For the Good of the Group? Balancing Individual and Collective Risks and Benefits in Community Psychology Research

Abstract

Balancing risks and benefits is a necessary ethical task for social science researchers. Community psychologists must often consider risks and benefits not just for individual participants, but also for the group, system, or society that those individuals inhabit. Little ethical guidance currently exists for how to navigate this ethical challenge across these multiple levels. In this article, we use a case example of social network research incorporating multiple levels of analysis to identify common risks and benefits and understand their relationship to each other. We conclude by discussing distinctive ethical considerations revealed by this case example.

A favorable ratio of risks to benefits is a core criterion on which human subjects research is evaluated (Emanuel, Wendler, & Grady, 2000), and many other criteria (e.g., respect for participants, informed consent) are ultimately strategies in pursuit of this goal. Community psychology’s commitment to multiple levels of analysis means that contexts are often the targets of research, so contextual (i.e., group, system, societal) risks and benefits must be considered. However, ethical guidance typically focuses on potential risks and benefits to individuals—and the benefits to society— without discussing risks and benefits to groups or how to effectively weigh those risks and benefits (see Weijer & Emanuel, 2000 for a notable exception). To inform these decisions, this article offers a narrative example of an ethical challenge in which we (a) identify risks and benefits across levels of analysis and (b) consider cross-level relationships between risks and benefits. We conclude by discussing distinctive ethical considerations that arise when weighing multilevel risks and benefits.

Case Example: Sorority Networks

Our efforts to understand sorority networks as a context for recovery after sexual assault provide an example of balancing individual and collective risks and benefits. Like any research project, this effort raised a set of ethical challenges that are, to a degree, idiosyncratic to characteristics of our setting, research questions, participants, IRB, and us as ethical decision-makers. Thus, although we offer considerations for social network analysis specifically, we also hope to illustrate a process of decision-making around multilevel risks and benefits that could inform ethical decisions regarding group-level research broadly.

After sexual assault, survivors’ friend networks comprise their main sources of support (Ullman, 1999). These networks are of particular importance because they can communicate norms about sexual assault and affect help-seeking decisions (Dworkin, Pittenger, & Allen, 2016). However, the ways that various characteristics of these networks can influence sexual assault survivors’ post-assault experiences remain largely unexamined. By asking every member of a closed group to provide information about their personal characteristics and report on their relationships with all other members, social network analysis (SNA) offers a unique way to summarize characteristics of a friend network that could be important to survivors’ disclosure decisions and recovery, including the attitudes (e.g., pervasiveness of rape myth acceptance in likely disclosure recipients), knowledge (e.g., likelihood that a survivor will disclose to someone who could refer her to a rape crisis center) and structure (e.g., interconnectedness or isolation of network members) of a given friend network. Although friend groups are often diffuse with difficult-to-determine boundaries (i.e., who is or is not in the friend group?), sororities are a naturally bounded friend group that could provide a unique setting in which these “climate-relevant” aspects of survivors’ recovery can be assessed. Further, collecting data from all members of such a setting would allow us to assess the attitudes, relationships, and knowledge of survivors’ friends directly, rather than relying on survivors’ perceptions of these variables. In addition to self-report measures assessing a variety of topics (e.g., mental health, rape myth acceptance, sexual assault history), we wanted to assess whether members considered other members to be friends and, for survivors, whether they had disclosed their assault to other members and found their responses helpful. Ultimately, we hoped to understand associations between survivors’ disclosure and recovery and 1) survivors’ direct and indirect connections to sorority members with varying levels of rape myth acceptance and knowledge about local resources, and 2) the degree of interconnectedness within survivors’ networks. We also hoped to conduct qualitative interviews to understand survivors’ subjective experience of the role of their network relationships in disclosure and recovery. Although whole-group social network analysis is associated with ethical risks (e.g., the necessity of using real names during data collection increases privacy risks), we proposed our design because it offered a new perspective on survivors’ recovery process that could help clarify the role of social contexts in recovery and ultimately inform the design of group-level interventions to promote survivor well-being.

Initial Considerations and Planning: Beginning to Identify Multilevel Risks and Benefits

Early hints of the ethical challenges inherent in this work emerged as we developed the initial IRB proposal. As violence-against-women researchers, we were aware of the potential individual-level risks of asking about sexual assault. We knew that a (very) small minority of survivors report feeling distressed by questions about their experiences (Campbell et al., 2010), and so we created a consent form that explained that we would ask about sexual assault experiences. Due to our value for attention to multiple levels of analysis, we were also conscious of both individual- and group- level privacy risks. We knew that survivors might be concerned about others finding out about their assault. We were accustomed to the protection provided by anonymous research, which is difficult to achieve with a research method that, by definition, links names to responses at the time of data collection. We also had concerns about risks posed to the group itself by network data collection (i.e., group-level risks). Because we wanted to collect data from participants who knew each other, as opposed to population-based research where participants are not necessarily interconnected, multiple individual privacy violations could have group-level consequences that they might not have in other types of research. For example, if there were a security breach and information about whom each person considered to be a friend (or not) was released, this might cause turmoil that could extend past a given dyad and affect the cohesion of the group as a whole, including relationships between nonparticipants. A group-level privacy violation could also affect the group’s willingness to permit access to other researchers or interventionists. We explored various data collection approaches for the survey, and in consultation with other network researchers, decided that online data collection would maximize the security of responses. Although names would be linked to responses during data collection, we developed a plan to replace names with identification numbers immediately upon downloading the data and retain an identity key only for those participants who agreed to be contacted regarding participation in the qualitative interviews.

We also considered group and individual benefits. Due to budget restrictions, we could not commit to paying the (potentially) 1000 participants a sufficient amount of money to incentivize participation. We also worried that we could not find an amount that would be incentivizing across sororities, because sororities presumably varied in the economic means of their members. On the suggestion of sorority insiders, we offered $500 to either the sorority or a charity of their choice for at least 85% participation, which would provide enough data to conduct a range of network analyses. These sororities were expected to raise money for a selected cause and to report the amount raised to their national-level organization, so a $500 donation to that effort would be highly desirable. Some sorority insiders also suggested that we send a letter to sororities’ national-level organization recognizing their participation, although conversations with other insiders led us to feel concerned that information about whether a particular chapter decided to participate could be used against the organization (i.e., a group-level privacy violation), so we decided against this additional incentive. As we gathered background information about sororities, however, we realized that our incentive plan presented another ethical challenge. Sororities varied widely in size: larger sororities consisted primarily of white women, and smaller sororities were comprised of women of color. If we included the smallest sororities, we would likely be spending a quarter of our budget for a very small amount of data (albeit data that would be consistent with our values of inclusivity). Because we planned to assess sexual assaults that had occurred since coming to college and/or joining the sorority, we expected the base rates of sexual assault to be somewhat low, so including these sororities would likely not yield enough participants who had been sexually assaulted for the data to be useful for our goals. We decided against limiting our inclusion criteria to larger sororities to be consistent with our values for justice and inclusivity.

Finally, we consulted SNA ethics sources regarding the types of information that we could ethically ask participants to provide regarding other sorority members (who might not consent to participate). Although it would have been useful to know whether members received sexual assault disclosures from other members in order to confirm that disclosures were made, this would violate the privacy of members who had made the disclosures. Asking members to provide this information about other members who might have intentionally declined to participate would violate their right to freely consent (or not) to research participation. In addition, this could raise individual-level risks, because names would be linked not only to information about individual participants, but also nonparticipants. These nonparticipants would thus need to be considered “secondary subjects,” or people who had participated as a function of others’ participation (Klovdahl, 2005), and risks to them would also need to be considered. We concluded that we could ask about participants’ perceptions of other sorority members (e.g., whether other sorority members were helpful when the research participant made a disclosure to them) without considering them secondary subjects, but if we asked about other sorority members’ actual behavior (e.g., whether other sorority members made disclosures of sexual assault to the research participant), secondary subjects concerns would apply (Borgatti & Molina, 2005).

Feedback from IRB: Honing in on Individual Risks

With optimism, we submitted our initial IRB proposal, but the IRB’s response made it clear that we had more ethical challenges to tackle. They had major concerns regarding individual risks. First, they felt that our plan to engage sorority gatekeepers (e.g., presidents) as recruitment liaisons was coercive, given that sorority settings are thought to encourage conformity. Second, they felt that using names in the survey without advance permission from individual members and linking responses to these names at the time of data collection posed risks to participants’ rights to confidentiality and free consent to participation, because other participants would be answering questions about people who had declined to participate. Finally, they were concerned that maintaining an identity key (which was necessary to conduct qualitative interviews with selected participants) raised privacy risks given the sensitivity of the topic. Feeling stuck, we asked to meet with our IRB contacts to determine a way forward. Beginning with this meeting, we worked with the IRB to concretize the core issues, share readings, and weigh options, with a focus on sharing the group-level risks and benefits that had informed our decision-making. Ultimately, we chose to withdraw our application in order to develop a new research proposal that would raise fewer concerns for the IRB.

Regrouping and Revising: Reducing Individual Risks

We made several key changes in our new proposal, which were mainly focused on reducing the individual risks identified by the IRB. First, we took steps to address privacy concerns. We decided to remove our qualitative component to obviate the need for an identity key. We considered allowing sorority members to opt in to having their names listed, but realized that participants would be able to determine who opted out by identifying names missing from the network list (Borgatti & Molina, 2005). We searched for an online platform that could dynamically insert member names in the survey without them ever existing together with responses, and came up short. Instead, we developed a “truly informed consent form” (Borgatti & Molina, 2005) that described the risks in greater detail than is typical for a consent form, including elaborating on various ways that our security measures could fail and providing examples of the harm that could result if this were to occur.

Ethical and Research Outcomes: Realizing that Individual and System Risks are Linked

We were excited to receive both approval and an invitation to co-present our project with our IRB partners as a case study at an IRB conference (Dworkin, Bagopal, Banks, & Allen, 2014). These early successes were not mirrored in our research outcomes, though. Ultimately, we gathered data from four sororities (four explicitly declined participation and 16 did not respond to recruitment emails), but—possibly due to our risk-minimizing decisions to (1) not involve sorority gatekeepers in recruitment efforts beyond deciding whether their chapter would participate, (2) increase perceptions of risk, and (3) limit incentives—response rates were dismally low (20–47% of sorority rosters; M = 29.75%), meaning that no sororities received the $500 incentive. Making matters worse, many participants did not answer enough questions to be included in network analyses (23–41% of participants; M = 31.25%). Because of these low response and completion rates, we were quite limited in the social network metrics we could calculate. For example, we were not able to calculate any metrics that relied on indirect connections between members (e.g., betweenness centrality). In addition, we could not be sure of the validity of some of the variables that we were able to calculate. Were people who cared more about ending sexual assault—those who would be less likely to endorse rape myth acceptance—more likely to participate because of how we framed the study, biasing group-level estimates of rape myth acceptance that we had hoped to use in multilevel models? Were friends of participants more likely to participate than non-friends, biasing density estimates? Thus, our efforts to minimize risk across levels also minimized the degree to which the study was able to contribute to broader knowledge or ameliorate the social problem of sexual assault.

Lessons Learned

Broadly speaking, our goal in sharing these experiences is to assist researchers in identifying risks and benefits across levels of analysis and understanding how they affect each other. Our experiences yielded key lessons in both domains.

Lesson 1: Analogous Risks and Benefits Exist Across Levels, But are Not Created Equal

We summarize the risks and benefits that we considered at individual, group, and societal levels in Table 1. We learned that, when identifying potential risks and benefits, it is helpful to keep in mind that individual risks often have analogous collective risks, and individual benefits often have analogous collective benefits. For example, both individual and group identifiability were risks, and it was possible to offer tangible incentives to both individuals and groups. We learned that research benefits might not translate across levels of analysis, and in particular, collective benefits might not be persuasive as individual benefits (Kadushin, 2005). Paying 800+ individual sorority members enough to incentivize participation would have been prohibitively expensive for an unfunded study, but group financial incentives ultimately did not encourage participation. Past research on this topic has found that some survivors participate in sexual assault research in order to contribute to the broader knowledge on this issue (Campbell et al., 2010)—a societal benefit leading to an individual benefit—but the degree to which this sufficiently incentivizes participation for enough participants is unclear. With the benefit of hindsight, we might have created tailored incentives based on the amount of participation to ensure that the group received some benefit. For example, sororities could have received $100 for 30% participation, $250 for 60% participation, and $500 for 90% participation. Although these lower tiers of participation would still be insufficient to compute many network statistics, they might have afforded us more power to test hypotheses that did not involve network statistics. Complicating matters, however, we learned that for groups within a system, benefits are not always one-size fits all. A $500 incentive for 85% participation to a 200-member sorority with the ability to fundraise on their own was not equivalent to the same incentive for a 20-member sorority with fewer resources.

Table 1.

Risks and Benefits Considered Across Levels of Analysis

	Risks	Benefits
Individual	Individual-level privacy violation Distress Violation of right to freely participate (i.e., coercion) Violation of right to decide whether one’s personal information is used in research (i.e., secondary subjects concerns)	Individual-level tangible incentive (e.g., money, class credit) Individual-level intangible incentive (e.g., increased understanding of research process, satisfaction of contributing to broader knowledge)
Group	Group-level privacy violation (e.g., identifiability of group to consumers of research), resulting in harm to nonparticipants’ relationships, or decreased group cohesion, reduced trust of researchers/interventionists Findings leveraged against group Group members participate but group does not receive tangible incentive	Group-level tangible incentive (e.g., money) Group-level intangible incentive (e.g., recognition from system for having participated)
System	Findings leveraged against system Findings fail to reflect marginalized groups within system	System recognized for contributing to broader knowledge
Society	Failure to contribute to knowledge on social problem Failure to ameliorate social problem	Contribution to broader knowledge Amelioration of social problem

We also believe that it is important to consider societal risks and benefits. Although societal benefits are often used to justify the ethical risks of studies, societal risks are considered less frequently. If a study fails to effectively contribute to generalizable knowledge or the amelioration of a social problem, this is not typically seen as something for which researchers are ethically accountable. However, when the potential societal benefits do not come to fruition, there is less to outweigh the risks to participants and groups, thus making them more ethically problematic. In addition, as more studies fail to produce these benefits, public trust in and support of research could erode.

Lesson 2: Changes to Risks and Benefits at One Level Affect Other Levels

The second focus of this paper is on the cross-level relationships between risks and benefits. It is well-known that altering risks can alter benefits at the individual level, but it is also noteworthy that changing risks at one level changes benefits at other levels, and changing benefits at one level changes risks at other levels. As an example of how increasing group benefits led to increased individual risks, we thought that conducting group data collection would help sororities reach 85% participation and receive their group-level benefit, but we worried that participants might feel social pressure to participate or not in a group setting (an individual-level risk). Moreover, as individual risks decreased, group, system, and societal benefits also decreased. For example, by preventing sorority gatekeepers from assisting with recruitment in any capacity to minimize the risk of coercion, and by increasing participants’ perception of the risk involved in participation with our “truly informed consent” form, we might have discouraged participation. Thus, no sorority achieved the response rate required to receive the incentive, which reduced the collective benefits. In turn, the high proportion of missing network data meant that we were unable to calculate many of the network metrics that we had hoped to use, and the lack of qualitative data prevented our ability to present rich idiographic descriptions, which together minimized the degree to which our study could have a broader societal benefit by contributing to generalizable knowledge.

Other Ethical Considerations When Balancing Multilevel Risks and Benefits

As we reflect on the experience of weighing these costs and benefits across levels of analysis, we are keenly aware that we, as researchers, are also stakeholders in this work whose decision-making is affected by a personal risk-benefit analysis. Researchers can benefit professionally from conducting novel, high-quality research, but risk wasting time and money if their findings are of low-quality. This could incentivize decisions that make the research worth doing but increase risks. We believe that it is crucial to recognize that we do not come to this decision-making process without vested interests that could lead to bias.

It is also important to note that researchers and IRBs ultimately have a great deal of power in making ethical decisions about risks and benefits, whereas the people and groups affected by the risks and benefits are typically not afforded a say. Of course, individuals can decide for themselves whether they want to participate, but in multilevel research, these decisions can affect and be affected by the groups they inhabit. In our study, for example, sorority leaders made decisions about whether the group would participate, which meant that they—and not individual members—decided whether individual members’ names would be listed in the social network measure. Members’ decisions to participate, in turn, affected the groups’ receipt of the $500 incentive. It is relatively rare that members of the group, or the group as a whole, is given approval power for the types of research questions asked or the release of findings, with the exception of participatory action research and research conducted with Native tribes. When groups are involved in such negotiations, though, it is possible that they may request group-level benefits that could pose more individual-level risks, such as a report of findings that could help the group make changes but could also lead to privacy violations and retribution against individual group members (Kadushin, 2005). It is also necessary to consider that group leadership might not speak for the interests of all individual members.

We considered whether risks and benefits should be weighed only during the research design phase, or also during later stages. There are clear examples of the importance of this consideration in the research publication phase. Currently, we are evaluating which of our original research questions can be addressed despite limited participation, and we are contemplating whether to include sorority characteristics including demographics, network cohesion, average rape myth acceptance, sorority-level prevalence of sexual assault, and pictorial representations of the relationships between network members in any manuscript we submit. It is possible that including these details could increase the individual- or group-level risks, because they could be used to identify sororities or their members. For example, if we presented information to suggest that a certain sorority lacked cohesiveness or had a particularly high prevalence of sexual assault, and that sorority was identified by readers from its demographic characteristics, this could affect the sorority’s ability to recruit new members (a group-level risk). Indeed, the publication of system-level sexual assault data has had system-level consequences: Harvard University recently decided to ban members of Greek organizations from holding leadership positions or receiving fellowships following the release of a campus study highlighting the high prevalence of sexual assault in sororities (Fahs, 2016), which represents both a societal benefit (i.e., the use of research to prevent sexual assault) and a systemic risk (i.e., the removal of resources from a system). However, not including such details might undermine the degree to which the study contributes to generalizable knowledge, thus minimizing societal benefits.

Conclusions and Next Steps

Applying community psychology’s attention to multiple levels of analysis to ethical decision-making raises more questions than answers, as is the case in many discussions of ethical issues (Kadushin, 2005). Indeed, it is one thing to know what risks and benefits are across levels of analysis, but it is another to balance them effectively. In what cases does risk to an individual outweigh systemic benefits? How much systemic benefit is enough to justify individual-level risk? These decisions will differ based on the nature of the research project, and thus, no sweeping guidelines can be offered. To fully attend to these multiple levels, community psychologists’ ethical guidelines might need to expand into a multidisciplinary space (Campbell, 2016). Although community psychologists commonly use the American Psychological Association’s ethical principles to guide such decisions, they lack the attention to the social ecology present in other ethical guidelines (e.g., those of the American Evaluation Association [1994/2007]). It is also possible that individual-level principles could be broadened in their focus. For example, when determining whether risks and benefits have been appropriately balanced, researchers routinely consider the ethical principles of nonmaleficence, beneficence, and nonexploitation, which are typically defined either explicitly or implicitly at the individual level (Emanuel, Wendler, and Grady, 2000). Explicitly linking these principles to multiple levels of analysis (e.g., doing no harm to individuals, groups, systems, or society) could more effectively draw researchers’ attention to these issues.