Social media are any Web-based computer-mediated tools to cocreate, share or exchange information, ideas, pictures or videos in virtual communities and networks (such as message boards, social networks, patient forums, Twitter, blogs, and Facebook) [1]. The availability of social media opens up new avenues for researchers to easily collect data, especially from sources that may have previously been difficult to access. This has led to a massive surge in social media analytics (whereby posts or chats are analyzed via qualitative methods or aggregate numerical data collection). The order of magnitude of data and the speed with which it is made available (approaching real time) make social media a potential tool to revolutionize health research [2]. Health-related social media analytics has taken many forms, including drug or product surveillance, monitoring disease or health patterns, or views or experiences of patients. Examples are pharmacovigilance (such as the discovery of adverse events), monitoring outbreaks of the flu epidemic, illicit drug usage or suicide patterns, and views on vaccinations and health service quality [3-29].

However, these new research avenues are not without ethical challenges [30-34]. In common with other research, potentially difficult considerations surround the purpose and value of the research, benefits and harm to participants, as well as privacy, informed consent, and confidentiality. However, Internet research is very different from traditional research and as such brings about many different ethical challenges. Whereas procedures are well established for obtaining ethical approval for traditional research, how far these can be transferred directly to Internet-mediated research is difficult to decipher. Whereas the ethical issues of social media research have been much debated [34-37], the attitudes of social media users (either posters or lurkers) and researchers have rarely been sought [38-42]. Researchers currently seek guidance from a wide variety of sources, such as individual institutions, research supervisors, subject specialist guidance [43], and increasing guidelines proposed specifically for research using social media [39,44-49].

This systematic review summarizes the existing research that has evaluated attitudes on the ethical considerations of research using social media. This will help to contribute to, and consolidate, current research practice as well as to clarify those ethical issues most pertinent to the public and researchers. This, in turn, will help guideline developers to formulate evidence-based guidelines for researchers conducting research using social media.

This study aimed to systematically review the research evidence that has evaluated attitudes of social media users, researchers, and other stakeholders on the ethical considerations of using social media as a data source for research.

Due to the anticipated dearth of studies specific to health-related research and the potential for generalizability of other research, any research area was considered. To reflect the qualitative or mixed methods nature of the research, we adopted SPIDER (sample, phenomenon of interest, design, evaluation, research type) for defining the inclusion criteria [50]:

S- Sample: Any sample of people (such as social media posters or lurkers, researchers, academics or other stakeholders). No minimum sample size was implemented.

P- Phenomenon of interest: Attitudes held on the ethical implications of conducting research using social media analytics.

D- Design: Any qualitative data collection methods (eg, surveys, questionnaires, interviews, observations, or focus groups) independent of the analysis conducted. Discussion papers were excluded.

E- Evaluation: Any information on attitudes to the ethical implications of research using social media. Such information may be the primary or secondary focus of the study.

R- Research type: Qualitative (such as interviews or focus groups), quantitative (such as surveys or questionnaires with fixed responses only) or mixed methods (such as research which collates a combination of fixed and open-ended responses).

No date, language or publication type restrictions were applied to the inclusion criteria. However, financial and logistical restraints did not enable translation from all languages.

Papers that are non–research based such as discussion papers were excluded. Such papers have been summarized elsewhere [35].

Research on the ethics of individual “look-ups” were excluded, for example, employers seeking information on employees or prospective employees, parents viewing their children’s posts, and health professionals seeking information on patients (or vice versa). Research on individual privacy or security issues such as fraud, cyberbullying, grooming, and child protection were also excluded.

A total of 16 databases and 2 Internet search engines were searched in addition to handsearching journals and conferences (Multimedia Appendix 1). Databases were carefully selected to reflect the multidisciplinary nature of the review. Other methods included reference checking all included articles and any related systematic reviews, citation searching of key papers on Google Scholar and Web of Science, and contacting authors and experts.

The search strategies contained 3 facets—social media, ethics, and qualitative research. A date restriction of 1996 onwards was placed on the searches as blogging first began in 1997. No language limits were placed on the searches. The Embase search strategy is contained in Multimedia Appendix 2 and was translated as appropriate for each database.

The results of the searches were entered into an EndNote library and duplicates were removed. The titles and abstracts were screened by 2 researchers independently (SG and GN). The full text of any potentially relevant articles was assessed for eligibility by 2 researchers independently (SG and GN or SA). Disagreements were resolved by consensus based discussion and, if necessary, a third reviewer.

Information was collected on the research question, the numbers of respondents, the characteristics of the participants (such as age and gender), the methods used to ascertain attitudes (such as interviews and survey), sampling methods or survey distribution (such as email and snowballing), questions or methods used to ascertain ethical considerations, and key findings. Data were extracted independently by 2 reviewers (SG and SA). Any disagreements were resolved by discussion or a third reviewer where necessary.

Two reviewers (SG, SA) conducted independent quality assessment using the methodological assessment tool—quality of reporting tool (QuaRT). This involves checking the reporting quality of the articles using 4 elements: (1) the question and study design, (2) recruitment and selection, (3) methods of data collection, and (4) analysis. Studies were categorized as “adequately reported” when a “yes” had been assigned against 2 or more criteria or “inadequately reported” where a study was assigned a single “yes” response, or no yes responses [51]. Any disagreements were resolved by discussion or a third reviewer where necessary.

Although the quality of the reporting of the included papers was assessed, no quality threshold was implemented. All studies which provided insight or contributed to the analysis were included.

We chose an inductive theme analysis with descriptive analysis rather than a framework approach. We acknowledge that a framework approach is an equally valid approach. However, we did not identify any framework completely compatible with our intended purpose. For instance, the most relevant framework identified was restricted to Twitter [52]. In addition, framework analysis has a risk of suppressing “interpretative creativity” and thus reducing some of the “vividness of insight” [53].

Inductive thematic synthesis aims to identify salient themes via coding of the data without the use of a preexisting coding frame, or any preconceptions held by the analysts [54] . We undertook coding in QSR NVivo Pro 11 by assigning text on a line-by-line basis to nodes developed by one author (SG) and then checked by a second author (SA). An aggregation approach to the synthesis of the data was applied with data from each study extracted and grouped together to form themes with supporting quotations. Finally, interrelationships between themes were assessed and organized into a structure to produce synthesized findings.

The enhancing transparency in reporting the synthesis of qualitative research (ENTREQ) statement of 21 items was used to report the stages of this review [55].

Of the 3340 records (4636 before duplicates removed) identified by the original searches in February 2016 and a further 555 unique records by update searches in July or August 2016, 132 full-text papers were retrieved of which 112 were excluded. Excluded studies were mostly concerned with personal privacy or security (such as bank details), or with individual “look-ups” (such as seeking information on a particular person; Multimedia Appendix 3). Overall, 17 studies (from 20 publications) met the inclusion criteria (Figure 1).

Of the 17 studies, 12 explored the ethical concerns of social media users [38,40,56-67] and 5 asked researchers (mainly academics) [42,43,68-71] (Table 1 and Multimedia Appendix 4). The earliest study was published in 2001 and the latest in 2016, with 12 studies published after 2012.

Numbers of study participants were reported in 16 of the 17 studies. The smallest study involved 26 respondents and the largest study 2260. The total number of participants was over 5453.

In 7 studies the participants were from a single country: the United Kingdom (4), the United States (2), or Australia (1). The other studies either explicitly indicated an international participant coverage (3 studies) or implied an international coverage of participants (7 studies)—most of which used social media for recruitment or study observation.

8 studies used surveys of participants’ attitudes, whereas others carried out more in-depth research and analysis using interviews (6 studies), or focus groups (4 studies). A mixture of open and closed questions was used. Three studies used observational or experimental techniques; with an original design devised by the authors (4 studies used more than one technique).

The majority of the studies did not indicate the demographic details of the respondents; however, in those that did males and females were well represented. Some studies specifically targeted young people [38,59,65].

5 studies were specifically concerned with health-related research [38,57,64,66,69], whereas the remaining studies tended to be more general.

Only 3 studies (from 4 reports) were assessed as “inadequately reported” [42,56,67,71] (Multimedia Appendix 5). The contribution to the synthesis of these 3 “inadequately reported” studies was assessed as being limited. These 3 studies did not impact on the presence of concepts within the synthesis and had only a marginal effect on the detail within the concepts.

Whereas some included studies reported summary data from surveys or questionnaires, the majority of studies presented author interpretations supported by verbatim extracts from participants.

Some responses spanned multiple themes and attempted to categorize overall reactions to social media research use. Many closely related themes underpin the diverse attitudes exhibited by the study respondents. The framework we adopted from our emerging themes was: the 2 actors (researchers and social media users), managing their relationship (consent), and framing the context (responsibilities of the social media site (Tables 2 and Multimedia Appendix 6). Some studies assessed multiple themes in a more integrated and quantifiable way and we consider these first under general reactions.

Within the arena of “researchers,” we identified themes of the perceived benefit of the research, the affiliation or type of researcher, the validity of the methods used to conduct the research, and the risks to the researchers themselves. Those themes related to the “social media users” were concerned with the risks involved with particular concern for vulnerable groups. Linked to these risks were the intended purpose of the social media poster and their ability to self-regulate through personal censorship or privacy settings. The next theme related to “consent” and the importance of and difficulties of informed consent and research disclosure. The last theme was regarding “responsibilities”—either via the social media site (including the platform and site administrators) or legal requirements (Tables 2 and Multimedia Appendix 6).

Three studies attempted to quantify general reactions to social media research. Moreno found that over half (56.1%, 74/132) of university students “strongly like” or “somewhat like” the concept of using Facebook for research by university researchers, nearly a third (28.8%, 38/132) were “neutral” and only 15.2% (20/132) were “unsure or uneasy” or had an “overt concern” [38].

Williams found that 37.2% (N=564) of social media users were “not at all concerned” with their social media information being used by university researchers, whereas 46.4% (N=564) were “slightly concerned,” 11.2% (N=564) “quite concerned,” and 5.2% (N=564) “very concerned” [67]. Evans found that 60% (N=1250) of social media users felt that “sharing individuals’ social media data with third parties, such as the government or companies, for research purposes” should not happen and 32% (N=1250) felt that “sharing overall numbers of social media data with third parties, such as the government or companies, for research purposes (but not linked to individuals)” should not happen [59].

Respondents in Evans may have been less positive because they were given use of examples relating to government and companies rather than university researchers [38,59,67]. The more positive attitudes in Moreno may be linked to surveying younger people [38,59,67].

This systematic review demonstrates the need to understand a complex array of interrelated and challenging factors with respect to ethical considerations in social media research. Diverse important issues and concerns remain to be addressed, but consensus proves difficult to achieve. Respondent views varied considerably from complete “animosity” to being “overly positive.” In between were attitudes of conditional acceptance (for instance social media research only to be undertaken if informed consent is gained or the site is public) and complete apathy. In addition, even views from the same respondent were conflicting with many being well aware of their contradictory views. The comprehensiveness of this review enables us to develop a broader view of the contrasting issues rather than being restricted to findings from a single study or single population that is not as transferable.

The differing attitudes can partially be explained by the heterogeneity of respondents, their different understandings of social media research, and different methodologies used in each included study. In particular, Twitter users and the younger generation tended to be more accepting of social media research. Research using social media is a relatively new phenomenon. Some respondents had little understanding of social media research particularly in the earlier studies or were unaware of the range of examples. Social media users were more likely than researchers to evaluate the perceived benefit of the research and the validity of the research when considering the acceptance of social media research. Researchers, on the other hand, were also more likely than social media users to be concerned about risks to researchers themselves.

There was general agreement on several issues. Whereas there was strong support for social media research that is for the “good” of society (and thus greater acceptance of university researchers)—this should not offer researchers carte blanche or an opportunity to ignore ethical principles. There are different definitions of “beneficial” research and some people could still be at risk of harm. Respondents acknowledged possible risks with these being more complex for such groups as, children, adolescents or vulnerable adults [38,72-75] or for discussions on sensitive topics. Risks to social media group dynamics and freedom of expression have been highlighted in the literature [36,76] and this review demonstrates the strength of feeling about the importance of keeping safe and supportive environments for people to post on. The potential benefits, therefore, should still be weighed up against any potential harm. However, often researchers considered general principles such as “do no harm” as difficult to follow [70].

Respondents were much more likely to support the use of numerical aggregate data (such as overall statistics) than qualitative research involving quotes or interpretation of quotes. This view was almost unanimous [40]. Respondents agreed overwhelmingly that the terms of service of social media platforms are infrequently read and should not be relied upon. This finding may be related to newsworthy cases where terms of service were relied upon. Over time much has been learnt through trial and error. As a consequence, platforms such as Twitter and Facebook have changed their terms of service with much tighter control over external research [76,77]. However, this review does not suggest that users feel this is the best way forward.

Respondents also thought that the conduct of social media research should not just obey the principles of the law (although there was little agreement as to what the law is) but also follow ethical principles and a moral obligation.

Many issues encountered little agreement and conflicting attitudes. The boundaries between public availability and privacy were particularly complex. The public nature of social media has been used to support current practice, whereby a minority of studies apply for institutional ethics board or other approval and most studies do not mention ethical approval [43,73,78-80]. Social media research, therefore, often involves a lack of informed or valid consent with group members often unaware that they are being monitored [76,80,81]. Whereas some social media users seemed happy with this lack of consent, quoting the public nature of social media and the potential to self-regulate, others strongly opposed this on the grounds of privacy, original intent of the post, risks to users, and ownership of the data. These arguments are similarly rehearsed in the literature [34,43,78,82-86].

Rooted within the concepts of risk to users are the issues of privacy and traceability or anonymity of the poster [34,38,39,41,48,71-73,77,81,87-93]. Whereas some respondents were happy for use of anonymous posts, others wanted to be cited (as is the case for published works). Both users and researchers displayed a lack of understanding of the difficulty in upholding anonymity [80]. In 2006, deanonymization of social media users was carried out by journalists of the New York Times [94], and in 2011 data aggregated and allegedly anonymized by researchers from Harvard were deanonymized [83,95]. Covert participation inside Web-based communities has also occasioned controversy [96].

Respondents lacked clarity with regard to informed consent, when it should be implemented, and how [42,71]. However, users agreed that it is unacceptable to use names and direct quotes without consent and that it is practically difficult to obtain informed consent. The issue of informed consent was highlighted in 2012 when Facebook manipulated users’ posts in an experiment to influence people’s mood. This caused much anger and dismay among users who were unaware they were part of any experiment [97,98]. The included studies in this review, however, did not examine the ethics of manipulation of sites and this needs separate consideration.

Researchers may access content posted on the Web without interacting with the individuals who wrote the posts or even considering them “human subjects.” The “human subject” or “published author” debate, however, was little discussed in the included studies, yet, has important implications for ethical approval and is often debated in the literature [70].

The absence of an overarching consensus with regard to social media research ethics is apparent. Many complex ethical dilemmas persist [73]. Whereas researchers have found it useful to read and understand ethical considerations faced by other researchers, they find it challenging to translate approaches used in another research context to their own particular research [70]. Each research project, therefore, requires individual consideration of its ethical issues. Just as a blanket approach to the ethics of traditional research (such as surveys or focus groups) would prove unsuitable so, too, we should resist a “one rule fits all” to social media research. Researchers should consider the type of research (such as aggregate, qualitative); the nature of their topic (whether sensitive or trivial); as well as issues of anonymization, confidentiality, informed consent, privacy, and the benefits and risks involved. The ability to undertake Internet research may depend on the level of trust and confidence the public have in the research being undertaken when people choose to contribute to studies or make posts publically available [77]. Koene [77] considers a possible revolt with a call for a public backlash and boycott of Internet-mediated research. This review does not suggest that social media users feel as strongly as Koene [77] suggests, but we should still be mindful of these issues.

Although the heterogeneity of included studies contributes strength to this review, this also makes it challenging to arrive at definitive conclusions. In particular, studies differed in their methodological approaches and by their presentation of results.

Social media are constantly evolving and have changed considerably since the first of the included studies was undertaken in 2001. It would be interesting to uncover how people’s perception of the ethical issues in social media research have changed; however, unfortunately the sparsity and heterogeneity of studies made it challenging to reveal any time trends.

Whereas additional non-database searches (such as citation searching and reference checking) were used in an attempt to overcome the difficulties in searching for this type of research, it is likely that some relevant studies remain unidentified. The review was also limited to papers in English or for which a translation could easily be obtained.

It remains unlikely that a consensus on the ethical considerations on using social media research will ever be reached. Each Internet research project requires an individual assessment of its ethical issues and selection of the most appropriate methodological approach.

Whereas the issues raised in this review suggest that ethical considerations in using social media for research are complex and require thoughtful consideration, this should not deter researchers from conducting social media research. Contributions from social media offer a more immediate time window than experiences documented in formally published qualitative research. This is because social media data can be analyzed immediately whereas published qualitative findings may take at least two years from data collection, through peer review and the wider editorial process, to publication. To a large degree such contributions may be unfiltered and unfettered and less subject to the influence of the researcher. They are less constrained by the temporal and spatial limitations encountered when planning and conducting qualitative research. However, social media contributions are public offerings largely written in the knowledge that they could be read by a wider audience. They offer a perspective that is often stripped, or at the very least, lacking a grounding in context and that may challenge representation and interpretation. Thus, we cannot afford to miss the considerable potential of social media research and its unique contribution to knowledge.

Guidelines for ethical conduct should be based on the available best practices and standardized to avoid discrepancies and duplication from one institution to another. This methodological review is offered to initiate ongoing discussion within the research community of how such guidelines might be formulated. It highlights the importance of properly conducted social media research of benefit to the public. It also highlights the need to consider informed consent and privacy and researchers should not rely solely on regulation but have a moral and ethical duty to consider social media users and the main purpose of social media groups.