Varieties of consent

1. Study-specific consent. Traditional research consent is study-specific, and discloses a great deal of information about the nature and purpose of a particular study, prior to data collection, in order to enable a prospective participant to decide whether to participate or not. This kind of consent will rarely be feasible in collecting large scale data from online platforms. Even when partnering with a platform that has the ability to push informed consent mechanisms (e.g., pop-up dialogue boxes with forced tutorials23andMe uses such a mechanism not for research but to secure user informed consent before they access sensitive health results (those pertaining to breast cancer, Parkinson’s, and Alzheimer’s risk).
   
   ) out to users, platforms will (understandably) be unwilling to interrupt the user experience every time a researcher wishes to collect or analyze user data.

2. Broad and blanket consent. Such a mechanism might be deployed for particular studies on rare occasions, but ordinarily, consent will have to be “broad,” in which a user consents to unspecified future research within a broad scope (e.g., “research about information consumption” or “research that can help us better understand civil engagement”), or “blanket,” in which the scope of research questions is boundless. Forced choice mechanisms (with or without tutorials and/or competence quizzes to ensure informednessHarvard’s Personal Genome Project requires prospective participants to score 100% on a quiz testing their knowledge of genetics and the risks of posting their whole genome sequence and health data on the open Internet before they are accepted into the study. NIH’s All of Us Research Program (formerly the Precision Medicine Initiative) uses an e-Consent platform developed by Sage Bionetworks that uses formative assessments of comprehension during the consent process to reinforce learning but does not require any score to join.
   
   ) could be used as one-time means to secure such broad or blanket consent (they could also be deployed on a regular basis, e.g., annually, to ensure that the user’s preferences have not changedA variant of broad consent is “dynamic consent,” which involves a two-way technological interface between participant and researcher that allows unanticipated research uses that might stretch the bounds of the original consent to be communicated to participants and also allows participants with evolving preferences to make different decisions about their data (Kaye et al. 2015).
   
   ). Whatever their legal status, platform Terms of Service (TOS) essentially never provide ethically meaningful informed consent. Virtually no one reads TOS, nor—rationally speaking—should they take the time and effort to do so, since they are largely incomprehensible contracts of adhesion (Meyer 2015).

Although broad consent is not without its critics among ethicists (e.g., Caulfield and Kaye 2009), the dominant view among ethics scholars appears to be that (well-executed) broad consent is ethically permissible. Biobank researchers have long used broad and blanket consent by availing themselves of the Common Rule’s provisions for waiver or alteration of consent, and the revised Common Rule explicitly provides for broad consent as an option for the collection, maintenance, and secondary use of identifiable biospecimens or identifiable, private information.45 C.F.R. § 46.116(d). These provisions are limited to identifiable biospecimens and data because the Common Rule does not apply to secondary research on non-identifiable biospecimens or data, nor does it apply to secondary analysis of identified or identifiable but non-“private” information. Id. § 46.102(e).

Broad consent is a policy compromise that recognizes two sets of realities. On the one hand, human tissue is expensive to collect and maintain but a critically important resource in advancing human welfare. Every possible appropriate research use of such tissue or the information derived from it cannot be foreseen at the time of collection, nor is it feasible to recontact and reconsent all tissue or data sources when new research uses are contemplated. There is also considerable value in allowing as many researchers as possible to access this valuable resource. On the other hand, broad consent preserves the ability of individuals to exercise control over their tissue and/or data by declining participation.

Broad consent should be familiar to citizens of representative democracies; in both cases, subjects agree to allow a governance body to make decisions on their behalf. When data resulting from tissue collected under broad or blanket consent are deposited into repositories for broader access by other researchers, a Data Access Committee (DAC) reviews data access requests, asking whether the proposed secondary analysis falls within the scope of the broad consent or runs afoul of any other Data Use Limitations (DUL) provided in the consent (e.g., a promise that researchers will not attempt to re-identify subjects or a restriction to use by affiliates of non-profit research organizations). Care should be taken when drafting broad consent language to ensure that laypersons and researchers share an understanding of the terms used. For instance, many researchers would understand “research related to health” to encompass the social determinants of health (e.g., educational attainment, income, zip code), but most laypersons are not familiar with that concept, leading to secondary research that does not align with user expectations.

3. Tacit consent. A final form of consentTwo other recently articulated kinds of consent are worth noting: meta consent, which invites people to determine how and when they would like to be presented with a request for consent (Ploug and Holm 2016), and dynamic consent, which entails ongoing communication between researchers and data subjects about evolving research questions and user preferences (Budin-Ljøsne et al. 2017).
   
   is tacit consent, which rests on an individual’s failure to actively opt out of passive research participation. For such consent to be ethically meaningful, it must satisfy the publicity principle: individuals must know that they have the ability to opt out, they must appreciate the consequences of not opting out, and opting out must be fairly easy. Like other nudges, opt-out consent can be an effective way of enrolling more participants in research who are not especially opposed to it by leveraging inertia or indifference, while still allowing those with clear views to easily make a different choice. Platforms could enable opt-out consent. Again, however, inviting an opt-out process in the TOS will not suffice; the invitation must be much more prominent than that. Fiesler and Proferes (2018) recommend that researchers using online data find ways to allow users to opt in or out of particular kinds of research: “This could be, for example, a flag set in the user profile or a black/white list included as part of the API. Another potential design would be to build a system that could provide public notices when data collection begins from a specific hashtag, informs users when their tweets are included in a dataset, and/or links those who have had their tweets used back to a published paper based on the results.”

When consent is not legally and (arguably) not ethically required

Several categories of data or research of relevance to ISSOD fall outside the scope of the Common Rule—and hence its default requirement of informed consent.

1. Research with non-identifiable data. First, research with non-“identifiable” data that were collected for a purpose other than a particular study (i.e., the researcher does not intervene or interact with subjects to obtain the data) does not involve “human subjects” as the Common Rule defines that term. Data are “identifiable” if “the identity of the subject is or may readily be ascertained by the investigator or associated with the information.”45 C.F.R. § 46.102(e)(5).
   
   Note that, depending on how one interprets “readily ascertainable,” this definition of “identifiable” is quite weak. Federal regulators are aware of improvements in reidentification techniques and the revised regulations require Common Rule agencies, within one year and at least every four years thereafter, to a) reconsider this definition and b) consider whether any analytic technologies or techniques should be considered to necessarily produce identifiable data.Id. § 46.102(e)(7).
   
   For present purposes, the point is that federal research regulation codifies a reasonable policy choice that treats research with non-identifiable data (however defined) quite differently than research with identifiable data, and excludes the former from regulation.HIPAA makes the same distinction, with the same consequences for (lack of) consent (“authorization”) for research use of de-identified data, although HIPAA sets a higher bar for when data are “de-identified” than does the Common Rule with respect to non-“identifiable” data.
   
   

Moreover, even if data aren’t likely to be associated with individual identities—or were already publicly available in identified or identifiable form—people may still have autonomy interests in controlling those data. For instance, data subjects may have an interest in avoiding complicity in research with which they disapprove by being able to prevent their data from contributing to such a project. The classic case is that of the Havasupai Tribe of Arizona, many of whose members entered into an oral agreement with academic researchers to give blood in order to study the tribe’s epidemic of diabetes. A written consent form, however, provided for broad use of the blood for behavioral or health research, and the data were eventually used to study schizophrenia, consanguinity, and tribal migration patterns, all of which were highly objectionable to both the participants and to the other members of the tribe who felt equally stigmatized even without have provided blood.

In the case of ISSOD, it is not difficult to imagine that some data subjects might object to what they perceive to be partisan research on U.S. politics. Some data repositories (e.g., NIH’s All of Us Research Program) have attempted to define “sensitive research” and develop processes for reviewing proposals to conduct it. The challenges of defining in advance categories of research that some (a majority?) of data subjects will find substantively objectionable, however, are formidable.

2. Research with public data. Under the Common Rule, “human subjects” are similarly not involved if researchers study identifiable but non-“private” information. “Private” information is “information about behavior that occurs in a context in which an individual can reasonably expect that no observation or recording is taking place, and information that has been provided for specific purposes by an individual and that the individual can reasonably expect will not be made public (e.g., a medical record).”Id. § 46.102(e)(4).
   
   A great deal of data of interest to ISSOD falls into this category, including non-protected tweets, Facebook posts set to “public,” public Twitter and Facebook profile information, and comments on sites such as Reddit and 4/8Chan. Although, as discussed below, some users and some ethicists believe that what constitutes meaningfully “public” data and spaces is or should be contested, these data are not “private” under any plausible interpretation of the Common Rule, and hence research with them—even with identities fully attached—does not involve “human subjects” and therefore is not subject to the Common Rule, including to its requirements of consent or IRB review.

There are some important complications to the claim that research with “public” data falls outside the scope of the Common Rule and does not require consent. First, sometimes a platform’s TOS will either deem platform data to be “private” or, more commonly, will prohibit research use of the data. One prominent advisory body has suggested that in both of these cases, seemingly public data should be considered private for purposes of the Common Rule (Secretary’s Advisory Committee on Human Research Protections 2013). Others, however, have argued that violating TOS in order to conduct research must be ethically justified, but is possible (Bruckman 2016).

A second complication is that although the Common Rule deems data to be private only if a data subject’s expectations of privacy are “reasonable,” many data users clearly do misunderstand the public nature of the data they produce. These misunderstandings may indeed be unreasonable, but from an ethical perspective, it is not obvious in all cases that researchers should take advantage of the user’s mistake, whether reasonable or not. For instance, a recent exploratory survey of Amazon Mechanical Turk workers who reported being current or former Twitter users (Fiesler and Proferes 2018) found that 61% had not previously realized that public tweets might be used in research, 43% (incorrectly) believed that researchers are not permitted to use tweets in research without user permission (because doing so would violate Twitter TOS, copyright law, or—most commonly—research ethics rules).Although the total sample size for this survey was 368, the n for most questions discussed here was only 268. One limitation of this study is that MTurkers are likely habituated to written informed consent disclosures and may incorrectly assume that all human subjects research requires such explicit, study-specific informed consent, which may have affected some of their responses about the appropriateness of researcher use of public tweets and profile information. Another limitation is that the survey did not elicit respondents’ overall preferences for research use of public tweets and Twitter histories in light of the tradeoffs involved if individual, study-specific consent for each research use of a public tweet were required.

Facebook users might misunderstand the platform’s privacy settings, which for a long time were rather non-intuitive. In some cases, users would not have contributed sensitive data had they realized how public it was and what kinds of things might be done with it.

Some scholars go further and believe that producers of “public” tweets, public Facebook posts, and the like in fact have reasonable expectations that their data will not be used in research without their explicit permission (Zimmer 2010; Williams, Burnap, and Sloan 2017). Many overlapping reasons have been offered for this. For instance, many users seem to adhere to the principle of “privacy by obscurity” (Hartznog and Stutzman 2013): although an unprotected tweet can technically be accessed by anyone with an Internet connection, ordinarily, the user with only a handful of followers can reasonably expect that it will in fact reach only a small audience. Similarly, although a user voluntarily provides each public datum, the fact that it is generally infeasible for others to aggregate and analyze her entire platform profile (perhaps discovering new, additional information about the user in the process of analysis that the user didn’t intend to share and may even have been unaware of herself) provides a sort of informational privacy. Data repositories that aggregate and organize data for easy searching eliminate or at least reduce that privacy-protecting obscurity. Similarly, depending on how public data are stored and reported in published research, research can have the effect of greatly amplifying attention to what were public-but-obscure data. Other scholars and users have, under the capacious label of “privacy,” expressed concerns about data decontextualization: studying data out of context (e.g., a tweet written in the heat of the moment, or one that is best understood in the context of a thread or of an unnamed exogenous event, e.g., political event) can portray the data subject in a poor or unfair light (Nissenbaum 2004).

A final complication is that it is unclear when a space is public or private and people do not share the same view about this. For instance, if one needs to register or create an account to join a platform, but anyone can do so (i.e., the registration process is entirely undiscriminating), is that a public or a private space? Sometimes, platforms created for particular purposes will generate expectations about how and why information is shared. For instance, someone who shares sexual preferences on an online dating site for the purpose of meeting a compatible companion is not necessarily thereby saying that that information is free for the taking, to be used by those outside that community for other purposes, simply because they registered (e.g., M. N. Meyer 2018b).

3. Exempt research. Two exemptions from the Common Rule mirror the two ways, described above, that research can be deemed not to involve human subjects (and so are subject to the same caveats described above). First, human subjects research involving “observation of public behavior (including visual or auditory recording)” is nevertheless exempt from the Common Rule (and, hence, its consent default requirement) if any one of the following criteria is met: the data are recorded by researchers in such a way that the subject’s identify “cannot readily be ascertained, directly or through identifiers linked to the subjects;” the data are not sensitive; or the data are identifiable and sensitive but an IRB conducts a limited review of the data security plan.Id. § 46.104(d)(2).
   
   Notably, with respect to the first prong, OHRP guidance advises that if the researcher and the holder of a key to coded private information enter into an agreement under which the researcher will never be given access to the key (or will not be given access until the subjects are deceased, at which point they are no longer “human subjects” under the Common Rule), then this meets the requirement that subjects’ identities cannot be readily ascertained (Office for Human Research Protections 2008). A platform could serve as this data broker, providing coded data to ISSOD (and, through it, researchers). A separate exemption is available permitting nonconsensual secondary research with identifiable, private data if the data are either “publicly available” or are recorded in a non-identifiable manner, as per the previous exemption.45 C.F.R. § 46.104(d)(4).
   
   

4. Minimal risk research in the public good. The Common Rule permits research that is no more than minimal risk to be conducted without informed consent if: the research “could not practicably be carried out” without the waiver or (when identifiable data are used) with non-identifiable data; if waiving consent would not adversely affect the rights or welfare of subjects; and if, whenever appropriate, subjects are debriefed after the research.Id. § 46.116(f).
   
   Although IRBs do not always agree about when research with consent would not be “practicable,” two common qualifying reasons are if the researchers have no feasible means of interacting with subjects (typically the case with big data research) and if informing subjects about a specific study in advance would bias the results (often the case when researchers wish to study behavior).

Some ethicists have argued compellingly that certain minimal risk research should be permitted without consent—whether or not obtaining consent is practicable. These scholars point out that although individuals and groups have legitimate interests in privacy, they also have legitimate interests in public goods such as better quality health care, and that research policy must better balance these sets of interests than it currently does. Some have even argued that citizens have an ethical obligation to participate in low-risk, passive research such as health data research (Ballantyne and Schaefer 2018) and digital epidemiology (Mittelstadt et al. 2018).

Other ways of respecting data subjects

When consent is infeasible or otherwise inappropriate, other ways of respecting data subjects as persons should be pursued. For instance, researchers can often provide notice of the research, either before or after the fact, and either study-specific or broad. Notice can also include return of aggregate results of research to data subjects. Some empirical studies of public perceptions of research have found that many (though not all) people would find notice of the purpose, nature, and/or results of research to be nearly as satisfying as consent for minimal risk research. For instance, in the aforementioned survey of MTurk Twitter users, although most thought that researchers should ask permission before collecting or using public tweets, “a number of respondents specifically framed their desire to both know about the research and to see it when it’s finished as an issue of respect. Informed consent can be seen as both informing and consenting, and for many respondents, the former would be sufficient”(Fiesler and Proferes 2018). A national survey of public perspective on pragmatic randomized trials found that although most respondents preferred traditional, written informed consent (despite the majority recognizing that the trials depicted did not pose additional risk on participants), a substantial minority—nearly 40% in one arm—not only tolerated but preferred general notification (Nayak et al. 2015).

ISSOD could require all publications based on ISSOD data to be made publicly available on the ISSOD site, accompanied by accessible lay summaries. A more ambitious version would involve a technology-enabled platform that tells data subjects exactly which publications their data contributed to.23andMe has such a feature: when logged into the platform, the user sees a message such as, “Michelle, you contributed to 17 published discoveries!” followed by a list of those publications with links to the journal (unfortunately, often not open access).

ISSOD could also host videos of lightning talks that explain to the public the major discoveries enabled by their data. Finally, in addition to returning aggregate research results, it can be valuable to return individual research results. The National Information Study that ISSOD envisions might, for instance, show each survey respondent how their answers compare to others.The Facebook app Genes for Good offers something like this. Participants are asked to complete numerous phenotype surveys, but when asked, e.g., how frequently they smoke, they are rewarded through a simple data visualization showing how their answer compares to those of others, e.g., the respondent is in the top quintile for smoking activity.

Another way of respecting data subjects and their contribution to research is by ensuring that ISSOD enables reproducible science. Reproducible science is an ethical (not merely scientific) issue because whatever risks data subjects bear are for naught if the resulting science is not reproducible. Social Science One, for instance, will require researchers who access their platform to adhere to the “replication standard”: all funded research must archive replication data files, including code, methods and metadata (King and Persily 2018). In addition, ISSOD could, similar to NIH and clinicaltrials.gov, require preregistration of hypotheses (if any) and data analysis plans and require public posting of results, including null results, within a specified time period (subject to extension) on an ISSOD site (whether or not the results are also reported in a peer-reviewed journal).

Group harm, vulnerable populations, and participant engagement

Certain populations may both perceive that they are at increased risk by having their data aggregated and shared and may in fact be at increased risk of harm. For instance, one survey study found that “[l]esbian, gay and bisexual (LGB) respondents were more likely to express concern over their Twitter posts being used in government (odds increase of 2.12) and commercial settings (odds increase of 1.92), compared to heterosexual respondents,” perhaps owing to historical online abuse of LGBT persons and/or antagonistic relationships between these communities and some governments (Williams, Burnap, and Sloan 2017). If ISSOD expects to enable research on vulnerable populations, it should consider engaging those communities early in the process of developing the platform to see whether risks can be mitigated or community representatives can be included in ISSOD oversight structures. More generally, it is a good idea to engage the public (whether vulnerable or not) in development and oversight of the research platform.

Access to data

Data repositories typically calibrate data access according to the sensitivity and/or identifiability of the data, providing different access mechanisms for different kinds of data. At one end the spectrum, the most sensitive data might only be accessible through virtual or actual data enclaves. At the other end of the spectrum, metadata and innocuous, individual-level raw data might be made publicly accessible. In between are myriad options (see Table 1 in M. N. Meyer (2018a), and Appendix) that involve trade-offs between data security, on the one hand, and transparency and fair access, on the other. For instance, it is common to restrict data access to “qualified researchers,” usually permanent faculty (i.e., those with “PI privileges”) affiliated with a research institution (e.g., Social Science One). The purpose of this restriction is that the institution can be required to be a party to the data use agreement, thereby lending that agreement some teeth. But this practice excludes citizen scientists, independent academics, and journalists, who may have legitimate interests in the data and who are likely to be more representative, politically and in other ways, of the general population whose data comprises the repository. A data repository that purports to serve the public good but is accessible only by academics may be viewed with skepticism by citizens on the political right, who may associate academia with biased, ideologically-driven research.

Data use agreements should prohibit attempts by researchers to re-identify or contact data subjects without the explicit permission of ISSOD (which can review proposals for, e.g., re-identification research).

Commercialization of research

It is essentially unheard of for data sources to be compensated for the research use of their data or to share in any profits derived from research, whether those data derive from human tissue or online behavior. Nevertheless, some people feel differently about the issue of their data in commercial as opposed to non-profit purposes, and a persistent minority of data subjects believe they should be financially compensated for use of what many regard as their “property” (Fiesler and Proferes 2018). For instance, one of several objections by some to the case of Henrietta Lacks (in which physician-investigators collected and used leftover, quasi-pseudonomized clinical tissue for research without consent, as was then and remains today legal) is that, although those physician-investigators did not profit from the research, others downstream from the initial research did—and handsomely. Some ethicists have argued cogently that sources of passively collected research materials that the source would have produced anyway and which therefore entail no extra effort by or inconvenience to the data subject (e.g., leftover clinical tissue or, in the present context, digital data already produced for the user’s own purposes) are not morally owed compensation (Truog, Kesselheim, and Joffe 2012). Still, in response to this minority public sentiment, the revised Common Rule newly requires research consent to include, where appropriate, a “statement that the subject’s biospecimens (even if identifiers are removed) may be used for commercial profit and whether the subject will or will not share in this commercial profit.”45 C.F.R. § 46.116(c)(7) (2017). Note that because the Common Rule does not apply to research with non-identifiable biospecimens collected for a purpose other than the instant research project, these new requirements technically only apply to research in which researchers intervene or interact with tissue sources to newly collect identifiable or non-identifiable biospecimens.

The regulations do not, however, provide that tissue sources must, or even ought, to share in profits.

Although compensating each data subject for their contribution is unreasonable and infeasible, it does make sense to constrain researchers from commercializing ISSOD-enabled research in ways that would threaten public access to the benefits of that research. This is especially important if, as seems likely, the public will be passively contributing data to ISSOD and bearing some (minimal) risk for doing so. Social Science One, for instance, precludes researchers who are funded by the platform from patenting their results, but it does not preclude other for-profit uses of the data (e.g., writing a profitable trade book about the research results) (Social Science One 2018).

Partnering with, versus scraping, platforms

In general, partnering with platforms provides several potential advantages. First, platforms can incorporate notice of (if not consent to) data sharing with ISSOD into their user-facing materials. Similarly, platforms may be able to and help transmit aggregate or individual results as a gesture of respect. Third, the standard rule in biobank research is that samples from participants who withdraw are destroyed and no longer used in analyses going forward, but that their data is not removed from completed analyses. By analogy, partnering with a platform could entail a mirrored research database that refreshes in real time, enabling deleted tweets to be automatically disappeared from the research platform. (The evolving content of the database would have to be accounted for somehow to enable reproducibility of the analyses.) Conversely, scraping is often frowned upon by both platforms and users, especially if the platform is quasi-closed and community- or purpose-specific, where researchers may be viewed as interlopers. This can undermine trust, setting back the goal of large scale data sharing. The primary risk of partnering with platforms is that it would be important to insulate the research from influence by the platform.

NHANES

National Longitudinal Survey of Youth (NLSY)

Framingham Heart Study (FHS)

Wisconsin Longitudinal Study (WLS)

Health and Retirement Study (HRS)

General Social Survey (GSS)

dbGaP