was to use a personal Facebook account linked to the lead researcher. However, this led to further ethical challenges; research data could be compromised and the identity of the researcher’s friends, family, and ‘gate keepers’ for the research project could be exposed, as they would become visible to the research participants (a potential outcome of users becoming connected through shared links with the researcher as a third party) (Elovici et al., 2014). This potentially fuels the blurring of the private/public boundaries in the research setting (Johnson et al., 2013), and was complicated by the fact that the gatekeepers were bound by professional codes of conduct that governed their use of social media. Gatekeepers had requested not to be linked to the young people they had introduced to the project. Although the ethical principle of protecting the private lives of gatekeepers within research is not necessarily different from other types of qualitative research, new and different safeguards to protect their online information were needed. We proposed to solve this ethical challenge by ‘unfriending’ all existing Facebook friends (for the lead researcher) and only accepting research participants as friends. However, this uncovered a further ethical complication, with the emerging possibility that research participants would have access to researchers historical data thus again compromising the privacy of gatekeepers and the researcher’s friends and family (Facebook, 2016a). The chosen option, on balance, was for the researcher to close their own personal Facebook profile, open a new profile using the researcher’s real identity but including details of the fake research identity as a ‘nickname’ and only accepting participants as ‘friends’.

Addressing the public vs private data dichotomy

In this study, the participants had given informed consent to their tweets, Facebook content, and Instagram posts all being used as data. However, the issues raised by Halford (2017) relating to the evolving nature and recirculation of social media data did raise ethical concerns within this study, particularly evident with Facebook content.

The sharing nature of Facebook meant the project had access to data that was continually evolving as a result of additional content (e.g. comments, likes, tagging, and shares) posted by the participants’ wider social networks (Barnes et al., 2015; Robards, 2013). These data potentially gave a deeper insight into the participants’ own postings but also posed an ethical challenge around its use.

According to the literature, these data could be considered ‘private’ as it was posted on a social media platform that has restricted permissions. However, as the researcher could see the content, these users had not made use of the highest privacy restrictions and had not restricted posts to be viewed only by ‘friends’. So the question we faced when examining user posts was ‘is this a “public” or “private” posting’? Facebook’s own terms and conditions state that each user owns their own content and has control over how that content is shared. Although content posted by participants’ friends and then shared by the participants is ‘owned’ by the user, sharing it according to Facebook terms of use. However, the question still remains how ethical it is to use these data. In addition, the evolution and recirculation of the content was deemed important to the study, since it provided information on the performativity of alcohol consumption practices online.

We decided to only use content from non-participants as data when that content directly related to the research participants’ online performance of drinking or competing practices, and when the identity of that non-participant could be protected. This position is similar to the ethical conclusions reached by Kosinski et al. (2015), who propose that the use of data generated or reference to content from non-participants is acceptable, if analysis is aimed exclusively at study participants (Kosinski et al., 2015: 553).

Addressing informed consent

Two ethical challenges relating to consent and the use of social media were highlighted in the study: (1) negotiation of informed consent, and (2) informed consent and data ownership.

Negotiation of informed consent

Similar to other researchers who wish to gain institutional ethical approval, the researcher followed internal institutional guidance for a formal, paper-based consent procedure, which included providing accessible information and consent materials for potential participants. This process was agreed by the FEC who also noted that young people in a group situation may consent to participate in a research study, even if they do not wish to participate, as research is affected by its particular social context. Consequently, consent was given at an initial face-to-face meeting with participants, and then followed up by individual confirmation of consent within 24 hours. All the young people recruited to the study were given the researchers’ contact details, and encouraged to make contact with any concerns they had, or if they wanted to withdraw from the study for any reason. Since participants were recruited to the study through face-to-face contacts, gaining informed consent took the form of more traditional consent protocols with the researcher able to build rapport and trust during this time. Corroboration of consent was accomplished with the use of social media within the agreed timescale, with no participants declining.

While the FEC did not require adherence to the BSA recommendation that research over a prolonged period of time consent should be regarded as a process, which includes renegotiation, the researcher adopted this approach. At the beginning of each group interview with the research participants, we revisited the study aims and our expectations around participant contribution and involvement. The participants were then asked to confirm their continuing involvement in the research. Revisiting their consent in this way was of particular importance as the virtual lives of the young people were observed on an ongoing basis through social media platforms. The importance of renegotiating consent became apparent when the researcher contacted one of the participants to arrange a meeting, raising the issue of the participant having ‘fallen out’ with another study participant, and to enquire how they wished to proceed with the study. The participant was confused, and had to be reminded that she had posted about the disagreement on Facebook. The participant had either forgotten the information had been posted or alternatively had not remembered that the researcher was part of her online social network. Either way it demonstrates how easy it can be for participants to forget they are participating in an online research study.

None of the young people in the study withdrew their consent to participate in the study. Rather those who no longer wished to continue simply chose not to turn up at the pre-arranged times, or did not commit to arranging a meeting time for the interviews. The latter circumstance was the case with a complete friendship group of seven young people who did not commit to a further meeting; this was despite meeting with the researcher, signing consent forms, re-consenting 24 hours later and accepting the researchers ‘friend requests’ on social media. Their silence was interpreted as an indication of withdrawal of their consent to participate in the study, and that meaningful informed consent had not been given (Bourke and Loveridge, 2014). However, young people who had given informed consent and then participated in at least one friendship group discussion (before absenting themselves from later discussion groups) were considered to have consented to participate in the study. This assumption was confirmed through the social media conversations when reasons for not attending interviews included leaving the college where the discussion groups were held, and no longer being in friendships with other participants. The combination of online data collection with more traditional data collection methods proved a useful combination in the gaining (or otherwise) of consent and its renegotiation over time.

Informed consent and data ownership

The issue of informed consent and data ownership was also relevant to this study as participant’s social media content interacted, evolved and recirculated within their social networks. The researcher was inundated with an abundance of data not only generated by the research participant, but as a result of their interactions. The previous discussions in the literature review section lead us to ask whether it can be deemed public or private, who owns the data and also to assess the potential for embarrassment or risk of damage to an individual’s reputation. In this case, the data could be considered to be public – as the user had not restricted access to it – although when shared on Facebook the affordances for higher privacy settings are available thus making it private. Also although legal ownership of the data is no longer with the original user, we agreed moral ownership is, and we therefore decided it was only appropriate to use non-participants’ data if it related to participant’s own practices, could be anonymised and did not damage any user’s reputation or be likely to cause embarrassment.

Addressing anonymity

As was noted previously, a commitment to anonymity and privacy for all participants is a key principle within social science research guidance (BSA, 2002; SRA, 2003). Although we knew the offline identities of the participants, the question of how to protect their anonymity was raised during the course of the study.

In an attempt to prevent a breach of anonymity while maintaining authenticity and meaning in this study, the following actions were undertaken. First, all names, places, faces, and other obvious searchable terms in any content used in the reporting of this study were blanked or pixelated, and images were checked for metadata (BPS, 2013). This action was intended to preserve anonymity while not changing meaning or affecting authenticity. Second, the study participants were informed about the risk of being traced from any quotes or data used and participants were asked to consent to the use of their social media content in the reporting of the research (Dawson, 2014). Random searches of the edited quotes that were planned to be used were conducted using two different search engines and the users were not subsequently identified (Barnes et al., 2015). In addition, a decision had to be made about the use of tweets in published material as the reproduction of participant’s tweets alongside their user names would effectively expose their offline identities. For this reason, the decision was made to not reproduce tweets in published material. However, the changing nature of online technologies raises the possibility that published data may be traceable in the future. In addition, in our study research participants who did not necessarily know each other became visible to each other and therefore their anonymity could not completely be preserved. For these reasons, both Dawson (2014) and Beaulieu and Estalella (2012) suggest promises of complete anonymity to research participants cannot (and should not) be made, since this is impossible to deliver on.